DETAILED ACTION
Response to Arguments
Election to Restriction Requirement
In response to the Restriction Requirement, Applicant elects without traverse Group II (claims 4-20, and 30-34) for examination, Group I (1-3, and 35-38) are withdrawn.
The amendments and arguments filed 12/12/2025 have been entered and made of record, and have been considered but are moot in view of the new ground(s) of rejection because the Applicant has amended at least independent claim 4, regarding to the newly added limitation that “wherein the neural network includes an encoder and decoder, the encoder comprising an image classification neural network that is pre-trained for classification of objects different than aneurysms”, which is rejected by MIHALEF as modified by ITU, and further in view of newly found reference BYDLON (US 20240383133 A1, claims priority of US-Provisional-Application US 63219507, July 8, 2021);
although MIHALEF as modified by ITU disclose the neural network includes auto-encoding classifier (see MIHALEF; e.g., …[0048] The machine-learnt predictor, with or without deep learning, is trained to associate the categorical labels (output clinical decision of the selections) to the extracted values of one or more features. The machine- learning uses training data with ground truth to learn to select based on the input vector. The resulting machine-learnt network is a matrix for inputs, weighting, convolution kernels, and/or combinations to output a clinical decision. Using the learned network, the processor inputs the extracted values for features and outputs the selection. ….[0047] Any machine learning or training may be used. A probabilistic boosting tree, support vector machine, neural network, sparse auto-encoding classifier, Bayesian network, or other now known or later developed machine learning may be used. Any semi-supervised, supervised, or unsupervised learning may be used. Hierarchal or other approaches may be used. In one embodiment, the classification is by a machine-learnt classifier learnt with deep learning. As part of identifying features that distinguish between different outcomes, the classifier is also machine learnt. Any deep learning approach or architecture may be used. For example, a convolutional neural network is used….
[0049] The machine-learned network is trained to output a selection, outcome, risk, deployment, or other clinical support information. The information extracted from the model may include the predicted outcomes and risks under different therapy choices. The information may select the therapy configuration, such as (i) start and end locations of the flow diverter, (ii) the number of flow diverters deployed, (iii) the configuration (overlapping or deployment into different branches), (iv) the reference radius of the flow diverter to use, (v) mean and maximum porosity in the neck/non-vessel area, etc. The information output by the machine-learned network may include identification of similar patients and the therapy choices made for the similar patients identified from a patient database and their outcomes.--, in [0047]-[0049]);
MIHALEF as modified by ITU however do not explicitly disclose the neural network includes an encoder and decoder, the encoder comprising an image classification neural network that is pre-trained for classification of objects different than aneurysms;
BYDLON discloses the neural network includes an encoder and decoder, the encoder comprising an image classification neural network that is pre-trained for classification of objects different than aneurysms (see BYDLON: e.g., Fig. 8, --[0019] a training system to retrain in dynamic learning the model based on the shape measurements encountered during deployment… [0020] The model may be one of clustering, trained and built on training data including historic and/or synthetically generated shape measurements. The categories predicted or output are hence presented by clusters in the training data. Training may be done independent from imagery such as X-ray imagery….. If a clustering approach is used, the current measurement may be displayed concurrently with a graphical rendition of some or all clusters. The models envisaged herein for the logic are however not confined to cluster type models. Other (machine learning) models, such as artificial neural networks, encoder-autoencoder, and other still, are also envisaged herein in embodiments. In other words, the predicted or output category is not necessarily indicative of a cluster in the training data, but may relate instead to classifier results, etc. Graphical indications of the said classes may also be displayed.--, in [0019]-[0021], [0188]-[0190], [0197]-[0198], and, Fig. 9, and Fig. 11, -- [0213] The trained machine learning module M may be stored in one or more memories MEM or databases and can be made available as pre-trained machine learning models for use in system.--, in [0210]-[0213], and [0240]-[0241]);
MIHALEF (as modified by ITU) and BYDLON are combinable as they are in the same field of endeavor: machine learning and neural network in processing of medical image of vascular anatomy including segmentation and classification for medical procedure and treatments. Therefore it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to further modify MIHALEF(as modified by ITU)’s system using BYDLON’s teachings by including the neural network includes an encoder and decoder, the encoder comprising an image classification neural network that is pre-trained for classification of objects different than aneurysms to MIHALEF(as modified by ITU)’s neural network and machine learning model in order to perform clustering and segmentation based on shape measurements (see BYDLON: e.g. in [0019]-[0021], [0188]-[0190], [0197]-[0198], [0210]-[0213], and [0240]-[0241]).
Therefore, claims 4-20, and 30-34 are still not patentably distinguishable over the prior art reference(s). Further discussions are addressed in the prior art rejection section below.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 4-20, and 30-34 are rejected under 35 U.S.C. 103 as being unpatentable over MIHALEF (WO 2020064090 A1), in view of ITU (US 20210219935 A1), and further in view of BYDLON (US 20240383133 A1, claims priority of US-Provisional-Application US 63219507, July 8, 2021).
Re Claim 4, MIHALEF discloses a system for providing outcome predictions for intrasaccular implant devices based on digital imaging and clinical information (see MIHALEF: e.g., --[0046] The decision support uses the anatomical and/or physiological knowledge using any of the different machine learning models. For training, many samples with known ground truth (e.g., selections) are used. For the samples, the machine learning model inputs may be real (from patients, or bench-scale measurements), virtual (synthetic or not representative of a given patient or physical bench arrangement), or a combination of the two. For patient data, the input samples may contain the patient images, blood biomarkers, demographics, measurements and/or genetic data. For the ground truth, the input data includes the therapy that was chosen for the patient. Other ground truths may alternatively or additionally be used, such as the recorded outcome at different temporal points (e.g., 30 days, 90 days, etc.) and/or the occurrence or lack thereof of adverse events. …[0048] The machine-learnt predictor, with or without deep learning, is trained to associate the categorical labels (output clinical decision of the selections) to the extracted values of one or more features. The machine- learning uses training data with ground truth to learn to select based on the input vector. The resulting machine-learnt network is a matrix for inputs, weighting, convolution kernels, and/or combinations to output a clinical decision. Using the learned network, the processor inputs the extracted values for features and outputs the selection. ….[0047] Any machine learning or training may be used. A probabilistic boosting tree, support vector machine, neural network, sparse auto-encoding classifier, Bayesian network, or other now known or later developed machine learning may be used. Any semi-supervised, supervised, or unsupervised learning may be used. Hierarchal or other approaches may be used. In one embodiment, the classification is by a machine-learnt classifier learnt with deep learning. As part of identifying features that distinguish between different outcomes, the classifier is also machine learnt. Any deep learning approach or architecture may be used. For example, a convolutional neural network is used….
[0049] The machine-learned network is trained to output a selection, outcome, risk, deployment, or other clinical support information. The information extracted from the model may include the predicted outcomes and risks under different therapy choices. The information may select the therapy configuration, such as (i) start and end locations of the flow diverter, (ii) the number of flow diverters deployed, (iii) the configuration (overlapping or deployment into different branches), (iv) the reference radius of the flow diverter to use, (v) mean and maximum porosity in the neck/non-vessel area, etc. The information output by the machine-learned network may include identification of similar patients and the therapy choices made for the similar patients identified from a patient database and their outcomes.--, in [0044]-[0049]), the system comprising:
memory; and at least one processor coupled to the memory, and based at least in part on information stored in the memory, configured to: receive, as input from a user device, digital imaging information and the clinical information for an aneurysm patient (see MIHALEF: e.g., --the user may indicate the starting and ending points on an image of the vessel or vessel model (e.g., 3D mesh rendered to 2D display image). Any locations of packing to increase porosity may be indicated. Other user inputs of the device and/or placement may be used, such as configuration for multiple devices…[0044]… the decision support processor selects. A selection function may be used, such as relating one or more different patient- specific characteristics to characteristics of or specific ones of the available devices. In one embodiment, a machine-learned network selects the vascular implant and/or placement of the vascular implant….the machine learning model inputs may be real (from patients, or bench-scale measurements), virtual (synthetic or not representative of a given patient or physical bench arrangement), or a combination of the two. For patient data, the input samples may contain the patient images, blood biomarkers, demographics, measurements and/or genetic data. For the ground truth, the input data includes the therapy that was chosen for the patient. Other ground truths may alternatively or additionally be used, such as the recorded outcome at different temporal points (e.g., 30 days, 90 days, etc.) and/or the occurrence or lack thereof of adverse events.--, in [0043]-[0046]; and, --the implant model is bent to follow the curvature of the vessel centerline. In a last stage, the release of the crimping is simulated so the implant model expands and fills the lumen. Other non- clinical-based simulation may be used, such as outputting a deployment and corresponding deformation from a machine-learned network based on inputs of the scan data, vessel model, patient-specific information, and/or device model. …[0062] In act 23, the decision support processor calculates a porosity of the endovascular device from the deployment. For the flow diverter or other implant, the porosity, in part, controls the flow. In-situ implantation may produce variations in the metal coverage ratio (MCR) and porosity of the device, especially in the aneurysm and adjacent regions. Hemodynamics is consequently affected by such regions, so porosity is a consideration when performing the implantation. By simulating deployment, the porosity that results in general or by location may be determined for estimating outcome.--, in [0060]-[0063]; also see: -- for patients with aneurysms, the choices could be coiling the aneurysm or placement of any one of different models of flow diverters with multiple possible implantation configurations. The clinician decides if implantation of endovascular devices like flow diverters or stents would be beneficial to the patient, and if so, make an optimal choice of the device type and its parameters (e.g. diameter, length, porosity, metal coverage area, material mechanical properties, etc.). In the case of patients with aneurysms--, in [0001], and, -- The decision support for the treatment planning process is based on a systemic analysis of the above-mentioned factors along with simulated treatment of the patient’s aneurysm using a virtual implantation of a flow-diverting stent. Only the data acquired prior to the treatment planning (medical images, blood biomarkers, patient medical history and demographics) is used for further processing. The model may also utilize simulated outcomes using machine learning or physics-based simulators, such as computational fluid dynamics models. For such simulations, an anatomical model may be extracted using either manual, semi-automatic or automatic segmentation algorithms. The virtual deployment system is integrated into the decision support. [0022] Figure 1 shows a flow chart or outline of one embodiment of the decision support for endovascular procedure planning. The outline illustrates a workflow, including device modeling (fitting the device characteristics to a physics model), deployment of the device model in a representation of the vessel of the patient, and outcome prediction based on the deployment.--, in [0021]-[0022]);
generate, using a neural network trained for aneurysm outcome prediction, an outcome prediction for an aneurysm treatment of the aneurysm patient based on at least one treatment device for implant in an aneurysm sac of the aneurysm patient and the at least one of the imaging information as the input (see MIHALEF: e.g., --[0021] Aneurysm treatment planning in clinical practice depends on acquiring suitable medical images describing the pathologic vascular region. Based on the imaging data in combination with patient demographics (age, gender, weight, body mass index, etc.), past medical history (diabetes, current list of medications, previous stroke, etc.) and/or blood biomarkers (e.g., blood counts, clotting factors, hematocrit, blood viscosity, glucose level, etc.), a patient-individualized treatment strategy is set up and performed. The decision support for the treatment planning process is based on a systemic analysis of the above-mentioned factors along with simulated treatment of the patient’s aneurysm using a virtual implantation of a flow-diverting stent. Only the data acquired prior to the treatment planning (medical images, blood biomarkers, patient medical history and demographics) is used for further processing. The model may also utilize simulated outcomes using machine learning or physics-based simulators, such as computational fluid dynamics models. For such simulations, an anatomical model may be extracted using either manual, semi-automatic or automatic segmentation algorithms. The virtual deployment system is integrated into the decision support. [0022] Figure 1 shows a flow chart or outline of one embodiment of the decision support for endovascular procedure planning. The outline illustrates a workflow, including device modeling (fitting the device characteristics to a physics model), deployment of the device model in a representation of the vessel of the patient, and outcome prediction based on the deployment.--, in [0021]-[0022], and, --[0026] In stage 16, a machine-learned network predicts outcome based on the simulated deployment, imaging data (e.g., segmented vessel), and/or other patient-specific information. The machine-learned network may be used to output implant information (e.g., type, size, placement, number of devices, configuration of device, etc.) used for the simulation in stage 14. Alternatively or additionally, the machine-learned network may be used to predict outcome or prognosis for a simulated deployment. The decision support system may be trained on outcome data, if available, to predict not only short-term physiological response, but also long-term outcomes and/or costs. The outcome or prognosis may be used to compare or select implant information, or the network directly outputs the selection.
[0027] In stage 18, the simulated deployment, predicted outcome, and/or a calculated hemodynamic characteristic based on the deployment are provided to the clinician. The information may be used to support decisions for planning an endovascular implantation.--, in [0026]-[0027]);
MIHALEF however does not explicitly disclose generate, using a neural network trained for aneurysm, an outcome prediction a outcome prediction based on the clinical information received as the input;
ITU discloses generate, using a neural network trained for aneurysm, an outcome prediction a outcome prediction based on the clinical information received as the input (see ITU: e.g., -- FIG. 8 shows an example of the normal distribution of the root radius of a coronary left arterial tree. Known, estimated, or standard normal distributions may be used. The synthetic examples generated are assigned the value for the root radius of the coronary left arterial tree based on the distribution (e.g., probability of a given value per example assigned using the distribution).--, in [0084], and,
-- [0102] Other features extracted include parameters for one or more abnormalities of the vessel structure. Abnormal morphology may be characterized by characteristics of calcification, characteristics of the plaque (e.g., fibrous tissue, lipid tissue, necrotic tissue, calcified tissue), characteristics of thrombus, characteristics of diffuse disease, presence of total or sub-total occlusion, presence of myocardial bridging (superficial and/or deep), congenital anomalies of coronary arteries (e.g., anomalous origin of a coronary artery from an abnormal sinus of Valsalva with an inter-arterial course between the great arteries, anomalous origin of one coronary artery from the pulmonary trunk, or others), aneurysmal dilatation and superimposed atherosclerosis, “high take off” coronary artery (e.g., the ostium is several millimeters above the sino-tubular junction (the artery may have a sharp downward angle and runs partially through the aortic wall)), myocardial bridging: superficial and deep, coronary fistula, coronary artery dissection, coronary vasculitis (e.g., rheumatoid arthritis, systemic lupus erythematosus (SLE), or Behçet's disease, Kawasaki disease, polyarteritis nodosa, and/or persisting (post) inflammatory aneurysms), fibromuscular dysplasia, coronary micro embolization, and/or left or right dominance. Additional, different, or fewer abnormality features may be used.--, in [0102], and, -- A local ischemia weight value is estimated independently for each root/branch/leaf segment using geometric features of the segment, such as the reference radius, length, tapering rate and other features…. An average value of healthy radiuses of the entire branch or a part of the branch, an average value of healthy radiuses obtained when excluding the largest x % and the smallest y % of the radius values of the entire branch or a part of the branch, or maximum or minimum value of healthy radii of the entire branch or part of the branch are computed.--, in [0110]-[0112]; and,
-- [0146] The in vitro model 23 and the flow conditions may be modified in numerous ways to generate a large number of setups. For example, the number, position and shape of the occluders is altered. As another example, the resistance at one or more locations is altered. In yet another example, the operation of the pump is altered. The number of side branches and any occlusions may be altered. Other alterations of combinations of different alterations are used to create different models with corresponding features and resulting flow characteristics.--, in [0145]-[0146], and, -- [0152] More than one classifier may be created. Since different types of branches and regions are present in a vessel tree, different classifiers may be machine trained for the different branches and/or regions. For example, different classifiers are trained for main and side branches, bifurcation regions and single branch regions, different types of pathologic regions such as different types of single branch stenotic regions (e.g., focal, long, diffuse, restenosis, or other), different types of bifurcation stenoses (e.g. a separate model for each bifurcation stenosis type in the medina classification), different types of aneurysms, different types of plaque, different types of total and/or sub-total occlusions, stenotic and regurgitant valves, various pathologies of the heart (e.g., past infarct or myopathies), or different types of branches (e.g. in case of coronary arterial trees: LM, LAD, LCx, RCA, Diagonal, OM, or other). Since the training is based on synthetic geometries, a large enough number of training instances may be generated for each of these different classifiers. Another possibility is to divide the geometry into separate segments (e.g. for coronary geometries: proximal LAD, mid LAD, and distal LAD) and to extract the features discussed in the previous sections separately for each segment. Afterwards these features may either be combined into cumulative features or used separately for a single or multiple machine learning algorithms for predicting a hemodynamic metric of interest.--, in [0152], and, -- In act 60, uncertainty is assigned to one or more features. The uncertainty is a distribution of possible values for the feature. For example, the radius may be measured as 0.25 cm, but the accuracy or tolerance in the measurement provides that the radius is between 0.20 cm and 0.30 cm with greater probability for the values closer to 0.25 cm. Any distribution of possible or probable values may be used, such as a normal distribution, a distribution from a study, or from another source…. Using the machine-learnt algorithm, the confidence of the estimated hemodynamic metric is provided. A confidence or probability is provided for one value of the metric. Alternatively, the predictions from the learnt model may also be ranges or confidence intervals within which the predicted quantity is expected. The predicted confidence interval for the patient could be either directly predicted from the model or estimated from a set of similar anatomies from a saved database of synthetic models. --, in [0178]-[0180]);
MIHALEF and ITU are combinable as they are in the same field of endeavor: machine learning and neural network in vascular implant treatment outcome prediction. Therefore it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to further modify MIHALEF’s apparatus using ITU’s teachings by including generate, using a neural network trained for aneurysm, an outcome prediction a outcome prediction based on the clinical information received as the input to MIHALEF’s outcome prediction in order to provide confidence or probability for the features associated with the outcome prediction for an intrasaccular implant (see ITU: e.g. in [0084], [0102], [0110]-[0112], [0145]-[0146], [0152], and [0178]-[0180]);
although MIHALEF as modified by ITU disclose the neural network includes auto-encoding classifier (see MIHALEF; e.g., …[0048] The machine-learnt predictor, with or without deep learning, is trained to associate the categorical labels (output clinical decision of the selections) to the extracted values of one or more features. The machine- learning uses training data with ground truth to learn to select based on the input vector. The resulting machine-learnt network is a matrix for inputs, weighting, convolution kernels, and/or combinations to output a clinical decision. Using the learned network, the processor inputs the extracted values for features and outputs the selection. ….[0047] Any machine learning or training may be used. A probabilistic boosting tree, support vector machine, neural network, sparse auto-encoding classifier, Bayesian network, or other now known or later developed machine learning may be used. Any semi-supervised, supervised, or unsupervised learning may be used. Hierarchal or other approaches may be used. In one embodiment, the classification is by a machine-learnt classifier learnt with deep learning. As part of identifying features that distinguish between different outcomes, the classifier is also machine learnt. Any deep learning approach or architecture may be used. For example, a convolutional neural network is used….
[0049] The machine-learned network is trained to output a selection, outcome, risk, deployment, or other clinical support information. The information extracted from the model may include the predicted outcomes and risks under different therapy choices. The information may select the therapy configuration, such as (i) start and end locations of the flow diverter, (ii) the number of flow diverters deployed, (iii) the configuration (overlapping or deployment into different branches), (iv) the reference radius of the flow diverter to use, (v) mean and maximum porosity in the neck/non-vessel area, etc. The information output by the machine-learned network may include identification of similar patients and the therapy choices made for the similar patients identified from a patient database and their outcomes.--, in [0047]-[0049]);
MIHALEF as modified by ITU however do not explicitly disclose the neural network includes an encoder and decoder, the encoder comprising an image classification neural network that is pre-trained for classification of objects different than aneurysms;
BYDLON discloses the neural network includes an encoder and decoder, the encoder comprising an image classification neural network that is pre-trained for classification of objects different than aneurysms (see BYDLON: e.g., Fig. 8, --[0019] a training system to retrain in dynamic learning the model based on the shape measurements encountered during deployment… [0020] The model may be one of clustering, trained and built on training data including historic and/or synthetically generated shape measurements. The categories predicted or output are hence presented by clusters in the training data. Training may be done independent from imagery such as X-ray imagery….. If a clustering approach is used, the current measurement may be displayed concurrently with a graphical rendition of some or all clusters. The models envisaged herein for the logic are however not confined to cluster type models. Other (machine learning) models, such as artificial neural networks, encoder-autoencoder, and other still, are also envisaged herein in embodiments. In other words, the predicted or output category is not necessarily indicative of a cluster in the training data, but may relate instead to classifier results, etc. Graphical indications of the said classes may also be displayed.--, in [0019]-[0021], [0188]-[0190], [0197]-[0198], and, Fig. 9, and Fig. 11, -- [0213] The trained machine learning module M may be stored in one or more memories MEM or databases and can be made available as pre-trained machine learning models for use in system.--, in [0210]-[0213], and [0240]-[0241]);
MIHALEF (as modified by ITU) and BYDLON are combinable as they are in the same field of endeavor: machine learning and neural network in processing of medical image of vascular anatomy including segmentation and classification for medical procedure and treatments. Therefore it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to further modify MIHALEF(as modified by ITU)’s system using BYDLON’s teachings by including the neural network includes an encoder and decoder, the encoder comprising an image classification neural network that is pre-trained for classification of objects different than aneurysms to MIHALEF(as modified by ITU)’s neural network and machine learning model in order to perform clustering and segmentation based on shape measurements (see BYDLON: e.g. in [0019]-[0021], [0188]-[0190], [0197]-[0198], [0210]-[0213], and [0240]-[0241]);
MIHALEF as modified by ITU and BYDLON further disclose send, to a display, the outcome prediction for the aneurysm treatment for display at the user device (see MIHALEF: e.g., ….[0047] Any machine learning or training may be used. A probabilistic boosting tree, support vector machine, neural network, sparse auto-encoding classifier, Bayesian network, or other now known or later developed machine learning may be used. Any semi-supervised, supervised, or unsupervised learning may be used. Hierarchal or other approaches may be used. In one embodiment, the classification is by a machine-learnt classifier learnt with deep learning. As part of identifying features that distinguish between different outcomes, the classifier is also machine learnt. Any deep learning approach or architecture may be used. For example, a convolutional neural network is used….
[0049] The machine-learned network is trained to output a selection, outcome, risk, deployment, or other clinical support information. The information extracted from the model may include the predicted outcomes and risks under different therapy choices. The information may select the therapy configuration, such as (i) start and end locations of the flow diverter, (ii) the number of flow diverters deployed, (iii) the configuration (overlapping or deployment into different branches), (iv) the reference radius of the flow diverter to use, (v) mean and maximum porosity in the neck/non-vessel area, etc. The information output by the machine-learned network may include identification of similar patients and the therapy choices made for the similar patients identified from a patient database and their outcomes.--, in [0044]-[0049], and, --[0077] In act 26, the decision support processor generates an image, and a display screen displays the image. The output (e.g., outcome, hemodynamic quantity, porosity, visual depiction of the deployment or end result of deployment, or combinations thereof) is transmitted to a display…..[0079] In one embodiment, the clinical decisions are visualized, either as text or in a graphical way (e.g. overlaid on the medical images) and presented to the clinician. Decision support information, such as treatments, risks, guidelines, or other information, may be output. Diagnostic rules for treatment, such as based on guidelines or studies, may be output as decision support.
[0080] Any display of the decision or decisions may be used. In one embodiment, a decision tree shows the clinical decision, other possible decisions, and further treatment options resulting from the clinical decision. Besides a basic text-based display, another option is to display in a hierarchy not only the currently selected clinical decision but also possible subsequent clinical decisions.
[0081] The displayed image from the deployment simulation may be used to visually guide the placement of the device or devices. For instance, the imaging during the procedure may utilize the output to overlay planes for the start and end locations of the flow diverter to aid in device placement.—[0077]-[0081]).
Re Claim 5, MIHALEF as modified by ITU and BYDLON further disclose the at least one processor is configured to process the imaging information and the clinical information using the neural network to generate the outcome prediction for the aneurysm treatment (see MIHALEF: e.g., ….[0047] Any machine learning or training may be used. A probabilistic boosting tree, support vector machine, neural network, sparse auto-encoding classifier, Bayesian network, or other now known or later developed machine learning may be used. Any semi-supervised, supervised, or unsupervised learning may be used. Hierarchal or other approaches may be used. In one embodiment, the classification is by a machine-learnt classifier learnt with deep learning. As part of identifying features that distinguish between different outcomes, the classifier is also machine learnt. Any deep learning approach or architecture may be used. For example, a convolutional neural network is used….
[0049] The machine-learned network is trained to output a selection, outcome, risk, deployment, or other clinical support information. The information extracted from the model may include the predicted outcomes and risks under different therapy choices. The information may select the therapy configuration, such as (i) start and end locations of the flow diverter, (ii) the number of flow diverters deployed, (iii) the configuration (overlapping or deployment into different branches), (iv) the reference radius of the flow diverter to use, (v) mean and maximum porosity in the neck/non-vessel area, etc. The information output by the machine-learned network may include identification of similar patients and the therapy choices made for the similar patients identified from a patient database and their outcomes.--, in [0044]-[0049], and, --[0077] In act 26, the decision support processor generates an image, and a display screen displays the image. The output (e.g., outcome, hemodynamic quantity, porosity, visual depiction of the deployment or end result of deployment, or combinations thereof) is transmitted to a display…..[0079] In one embodiment, the clinical decisions are visualized, either as text or in a graphical way (e.g. overlaid on the medical images) and presented to the clinician. Decision support information, such as treatments, risks, guidelines, or other information, may be output. Diagnostic rules for treatment, such as based on guidelines or studies, may be output as decision support.
[0080] Any display of the decision or decisions may be used. In one embodiment, a decision tree shows the clinical decision, other possible decisions, and further treatment options resulting from the clinical decision. Besides a basic text-based display, another option is to display in a hierarchy not only the currently selected clinical decision but also possible subsequent clinical decisions.
[0081] The displayed image from the deployment simulation may be used to visually guide the placement of the device or devices. For instance, the imaging during the procedure may utilize the output to overlay planes for the start and end locations of the flow diverter to aid in device placement.—[0077]-[0081]).
Re Claim 6, MIHALEF as modified by ITU and BYDLON further disclose image classification neural network is pre-trained for the classification of objects different including categories for one or more of common household objects, inanimate objects, or animals (see MIHALEF: e.g., --[0021] Aneurysm treatment planning in clinical practice depends on acquiring suitable medical images describing the pathologic vascular region. Based on the imaging data in combination with patient demographics (age, gender, weight, body mass index, etc.), past medical history (diabetes, current list of medications, previous stroke, etc.) and/or blood biomarkers (e.g., blood counts, clotting factors, hematocrit, blood viscosity, glucose level, etc.), a patient-individualized treatment strategy is set up and performed. The decision support for the treatment planning process is based on a systemic analysis of the above-mentioned factors along with simulated treatment of the patient’s aneurysm using a virtual implantation of a flow-diverting stent. Only the data acquired prior to the treatment planning (medical images, blood biomarkers, patient medical history and demographics) is used for further processing. The model may also utilize simulated outcomes using machine learning or physics-based simulators, such as computational fluid dynamics models. For such simulations, an anatomical model may be extracted using either manual, semi-automatic or automatic segmentation algorithms. The virtual deployment system is integrated into the decision support.--, in [0021], and, -- [0047] Any machine learning or training may be used. A probabilistic boosting tree, support vector machine, neural network, sparse auto-encoding classifier, Bayesian network, or other now known or later developed machine learning may be used. Any semi-supervised, supervised, or unsupervised learning may be used. Hierarchal or other approaches may be used. In one embodiment, the classification is by a machine-learnt classifier learnt with deep learning. As part of identifying features that distinguish between different outcomes, the classifier is also machine learnt. Any deep learning approach or architecture may be used. For example, a convolutional neural network is used. The network may include convolutional, sub-sampling (e.g., max pooling), fully connected layers, and/or other types of layers. By using convolution, the number of possible features to be tested is limited. The fully connected layers operate to fully connect the features as limited by the convolution layer after maximum pooling. Other features may be added to the fully connected layers, such as non-imaging or clinical information. Any combination of layers may be provided. Hierarchical structures are employed, either for learning features or representation or for classification or regression. The computer-based decision support system employs a machine learning algorithm for automated decision making.--, in [0046]-[0047]; also see BYDLON: e.g., Fig. 8, --[0019] a training system to retrain in dynamic learning the model based on the shape measurements encountered during deployment… [0020] The model may be one of clustering, trained and built on training data including historic and/or synthetically generated shape measurements. The categories predicted or output are hence presented by clusters in the training data. Training may be done independent from imagery such as X-ray imagery….. If a clustering approach is used, the current measurement may be displayed concurrently with a graphical rendition of some or all clusters. The models envisaged herein for the logic are however not confined to cluster type models. Other (machine learning) models, such as artificial neural networks, encoder-autoencoder, and other still, are also envisaged herein in embodiments. In other words, the predicted or output category is not necessarily indicative of a cluster in the training data, but may relate instead to classifier results, etc. Graphical indications of the said classes may also be displayed.--, in [0019]-[0021], [0188]-[0190], [0197]-[0198], and, Fig. 9, and Fig. 11, -- [0213] The trained machine learning module M may be stored in one or more memories MEM or databases and can be made available as pre-trained machine learning models for use in system.--, in [0210]-[0213], and [0240]-[0241]).
Re Claim 7, MIHALEF as modified by ITU and BYDLON further disclose wherein the neural network is configured with a classification algorithm based on at least one of a random forest algorithm, a multilayer perceptron (MLP) neural network algorithm, a logistic regression algorithm, a naive Bayes machine learning algorithm, or a support vector machine (SVM) algorithm (see MIHALEF: e.g., --[0021] Aneurysm treatment planning in clinical practice depends on acquiring suitable medical images describing the pathologic vascular region. Based on the imaging data in combination with patient demographics (age, gender, weight, body mass index, etc.), past medical history (diabetes, current list of medications, previous stroke, etc.) and/or blood biomarkers (e.g., blood counts, clotting factors, hematocrit, blood viscosity, glucose level, etc.), a patient-individualized treatment strategy is set up and performed. The decision support for the treatment planning process is based on a systemic analysis of the above-mentioned factors along with simulated treatment of the patient’s aneurysm using a virtual implantation of a flow-diverting stent. Only the data acquired prior to the treatment planning (medical images, blood biomarkers, patient medical history and demographics) is used for further processing. The model may also utilize simulated outcomes using machine learning or physics-based simulators, such as computational fluid dynamics models. For such simulations, an anatomical model may be extracted using either manual, semi-automatic or automatic segmentation algorithms. The virtual deployment system is integrated into the decision support.--, in [0021], and, -- [0047] Any machine learning or training may be used. A probabilistic boosting tree, support vector machine, neural network, sparse auto-encoding classifier, Bayesian network, or other now known or later developed machine learning may be used. Any semi-supervised, supervised, or unsupervised learning may be used. Hierarchal or other approaches may be used. In one embodiment, the classification is by a machine-learnt classifier learnt with deep learning. As part of identifying features that distinguish between different outcomes, the classifier is also machine learnt. Any deep learning approach or architecture may be used. For example, a convolutional neural network is used. The network may include convolutional, sub-sampling (e.g., max pooling), fully connected layers, and/or other types of layers. By using convolution, the number of possible features to be tested is limited. The fully connected layers operate to fully connect the features as limited by the convolution layer after maximum pooling. Other features may be added to the fully connected layers, such as non-imaging or clinical information. Any combination of layers may be provided. Hierarchical structures are employed, either for learning features or representation or for classification or regression. The computer-based decision support system employs a machine learning algorithm for automated decision making.--, in [0046]-[0047]).
Re Claim 8, MIHALEF as modified by ITU and BYDLON further disclose wherein the input includes both the imaging information and the clinical information associated with the aneurysm patient and the at least one processor is configured to generate the outcome prediction for the aneurysm treatment based on both the imaging information and the clinical information (see MIHALEF: e.g., -- The decision support for the treatment planning process is based on a systemic analysis of the above-mentioned factors along with simulated treatment of the patient’s aneurysm using a virtual implantation of a flow-diverting stent. Only the data acquired prior to the treatment planning (medical images, blood biomarkers, patient medical history and demographics) is used for further processing. The model may also utilize simulated outcomes using machine learning or physics-based simulators, such as computational fluid dynamics models. For such simulations, an anatomical model may be extracted using either manual, semi-automatic or automatic segmentation algorithms. The virtual deployment system is integrated into the decision support. [0022] Figure 1 shows a flow chart or outline of one embodiment of the decision support for endovascular procedure planning. The outline illustrates a workflow, including device modeling (fitting the device characteristics to a physics model), deployment of the device model in a representation of the vessel of the patient, and outcome prediction based on the deployment.--, in [0021]-[0022], and, ….[0047] Any machine learning or training may be used. A probabilistic boosting tree, support vector machine, neural network, sparse auto-encoding classifier, Bayesian network, or other now known or later developed machine learning may be used. Any semi-supervised, supervised, or unsupervised learning may be used. Hierarchal or other approaches may be used. In one embodiment, the classification is by a machine-learnt classifier learnt with deep learning. As part of identifying features that distinguish between different outcomes, the classifier is also machine learnt. Any deep learning approach or architecture may be used. For example, a convolutional neural network is used--, in [0047]; [0049] The machine-learned network is trained to output a selection, outcome, risk, deployment, or other clinical support information. The information extracted from the model may include the predicted outcomes and risks under different therapy choices. The information may select the therapy configuration, such as (i) start and end locations of the flow diverter, (ii) the number of flow diverters deployed, (iii) the configuration (overlapping or deployment into different branches), (iv) the reference radius of the flow diverter to use, (v) mean and maximum porosity in the neck/non-vessel area, etc. The information output by the machine-learned network may include identification of similar patients and the therapy choices made for the similar patients identified from a patient database and their outcomes.--, in [0049], and, --[0077] In act 26, the decision support processor generates an image, and a display screen displays the image. The output (e.g., outcome, hemodynamic quantity, porosity, visual depiction of the deployment or end result of deployment, or combinations thereof) is transmitted to a display…..[0079] In one embodiment, the clinical decisions are visualized, either as text or in a graphical way (e.g. overlaid on the medical images) and presented to the clinician. Decision support information, such as treatments, risks, guidelines, or other information, may be output. Diagnostic rules for treatment, such as based on guidelines or studies, may be output as decision support.
[0080] Any display of the decision or decisions may be used. In one embodiment, a decision tree shows the clinical decision, other possible decisions, and further treatment options resulting from the clinical decision. Besides a basic text-based display, another option is to display in a hierarchy not only the currently selected clinical decision but also possible subsequent clinical decisions.
[0081] The displayed image from the deployment simulation may be used to visually guide the placement of the device or devices. For instance, the imaging during the procedure may utilize the output to overlay planes for the start and end locations of the flow diverter to aid in device placement.—[0077]-[0081]; also see ITU: e.g., -- FIG. 8 shows an example of the normal distribution of the root radius of a coronary left arterial tree. Known, estimated, or standard normal distributions may be used. The synthetic examples generated are assigned the value for the root radius of the coronary left arterial tree based on the distribution (e.g., probability of a given value per example assigned using the distribution).--, in [0084], and,
-- [0102] Other features extracted include parameters for one or more abnormalities of the vessel structure. Abnormal morphology may be characterized by characteristics of calcification, characteristics of the plaque (e.g., fibrous tissue, lipid tissue, necrotic tissue, calcified tissue), characteristics of thrombus, characteristics of diffuse disease, presence of total or sub-total occlusion, presence of myocardial bridging (superficial and/or deep), congenital anomalies of coronary arteries (e.g., anomalous origin of a coronary artery from an abnormal sinus of Valsalva with an inter-arterial course between the great arteries, anomalous origin of one coronary artery from the pulmonary trunk, or others), aneurysmal dilatation and superimposed atherosclerosis, “high take off” coronary artery (e.g., the ostium is several millimeters above the sino-tubular junction (the artery may have a sharp downward angle and runs partially through the aortic wall)), myocardial bridging: superficial and deep, coronary fistula, coronary artery dissection, coronary vasculitis (e.g., rheumatoid arthritis, systemic lupus erythematosus (SLE), or Behçet's disease, Kawasaki disease, polyarteritis nodosa, and/or persisting (post) inflammatory aneurysms), fibromuscular dysplasia, coronary micro embolization, and/or left or right dominance. Additional, different, or fewer abnormality features may be used.--, in [0102], and, -- A local ischemia weight value is estimated independently for each root/branch/leaf segment using geometric features of the segment, such as the reference radius, length, tapering rate and other features…. An average value of healthy radiuses of the entire branch or a part of the branch, an average value of healthy radiuses obtained when excluding the largest x % and the smallest y % of the radius values of the entire branch or a part of the branch, or maximum or minimum value of healthy radii of the entire branch or part of the branch are computed.--, in [0110]-[0112]; and,
-- [0146] The in vitro model 23 and the flow conditions may be modified in numerous ways to generate a large number of setups. For example, the number, position and shape of the occluders is altered. As another example, the resistance at one or more locations is altered. In yet another example, the operation of the pump is altered. The number of side branches and any occlusions may be altered. Other alterations of combinations of different alterations are used to create different models with corresponding features and resulting flow characteristics.--, in [0145]-[0146], and, -- [0152] More than one classifier may be created. Since different types of branches and regions are present in a vessel tree, different classifiers may be machine trained for the different branches and/or regions. For example, different classifiers are trained for main and side branches, bifurcation regions and single branch regions, different types of pathologic regions such as different types of single branch stenotic regions (e.g., focal, long, diffuse, restenosis, or other), different types of bifurcation stenoses (e.g. a separate model for each bifurcation stenosis type in the medina classification), different types of aneurysms, different types of plaque, different types of total and/or sub-total occlusions, stenotic and regurgitant valves, various pathologies of the heart (e.g., past infarct or myopathies), or different types of branches (e.g. in case of coronary arterial trees: LM, LAD, LCx, RCA, Diagonal, OM, or other). Since the training is based on synthetic geometries, a large enough number of training instances may be generated for each of these different classifiers. Another possibility is to divide the geometry into separate segments (e.g. for coronary geometries: proximal LAD, mid LAD, and distal LAD) and to extract the features discussed in the previous sections separately for each segment. Afterwards these features may either be combined into cumulative features or used separately for a single or multiple machine learning algorithms for predicting a hemodynamic metric of interest.--, in [0152], and, -- In act 60, uncertainty is assigned to one or more features. The uncertainty is a distribution of possible values for the feature. For example, the radius may be measured as 0.25 cm, but the accuracy or tolerance in the measurement provides that the radius is between 0.20 cm and 0.30 cm with greater probability for the values closer to 0.25 cm. Any distribution of possible or probable values may be used, such as a normal distribution, a distribution from a study, or from another source…. Using the machine-learnt algorithm, the confidence of the estimated hemodynamic metric is provided. A confidence or probability is provided for one value of the metric. Alternatively, the predictions from the learnt model may also be ranges or confidence intervals within which the predicted quantity is expected. The predicted confidence interval for the patient could be either directly predicted from the model or estimated from a set of similar anatomies from a saved database of synthetic models. --, in [0178]-[0180]).
Re Claim 9, MIHALEF as modified by ITU and BYDLON further disclose wherein the clinical information includes at least one of: demographic information for the aneurysm patient, aneurysm information associated with the imaging information, dimension information for an aneurysm imaged in the imaging information, allergies of the aneurysm patient, medication information for the aneurysm patient, or pre-existing condition information for the aneurysm patient (see MIHALEF: e.g., --[0046] The decision support uses the anatomical and/or physiological knowledge using any of the different machine learning models. For training, many samples with known ground truth (e.g., selections) are used. For the samples, the machine learning model inputs may be real (from patients, or bench-scale measurements), virtual (synthetic or not representative of a given patient or physical bench arrangement), or a combination of the two. For patient data, the input samples may contain the patient images, blood biomarkers, demographics, measurements and/or genetic data. For the ground truth, the input data includes the therapy that was chosen for the patient. Other ground truths may alternatively or additionally be used, such as the recorded outcome at different temporal points (e.g., 30 days, 90 days, etc.) and/or the occurrence or lack thereof of adverse events. …[0048] The machine-learnt predictor, with or without deep learning, is trained to associate the categorical labels (output clinical decision of the selections) to the extracted values of one or more features. The machine- learning uses training data with ground truth to learn to select based on the input vector. The resulting machine-learnt network is a matrix for inputs, weighting, convolution kernels, and/or combinations to output a clinical decision. Using the learned network, the processor inputs the extracted values for features and outputs the selection. ….[0047] Any machine learning or training may be used. A probabilistic boosting tree, support vector machine, neural network, sparse auto-encoding classifier, Bayesian network, or other now known or later developed machine learning may be used. Any semi-supervised, supervised, or unsupervised learning may be used. Hierarchal or other approaches may be used. In one embodiment, the classification is by a machine-learnt classifier learnt with deep learning. As part of identifying features that distinguish between different outcomes, the classifier is also machine learnt. Any deep learning approach or architecture may be used. For example, a convolutional neural network is used….
[0049] The machine-learned network is trained to output a selection, outcome, risk, deployment, or other clinical support information. The information extracted from the model may include the predicted outcomes and risks under different therapy choices. The information may select the therapy configuration, such as (i) start and end locations of the flow diverter, (ii) the number of flow diverters deployed, (iii) the configuration (overlapping or deployment into different branches), (iv) the reference radius of the flow diverter to use, (v) mean and maximum porosity in the neck/non-vessel area, etc. The information output by the machine-learned network may include identification of similar patients and the therapy choices made for the similar patients identified from a patient database and their outcomes.--, in [0044]-[0049]; and, …[0062] In act 23, the decision support processor calculates a porosity of the endovascular device from the deployment. For the flow diverter or other implant, the porosity, in part, controls the flow. In-situ implantation may produce variations in the metal coverage ratio (MCR) and porosity of the device, especially in the aneurysm and adjacent regions. Hemodynamics is consequently affected by such regions, so porosity is a consideration when performing the implantation. By simulating deployment, the porosity that results in general or by location may be determined for estimating outcome.--, in [0060]-[0063];also see ITU: e.g., -- [0102] Other features extracted include parameters for one or more abnormalities of the vessel structure. Abnormal morphology may be characterized by characteristics of calcification, characteristics of the plaque (e.g., fibrous tissue, lipid tissue, necrotic tissue, calcified tissue), characteristics of thrombus, characteristics of diffuse disease, presence of total or sub-total occlusion, presence of myocardial bridging (superficial and/or deep), congenital anomalies of coronary arteries (e.g., anomalous origin of a coronary artery from an abnormal sinus of Valsalva with an inter-arterial course between the great arteries, anomalous origin of one coronary artery from the pulmonary trunk, or others), aneurysmal dilatation and superimposed atherosclerosis, “high take off” coronary artery (e.g., the ostium is several millimeters above the sino-tubular junction (the artery may have a sharp downward angle and runs partially through the aortic wall)), myocardial bridging: superficial and deep, coronary fistula, coronary artery dissection, coronary vasculitis (e.g., rheumatoid arthritis, systemic lupus erythematosus (SLE), or Behçet's disease, Kawasaki disease, polyarteritis nodosa, and/or persisting (post) inflammatory aneurysms), fibromuscular dysplasia, coronary micro embolization, and/or left or right dominance. Additional, different, or fewer abnormality features may be used.--, in [0102], and, -- A local ischemia weight value is estimated independently for each root/branch/leaf segment using geometric features of the segment, such as the reference radius, length, tapering rate and other features…. An average value of healthy radiuses of the entire branch or a part of the branch, an average value of healthy radiuses obtained when excluding the largest x % and the smallest y % of the radius values of the entire branch or a part of the branch, or maximum or minimum value of healthy radii of the entire branch or part of the branch are computed.--, in [0110]-[0112]; and,
-- [0146] The in vitro model 23 and the flow conditions may be modified in numerous ways to generate a large number of setups. For example, the number, position and shape of the occluders is altered. As another example, the resistance at one or more locations is altered. In yet another example, the operation of the pump is altered. The number of side branches and any occlusions may be altered. Other alterations of combinations of different alterations are used to create different models with corresponding features and resulting flow characteristics.--, in [0145]-[0146], and, -- [0152] More than one classifier may be created. Since different types of branches and regions are present in a vessel tree, different classifiers may be machine trained for the different branches and/or regions. For example, different classifiers are trained for main and side branches, bifurcation regions and single branch regions, different types of pathologic regions such as different types of single branch stenotic regions (e.g., focal, long, diffuse, restenosis, or other), different types of bifurcation stenoses (e.g. a separate model for each bifurcation stenosis type in the medina classification), different types of aneurysms, different types of plaque, different types of total and/or sub-total occlusions, stenotic and regurgitant valves, various pathologies of the heart (e.g., past infarct or myopathies), or different types of branches (e.g. in case of coronary arterial trees: LM, LAD, LCx, RCA, Diagonal, OM, or other). Since the training is based on synthetic geometries, a large enough number of training instances may be generated for each of these different classifiers. Another possibility is to divide the geometry into separate segments (e.g. for coronary geometries: proximal LAD, mid LAD, and distal LAD) and to extract the features discussed in the previous sections separately for each segment. Afterwards these features may either be combined into cumulative features or used separately for a single or multiple machine learning algorithms for predicting a hemodynamic metric of interest.--, in [0152]).
Re Claim 10, MIHALEF as modified by ITU and BYDLON further disclose wherein the outcome prediction comprises at least one of: one or more measurements for an aneurysm imaged in the imaging information, or a best predicted size of an intrasaccular device for the aneurysm imaged in the imaging information (see MIHALEF: e.g., --[0026] In stage 16, a machine-learned network predicts outcome based on the simulated deployment, imaging data (e.g., segmented vessel), and/or other patient-specific information. The machine-learned network may be used to output implant information (e.g., type, size, placement, number of devices, configuration of device, etc.) used for the simulation in stage 14. Alternatively or additionally, the machine-learned network may be used to predict outcome or prognosis for a simulated deployment. The decision support system may be trained on outcome data, if available, to predict not only short-term physiological response, but also long-term outcomes and/or costs. The outcome or prognosis may be used to compare or select implant information, or the network directly outputs the selection.
[0027] In stage 18, the simulated deployment, predicted outcome, and/or a calculated hemodynamic characteristic based on the deployment are provided to the clinician. The information may be used to support decisions for planning an endovascular implantation.--, in [0026]-[0027]; and, --[0046] The decision support uses the anatomical and/or physiological knowledge using any of the different machine learning models. For training, many samples with known ground truth (e.g., selections) are used. For the samples, the machine learning model inputs may be real (from patients, or bench-scale measurements), virtual (synthetic or not representative of a given patient or physical bench arrangement), or a combination of the two. For patient data, the input samples may contain the patient images, blood biomarkers, demographics, measurements and/or genetic data. For the ground truth, the input data includes the therapy that was chosen for the patient. Other ground truths may alternatively or additionally be used, such as the recorded outcome at different temporal points (e.g., 30 days, 90 days, etc.) and/or the occurrence or lack thereof of adverse events. …[0048] The machine-learnt predictor, with or without deep learning, is trained to associate the categorical labels (output clinical decision of the selections) to the extracted values of one or more features. The machine- learning uses training data with ground truth to learn to select based on the input vector. The resulting machine-learnt network is a matrix for inputs, weighting, convolution kernels, and/or combinations to output a clinical decision. Using the learned network, the processor inputs the extracted values for features and outputs the selection. ….[0047] Any machine learning or training may be used. A probabilistic boosting tree, support vector machine, neural network, sparse auto-encoding classifier, Bayesian network, or other now known or later developed machine learning may be used. Any semi-supervised, supervised, or unsupervised learning may be used. Hierarchal or other approaches may be used. In one embodiment, the classification is by a machine-learnt classifier learnt with deep learning. As part of identifying features that distinguish between different outcomes, the classifier is also machine learnt. Any deep learning approach or architecture may be used. For example, a convolutional neural network is used….
[0049] The machine-learned network is trained to output a selection, outcome, risk, deployment, or other clinical support information. The information extracted from the model may include the predicted outcomes and risks under different therapy choices. The information may select the therapy configuration, such as (i) start and end locations of the flow diverter, (ii) the number of flow diverters deployed, (iii) the configuration (overlapping or deployment into different branches), (iv) the reference radius of the flow diverter to use, (v) mean and maximum porosity in the neck/non-vessel area, etc. The information output by the machine-learned network may include identification of similar patients and the therapy choices made for the similar patients identified from a patient database and their outcomes.--, in [0044]-[0049]; and, …[0062] In act 23, the decision support processor calculates a porosity of the endovascular device from the deployment. For the flow diverter or other implant, the porosity, in part, controls the flow. In-situ implantation may produce variations in the metal coverage ratio (MCR) and porosity of the device, especially in the aneurysm and adjacent regions. Hemodynamics is consequently affected by such regions, so porosity is a consideration when performing the implantation. By simulating deployment, the porosity that results in general or by location may be determined for estimating outcome.--, in [0060]-[0063];also see ITU: e.g., -- [0102] Other features extracted include parameters for one or more abnormalities of the vessel structure. Abnormal morphology may be characterized by characteristics of calcification, characteristics of the plaque (e.g., fibrous tissue, lipid tissue, necrotic tissue, calcified tissue), characteristics of thrombus, characteristics of diffuse disease, presence of total or sub-total occlusion, presence of myocardial bridging (superficial and/or deep), congenital anomalies of coronary arteries (e.g., anomalous origin of a coronary artery from an abnormal sinus of Valsalva with an inter-arterial course between the great arteries, anomalous origin of one coronary artery from the pulmonary trunk, or others), aneurysmal dilatation and superimposed atherosclerosis, “high take off” coronary artery (e.g., the ostium is several millimeters above the sino-tubular junction (the artery may have a sharp downward angle and runs partially through the aortic wall)), myocardial bridging: superficial and deep, coronary fistula, coronary artery dissection, coronary vasculitis (e.g., rheumatoid arthritis, systemic lupus erythematosus (SLE), or Behçet's disease, Kawasaki disease, polyarteritis nodosa, and/or persisting (post) inflammatory aneurysms), fibromuscular dysplasia, coronary micro embolization, and/or left or right dominance. Additional, different, or fewer abnormality features may be used.--, in [0102], and, -- A local ischemia weight value is estimated independently for each root/branch/leaf segment using geometric features of the segment, such as the reference radius, length, tapering rate and other features…. An average value of healthy radiuses of the entire branch or a part of the branch, an average value of healthy radiuses obtained when excluding the largest x % and the smallest y % of the radius values of the entire branch or a part of the branch, or maximum or minimum value of healthy radii of the entire branch or part of the branch are computed.--, in [0110]-[0112]; and,
-- [0146] The in vitro model 23 and the flow conditions may be modified in numerous ways to generate a large number of setups. For example, the number, position and shape of the occluders is altered. As another example, the resistance at one or more locations is altered. In yet another example, the operation of the pump is altered. The number of side branches and any occlusions may be altered. Other alterations of combinations of different alterations are used to create different models with corresponding features and resulting flow characteristics.--, in [0145]-[0146], and, -- [0152] More than one classifier may be created. Since different types of branches and regions are present in a vessel tree, different classifiers may be machine trained for the different branches and/or regions. For example, different classifiers are trained for main and side branches, bifurcation regions and single branch regions, different types of pathologic regions such as different types of single branch stenotic regions (e.g., focal, long, diffuse, restenosis, or other), different types of bifurcation stenoses (e.g. a separate model for each bifurcation stenosis type in the medina classification), different types of aneurysms, different types of plaque, different types of total and/or sub-total occlusions, stenotic and regurgitant valves, various pathologies of the heart (e.g., past infarct or myopathies), or different types of branches (e.g. in case of coronary arterial trees: LM, LAD, LCx, RCA, Diagonal, OM, or other). Since the training is based on synthetic geometries, a large enough number of training instances may be generated for each of these different classifiers. Another possibility is to divide the geometry into separate segments (e.g. for coronary geometries: proximal LAD, mid LAD, and distal LAD) and to extract the features discussed in the previous sections separately for each segment. Afterwards these features may either be combined into cumulative features or used separately for a single or multiple machine learning algorithms for predicting a hemodynamic metric of interest.--, in [0152]).
Re Claim 11, MIHALEF as modified by ITU and BYDLON further disclose wherein the imaging information includes at least one annotation identifying a region of the aneurysm (see MIHALEF: e.g., --[0026] In stage 16, a machine-learned network predicts outcome based on the simulated deployment, imaging data (e.g., segmented vessel), and/or other patient-specific information. The machine-learned network may be used to output implant information (e.g., type, size, placement, number of devices, configuration of device, etc.) used for the simulation in stage 14. Alternatively or additionally, the machine-learned network may be used to predict outcome or prognosis for a simulated deployment. The decision support system may be trained on outcome data, if available, to predict not only short-term physiological response, but also long-term outcomes and/or costs. The outcome or prognosis may be used to compare or select implant information, or the network directly outputs the selection.
[0027] In stage 18, the simulated deployment, predicted outcome, and/or a calculated hemodynamic characteristic based on the deployment are provided to the clinician. The information may be used to support decisions for planning an endovascular implantation.--, in [0026]-[0027]; and, --[0046] The decision support uses the anatomical and/or physiological knowledge using any of the different machine learning models. For training, many samples with known ground truth (e.g., selections) are used. For the samples, the machine learning model inputs may be real (from patients, or bench-scale measurements), virtual (synthetic or not representative of a given patient or physical bench arrangement), or a combination of the two. For patient data, the input samples may contain the patient images, blood biomarkers, demographics, measurements and/or genetic data. For the ground truth, the input data includes the therapy that was chosen for the patient. Other ground truths may alternatively or additionally be used, such as the recorded outcome at different temporal points (e.g., 30 days, 90 days, etc.) and/or the occurrence or lack thereof of adverse events. …[0048] The machine-learnt predictor, with or without deep learning, is trained to associate the categorical labels (output clinical decision of the selections) to the extracted values of one or more features. The machine- learning uses training data with ground truth to learn to select based on the input vector. The resulting machine-learnt network is a matrix for inputs, weighting, convolution kernels, and/or combinations to output a clinical decision. Using the learned network, the processor inputs the extracted values for features and outputs the selection. ….[0047] Any machine learning or training may be used. A probabilistic boosting tree, support vector machine, neural network, sparse auto-encoding classifier, Bayesian network, or other now known or later developed machine learning may be used. Any semi-supervised, supervised, or unsupervised learning may be used. Hierarchal or other approaches may be used. In one embodiment, the classification is by a machine-learnt classifier learnt with deep learning. As part of identifying features that distinguish between different outcomes, the classifier is also machine learnt. Any deep learning approach or architecture may be used. For example, a convolutional neural network is used….
[0049] The machine-learned network is trained to output a selection, outcome, risk, deployment, or other clinical support information. The information extracted from the model may include the predicted outcomes and risks under different therapy choices. The information may select the therapy configuration, such as (i) start and end locations of the flow diverter, (ii) the number of flow diverters deployed, (iii) the configuration (overlapping or deployment into different branches), (iv) the reference radius of the flow diverter to use, (v) mean and maximum porosity in the neck/non-vessel area, etc. The information output by the machine-learned network may include identification of similar patients and the therapy choices made for the similar patients identified from a patient database and their outcomes.--, in [0044]-[0049]; and, …[0062] In act 23, the decision support processor calculates a porosity of the endovascular device from the deployment. For the flow diverter or other implant, the porosity, in part, controls the flow. In-situ implantation may produce variations in the metal coverage ratio (MCR) and porosity of the device, especially in the aneurysm and adjacent regions. Hemodynamics is consequently affected by such regions, so porosity is a consideration when performing the implantation. By simulating deployment, the porosity that results in general or by location may be determined for estimating outcome.--, in [0060]-[0063];also see ITU: e.g., -- [0102] Other features extracted include parameters for one or more abnormalities of the vessel structure. Abnormal morphology may be characterized by characteristics of calcification, characteristics of the plaque (e.g., fibrous tissue, lipid tissue, necrotic tissue, calcified tissue), characteristics of thrombus, characteristics of diffuse disease, presence of total or sub-total occlusion, presence of myocardial bridging (superficial and/or deep), congenital anomalies of coronary arteries (e.g., anomalous origin of a coronary artery from an abnormal sinus of Valsalva with an inter-arterial course between the great arteries, anomalous origin of one coronary artery from the pulmonary trunk, or others), aneurysmal dilatation and superimposed atherosclerosis, “high take off” coronary artery (e.g., the ostium is several millimeters above the sino-tubular junction (the artery may have a sharp downward angle and runs partially through the aortic wall)), myocardial bridging: superficial and deep, coronary fistula, coronary artery dissection, coronary vasculitis (e.g., rheumatoid arthritis, systemic lupus erythematosus (SLE), or Behçet's disease, Kawasaki disease, polyarteritis nodosa, and/or persisting (post) inflammatory aneurysms), fibromuscular dysplasia, coronary micro embolization, and/or left or right dominance. Additional, different, or fewer abnormality features may be used.--, in [0102], and, -- A local ischemia weight value is estimated independently for each root/branch/leaf segment using geometric features of the segment, such as the reference radius, length, tapering rate and other features…. An average value of healthy radiuses of the entire branch or a part of the branch, an average value of healthy radiuses obtained when excluding the largest x % and the smallest y % of the radius values of the entire branch or a part of the branch, or maximum or minimum value of healthy radii of the entire branch or part of the branch are computed.--, in [0110]-[0112]; and,
-- [0146] The in vitro model 23 and the flow conditions may be modified in numerous ways to generate a large number of setups. For example, the number, position and shape of the occluders is altered. As another example, the resistance at one or more locations is altered. In yet another example, the operation of the pump is altered. The number of side branches and any occlusions may be altered. Other alterations of combinations of different alterations are used to create different models with corresponding features and resulting flow characteristics.--, in [0145]-[0146], and, -- [0152] More than one classifier may be created. Since different types of branches and regions are present in a vessel tree, different classifiers may be machine trained for the different branches and/or regions. For example, different classifiers are trained for main and side branches, bifurcation regions and single branch regions, different types of pathologic regions such as different types of single branch stenotic regions (e.g., focal, long, diffuse, restenosis, or other), different types of bifurcation stenoses (e.g. a separate model for each bifurcation stenosis type in the medina classification), different types of aneurysms, different types of plaque, different types of total and/or sub-total occlusions, stenotic and regurgitant valves, various pathologies of the heart (e.g., past infarct or myopathies), or different types of branches (e.g. in case of coronary arterial trees: LM, LAD, LCx, RCA, Diagonal, OM, or other). Since the training is based on synthetic geometries, a large enough number of training instances may be generated for each of these different classifiers. Another possibility is to divide the geometry into separate segments (e.g. for coronary geometries: proximal LAD, mid LAD, and distal LAD) and to extract the features discussed in the previous sections separately for each segment. Afterwards these features may either be combined into cumulative features or used separately for a single or multiple machine learning algorithms for predicting a hemodynamic metric of interest.--, in [0152]).
Re Claim 12, MIHALEF as modified by ITU and BYDLON further disclose wherein the imaging information includes raw imaging information, and based at least in part on the information stored in the memory, the at least one processor is further configured to:
send, to the user device, an identification of a presence of the aneurysm in the imaging information in addition to the one or more measurements (see MIHALEF: e.g., --[0026] In stage 16, a machine-learned network predicts outcome based on the simulated deployment, imaging data (e.g., segmented vessel), and/or other patient-specific information. The machine-learned network may be used to output implant information (e.g., type, size, placement, number of devices, configuration of device, etc.) used for the simulation in stage 14. Alternatively or additionally, the machine-learned network may be used to predict outcome or prognosis for a simulated deployment. The decision support system may be trained on outcome data, if available, to predict not only short-term physiological response, but also long-term outcomes and/or costs. The outcome or prognosis may be used to compare or select implant information, or the network directly outputs the selection.
[0027] In stage 18, the simulated deployment, predicted outcome, and/or a calculated hemodynamic characteristic based on the deployment are provided to the clinician. The information may be used to support decisions for planning an endovascular implantation.--, in [0026]-[0027]; and, --[0046] The decision support uses the anatomical and/or physiological knowledge using any of the different machine learning models. For training, many samples with known ground truth (e.g., selections) are used. For the samples, the machine learning model inputs may be real (from patients, or bench-scale measurements), virtual (synthetic or not representative of a given patient or physical bench arrangement), or a combination of the two. For patient data, the input samples may contain the patient images, blood biomarkers, demographics, measurements and/or genetic data. For the ground truth, the input data includes the therapy that was chosen for the patient. Other ground truths may alternatively or additionally be used, such as the recorded outcome at different temporal points (e.g., 30 days, 90 days, etc.) and/or the occurrence or lack thereof of adverse events. …[0048] The machine-learnt predictor, with or without deep learning, is trained to associate the categorical labels (output clinical decision of the selections) to the extracted values of one or more features. The machine- learning uses training data with ground truth to learn to select based on the input vector. The resulting machine-learnt network is a matrix for inputs, weighting, convolution kernels, and/or combinations to output a clinical decision. Using the learned network, the processor inputs the extracted values for features and outputs the selection. ….[0047] Any machine learning or training may be used. A probabilistic boosting tree, support vector machine, neural network, sparse auto-encoding classifier, Bayesian network, or other now known or later developed machine learning may be used. Any semi-supervised, supervised, or unsupervised learning may be used. Hierarchal or other approaches may be used. In one embodiment, the classification is by a machine-learnt classifier learnt with deep learning. As part of identifying features that distinguish between different outcomes, the classifier is also machine learnt. Any deep learning approach or architecture may be used. For example, a convolutional neural network is used….
[0049] The machine-learned network is trained to output a selection, outcome, risk, deployment, or other clinical support information. The information extracted from the model may include the predicted outcomes and risks under different therapy choices. The information may select the therapy configuration, such as (i) start and end locations of the flow diverter, (ii) the number of flow diverters deployed, (iii) the configuration (overlapping or deployment into different branches), (iv) the reference radius of the flow diverter to use, (v) mean and maximum porosity in the neck/non-vessel area, etc. The information output by the machine-learned network may include identification of similar patients and the therapy choices made for the similar patients identified from a patient database and their outcomes.--, in [0044]-[0049]; and, …[0062] In act 23, the decision support processor calculates a porosity of the endovascular device from the deployment. For the flow diverter or other implant, the porosity, in part, controls the flow. In-situ implantation may produce variations in the metal coverage ratio (MCR) and porosity of the device, especially in the aneurysm and adjacent regions. Hemodynamics is consequently affected by such regions, so porosity is a consideration when performing the implantation. By simulating deployment, the porosity that results in general or by location may be determined for estimating outcome.--, in [0060]-[0063];also see ITU: e.g., -- [0102] Other features extracted include parameters for one or more abnormalities of the vessel structure. Abnormal morphology may be characterized by characteristics of calcification, characteristics of the plaque (e.g., fibrous tissue, lipid tissue, necrotic tissue, calcified tissue), characteristics of thrombus, characteristics of diffuse disease, presence of total or sub-total occlusion, presence of myocardial bridging (superficial and/or deep), congenital anomalies of coronary arteries (e.g., anomalous origin of a coronary artery from an abnormal sinus of Valsalva with an inter-arterial course between the great arteries, anomalous origin of one coronary artery from the pulmonary trunk, or others), aneurysmal dilatation and superimposed atherosclerosis, “high take off” coronary artery (e.g., the ostium is several millimeters above the sino-tubular junction (the artery may have a sharp downward angle and runs partially through the aortic wall)), myocardial bridging: superficial and deep, coronary fistula, coronary artery dissection, coronary vasculitis (e.g., rheumatoid arthritis, systemic lupus erythematosus (SLE), or Behçet's disease, Kawasaki disease, polyarteritis nodosa, and/or persisting (post) inflammatory aneurysms), fibromuscular dysplasia, coronary micro embolization, and/or left or right dominance. Additional, different, or fewer abnormality features may be used.--, in [0102], and, -- A local ischemia weight value is estimated independently for each root/branch/leaf segment using geometric features of the segment, such as the reference radius, length, tapering rate and other features…. An average value of healthy radiuses of the entire branch or a part of the branch, an average value of healthy radiuses obtained when excluding the largest x % and the smallest y % of the radius values of the entire branch or a part of the branch, or maximum or minimum value of healthy radii of the entire branch or part of the branch are computed.--, in [0110]-[0112]; and,
-- [0146] The in vitro model 23 and the flow conditions may be modified in numerous ways to generate a large number of setups. For example, the number, position and shape of the occluders is altered. As another example, the resistance at one or more locations is altered. In yet another example, the operation of the pump is altered. The number of side branches and any occlusions may be altered. Other alterations of combinations of different alterations are used to create different models with corresponding features and resulting flow characteristics.--, in [0145]-[0146], and, -- [0152] More than one classifier may be created. Since different types of branches and regions are present in a vessel tree, different classifiers may be machine trained for the different branches and/or regions. For example, different classifiers are trained for main and side branches, bifurcation regions and single branch regions, different types of pathologic regions such as different types of single branch stenotic regions (e.g., focal, long, diffuse, restenosis, or other), different types of bifurcation stenoses (e.g. a separate model for each bifurcation stenosis type in the medina classification), different types of aneurysms, different types of plaque, different types of total and/or sub-total occlusions, stenotic and regurgitant valves, various pathologies of the heart (e.g., past infarct or myopathies), or different types of branches (e.g. in case of coronary arterial trees: LM, LAD, LCx, RCA, Diagonal, OM, or other). Since the training is based on synthetic geometries, a large enough number of training instances may be generated for each of these different classifiers. Another possibility is to divide the geometry into separate segments (e.g. for coronary geometries: proximal LAD, mid LAD, and distal LAD) and to extract the features discussed in the previous sections separately for each segment. Afterwards these features may either be combined into cumulative features or used separately for a single or multiple machine learning algorithms for predicting a hemodynamic metric of interest.--, in [0152]).
Re Claim 13, MIHALEF as modified by ITU and BYDLON further disclose wherein the identification includes a contour outlining the aneurysm sac imaged in the imaging information (see MIHALEF: e.g., --[0026] In stage 16, a machine-learned network predicts outcome based on the simulated deployment, imaging data (e.g., segmented vessel), and/or other patient-specific information. The machine-learned network may be used to output implant information (e.g., type, size, placement, number of devices, configuration of device, etc.) used for the simulation in stage 14. Alternatively or additionally, the machine-learned network may be used to predict outcome or prognosis for a simulated deployment. The decision support system may be trained on outcome data, if available, to predict not only short-term physiological response, but also long-term outcomes and/or costs. The outcome or prognosis may be used to compare or select implant information, or the network directly outputs the selection.
[0027] In stage 18, the simulated deployment, predicted outcome, and/or a calculated hemodynamic characteristic based on the deployment are provided to the clinician. The information may be used to support decisions for planning an endovascular implantation.--, in [0026]-[0027]; and –[0036] The vessel may be segmented, or other tissue masked. The segmentation provides locations of the vessel. Alternatively, boundary or another vessel detection is applied to identify the interior and/or exterior surface of the vessel. The vessel may be identified by fitting a model, such as anatomical or statistical shape model, to the scan data. A machine-learned network may detect the vessel or vessel locations from the scan data.--, in [0036]).
Re Claim 14, MIHALEF as modified by ITU and BYDLON further disclose wherein, based at least in part on the information stored in the memory, the at least one processor is further configured to:
send, to the user device, an identification of at least one treatment device for the implant in the aneurysm sac of the aneurysm patient based on the at least one of the imaging information or the clinical information for the aneurysm patient, the at least one treatment device identified based on having a highest predicted likelihood of complete occlusion of the aneurysm sac imaged in the imaging information for the aneurysm patient from a set of potential treatment devices (see ITU: e.g., -- FIG. 8 shows an example of the normal distribution of the root radius of a coronary left arterial tree. Known, estimated, or standard normal distributions may be used. The synthetic examples generated are assigned the value for the root radius of the coronary left arterial tree based on the distribution (e.g., probability of a given value per example assigned using the distribution).--, in [0084], and,
-- [0102] Other features extracted include parameters for one or more abnormalities of the vessel structure. Abnormal morphology may be characterized by characteristics of calcification, characteristics of the plaque (e.g., fibrous tissue, lipid tissue, necrotic tissue, calcified tissue), characteristics of thrombus, characteristics of diffuse disease, presence of total or sub-total occlusion, presence of myocardial bridging (superficial and/or deep), congenital anomalies of coronary arteries (e.g., anomalous origin of a coronary artery from an abnormal sinus of Valsalva with an inter-arterial course between the great arteries, anomalous origin of one coronary artery from the pulmonary trunk, or others), aneurysmal dilatation and superimposed atherosclerosis, “high take off” coronary artery (e.g., the ostium is several millimeters above the sino-tubular junction (the artery may have a sharp downward angle and runs partially through the aortic wall)), myocardial bridging: superficial and deep, coronary fistula, coronary artery dissection, coronary vasculitis (e.g., rheumatoid arthritis, systemic lupus erythematosus (SLE), or Behçet's disease, Kawasaki disease, polyarteritis nodosa, and/or persisting (post) inflammatory aneurysms), fibromuscular dysplasia, coronary micro embolization, and/or left or right dominance. Additional, different, or fewer abnormality features may be used.--, in [0102], and, -- A local ischemia weight value is estimated independently for each root/branch/leaf segment using geometric features of the segment, such as the reference radius, length, tapering rate and other features…. An average value of healthy radiuses of the entire branch or a part of the branch, an average value of healthy radiuses obtained when excluding the largest x % and the smallest y % of the radius values of the entire branch or a part of the branch, or maximum or minimum value of healthy radii of the entire branch or part of the branch are computed.--, in [0110]-[0112]; and,
-- [0146] The in vitro model 23 and the flow conditions may be modified in numerous ways to generate a large number of setups. For example, the number, position and shape of the occluders is altered. As another example, the resistance at one or more locations is altered. In yet another example, the operation of the pump is altered. The number of side branches and any occlusions may be altered. Other alterations of combinations of different alterations are used to create different models with corresponding features and resulting flow characteristics.--, in [0145]-[0146], and, -- [0152] More than one classifier may be created. Since different types of branches and regions are present in a vessel tree, different classifiers may be machine trained for the different branches and/or regions. For example, different classifiers are trained for main and side branches, bifurcation regions and single branch regions, different types of pathologic regions such as different types of single branch stenotic regions (e.g., focal, long, diffuse, restenosis, or other), different types of bifurcation stenoses (e.g. a separate model for each bifurcation stenosis type in the medina classification), different types of aneurysms, different types of plaque, different types of total and/or sub-total occlusions, stenotic and regurgitant valves, various pathologies of the heart (e.g., past infarct or myopathies), or different types of branches (e.g. in case of coronary arterial trees: LM, LAD, LCx, RCA, Diagonal, OM, or other). Since the training is based on synthetic geometries, a large enough number of training instances may be generated for each of these different classifiers. Another possibility is to divide the geometry into separate segments (e.g. for coronary geometries: proximal LAD, mid LAD, and distal LAD) and to extract the features discussed in the previous sections separately for each segment. Afterwards these features may either be combined into cumulative features or used separately for a single or multiple machine learning algorithms for predicting a hemodynamic metric of interest.--, in [0152], and, -- In act 60, uncertainty is assigned to one or more features. The uncertainty is a distribution of possible values for the feature. For example, the radius may be measured as 0.25 cm, but the accuracy or tolerance in the measurement provides that the radius is between 0.20 cm and 0.30 cm with greater probability for the values closer to 0.25 cm. Any distribution of possible or probable values may be used, such as a normal distribution, a distribution from a study, or from another source…. Using the machine-learnt algorithm, the confidence of the estimated hemodynamic metric is provided. A confidence or probability is provided for one value of the metric. Alternatively, the predictions from the learnt model may also be ranges or confidence intervals within which the predicted quantity is expected. The predicted confidence interval for the patient could be either directly predicted from the model or estimated from a set of similar anatomies from a saved database of synthetic models. --, in [0178]-[0180]).
Re Claim 15, MIHALEF as modified by ITU and BYDLON further disclose wherein the identification includes a list of multiple treatment devices and a respective outcome prediction associated with each treatment device in the list of multiple treatment devices (see MIHALEF: e.g., --[0026] In stage 16, a machine-learned network predicts outcome based on the simulated deployment, imaging data (e.g., segmented vessel), and/or other patient-specific information. The machine-learned network may be used to output implant information (e.g., type, size, placement, number of devices, configuration of device, etc.) used for the simulation in stage 14. Alternatively or additionally, the machine-learned network may be used to predict outcome or prognosis for a simulated deployment. The decision support system may be trained on outcome data, if available, to predict not only short-term physiological response, but also long-term outcomes and/or costs. The outcome or prognosis may be used to compare or select implant information, or the network directly outputs the selection.
[0027] In stage 18, the simulated deployment, predicted outcome, and/or a calculated hemodynamic characteristic based on the deployment are provided to the clinician. The information may be used to support decisions for planning an endovascular implantation.--, in [0026]-[0027]; and, --[0046] The decision support uses the anatomical and/or physiological knowledge using any of the different machine learning models. For training, many samples with known ground truth (e.g., selections) are used. For the samples, the machine learning model inputs may be real (from patients, or bench-scale measurements), virtual (synthetic or not representative of a given patient or physical bench arrangement), or a combination of the two. For patient data, the input samples may contain the patient images, blood biomarkers, demographics, measurements and/or genetic data. For the ground truth, the input data includes the therapy that was chosen for the patient. Other ground truths may alternatively or additionally be used, such as the recorded outcome at different temporal points (e.g., 30 days, 90 days, etc.) and/or the occurrence or lack thereof of adverse events. …[0048] The machine-learnt predictor, with or without deep learning, is trained to associate the categorical labels (output clinical decision of the selections) to the extracted values of one or more features. The machine- learning uses training data with ground truth to learn to select based on the input vector. The resulting machine-learnt network is a matrix for inputs, weighting, convolution kernels, and/or combinations to output a clinical decision. Using the learned network, the processor inputs the extracted values for features and outputs the selection. ….[0047] Any machine learning or training may be used. A probabilistic boosting tree, support vector machine, neural network, sparse auto-encoding classifier, Bayesian network, or other now known or later developed machine learning may be used. Any semi-supervised, supervised, or unsupervised learning may be used. Hierarchal or other approaches may be used. In one embodiment, the classification is by a machine-learnt classifier learnt with deep learning. As part of identifying features that distinguish between different outcomes, the classifier is also machine learnt. Any deep learning approach or architecture may be used. For example, a convolutional neural network is used….
[0049] The machine-learned network is trained to output a selection, outcome, risk, deployment, or other clinical support information. The information extracted from the model may include the predicted outcomes and risks under different therapy choices. The information may select the therapy configuration, such as (i) start and end locations of the flow diverter, (ii) the number of flow diverters deployed, (iii) the configuration (overlapping or deployment into different branches), (iv) the reference radius of the flow diverter to use, (v) mean and maximum porosity in the neck/non-vessel area, etc. The information output by the machine-learned network may include identification of similar patients and the therapy choices made for the similar patients identified from a patient database and their outcomes.--, in [0044]-[0049]; and, …[0062] In act 23, the decision support processor calculates a porosity of the endovascular device from the deployment. For the flow diverter or other implant, the porosity, in part, controls the flow. In-situ implantation may produce variations in the metal coverage ratio (MCR) and porosity of the device, especially in the aneurysm and adjacent regions. Hemodynamics is consequently affected by such regions, so porosity is a consideration when performing the implantation. By simulating deployment, the porosity that results in general or by location may be determined for estimating outcome.--, in [0060]-[0063];also see ITU: e.g., -- [0102] Other features extracted include parameters for one or more abnormalities of the vessel structure. Abnormal morphology may be characterized by characteristics of calcification, characteristics of the plaque (e.g., fibrous tissue, lipid tissue, necrotic tissue, calcified tissue), characteristics of thrombus, characteristics of diffuse disease, presence of total or sub-total occlusion, presence of myocardial bridging (superficial and/or deep), congenital anomalies of coronary arteries (e.g., anomalous origin of a coronary artery from an abnormal sinus of Valsalva with an inter-arterial course between the great arteries, anomalous origin of one coronary artery from the pulmonary trunk, or others), aneurysmal dilatation and superimposed atherosclerosis, “high take off” coronary artery (e.g., the ostium is several millimeters above the sino-tubular junction (the artery may have a sharp downward angle and runs partially through the aortic wall)), myocardial bridging: superficial and deep, coronary fistula, coronary artery dissection, coronary vasculitis (e.g., rheumatoid arthritis, systemic lupus erythematosus (SLE), or Behçet's disease, Kawasaki disease, polyarteritis nodosa, and/or persisting (post) inflammatory aneurysms), fibromuscular dysplasia, coronary micro embolization, and/or left or right dominance. Additional, different, or fewer abnormality features may be used.--, in [0102], and, -- A local ischemia weight value is estimated independently for each root/branch/leaf segment using geometric features of the segment, such as the reference radius, length, tapering rate and other features…. An average value of healthy radiuses of the entire branch or a part of the branch, an average value of healthy radiuses obtained when excluding the largest x % and the smallest y % of the radius values of the entire branch or a part of the branch, or maximum or minimum value of healthy radii of the entire branch or part of the branch are computed.--, in [0110]-[0112]; and,
-- [0146] The in vitro model 23 and the flow conditions may be modified in numerous ways to generate a large number of setups. For example, the number, position and shape of the occluders is altered. As another example, the resistance at one or more locations is altered. In yet another example, the operation of the pump is altered. The number of side branches and any occlusions may be altered. Other alterations of combinations of different alterations are used to create different models with corresponding features and resulting flow characteristics.--, in [0145]-[0146], and, -- [0152] More than one classifier may be created. Since different types of branches and regions are present in a vessel tree, different classifiers may be machine trained for the different branches and/or regions. For example, different classifiers are trained for main and side branches, bifurcation regions and single branch regions, different types of pathologic regions such as different types of single branch stenotic regions (e.g., focal, long, diffuse, restenosis, or other), different types of bifurcation stenoses (e.g. a separate model for each bifurcation stenosis type in the medina classification), different types of aneurysms, different types of plaque, different types of total and/or sub-total occlusions, stenotic and regurgitant valves, various pathologies of the heart (e.g., past infarct or myopathies), or different types of branches (e.g. in case of coronary arterial trees: LM, LAD, LCx, RCA, Diagonal, OM, or other). Since the training is based on synthetic geometries, a large enough number of training instances may be generated for each of these different classifiers. Another possibility is to divide the geometry into separate segments (e.g. for coronary geometries: proximal LAD, mid LAD, and distal LAD) and to extract the features discussed in the previous sections separately for each segment. Afterwards these features may either be combined into cumulative features or used separately for a single or multiple machine learning algorithms for predicting a hemodynamic metric of interest.--, in [0152]).
Re Claim 16, MIHALEF as modified by ITU and BYDLON further disclose wherein the list of multiple treatment devices include different sizes of a same type of intrasaccular implant device having a most favorable outcome prediction from the set of potential treatment devices (see MIHALEF: e.g., --[0021] Aneurysm treatment planning in clinical practice depends on acquiring suitable medical images describing the pathologic vascular region. Based on the imaging data in combination with patient demographics (age, gender, weight, body mass index, etc.), past medical history (diabetes, current list of medications, previous stroke, etc.) and/or blood biomarkers (e.g., blood counts, clotting factors, hematocrit, blood viscosity, glucose level, etc.), a patient-individualized treatment strategy is set up and performed. The decision support for the treatment planning process is based on a systemic analysis of the above-mentioned factors along with simulated treatment of the patient’s aneurysm using a virtual implantation of a flow-diverting stent. Only the data acquired prior to the treatment planning (medical images, blood biomarkers, patient medical history and demographics) is used for further processing. The model may also utilize simulated outcomes using machine learning or physics-based simulators, such as computational fluid dynamics models. For such simulations, an anatomical model may be extracted using either manual, semi-automatic or automatic segmentation algorithms. The virtual deployment system is integrated into the decision support. [0022] Figure 1 shows a flow chart or outline of one embodiment of the decision support for endovascular procedure planning. The outline illustrates a workflow, including device modeling (fitting the device characteristics to a physics model), deployment of the device model in a representation of the vessel of the patient, and outcome prediction based on the deployment.--, in [0021]-[0022], and, --[0026] In stage 16, a machine-learned network predicts outcome based on the simulated deployment, imaging data (e.g., segmented vessel), and/or other patient-specific information. The machine-learned network may be used to output implant information (e.g., type, size, placement, number of devices, configuration of device, etc.) used for the simulation in stage 14. Alternatively or additionally, the machine-learned network may be used to predict outcome or prognosis for a simulated deployment. The decision support system may be trained on outcome data, if available, to predict not only short-term physiological response, but also long-term outcomes and/or costs. The outcome or prognosis may be used to compare or select implant information, or the network directly outputs the selection.
[0027] In stage 18, the simulated deployment, predicted outcome, and/or a calculated hemodynamic characteristic based on the deployment are provided to the clinician. The information may be used to support decisions for planning an endovascular implantation.--, in [0026]-[0027]; also see ITU: e.g., -- FIG. 8 shows an example of the normal distribution of the root radius of a coronary left arterial tree. Known, estimated, or standard normal distributions may be used. The synthetic examples generated are assigned the value for the root radius of the coronary left arterial tree based on the distribution (e.g., probability of a given value per example assigned using the distribution).--, in [0084], and,
-- [0102] Other features extracted include parameters for one or more abnormalities of the vessel structure. Abnormal morphology may be characterized by characteristics of calcification, characteristics of the plaque (e.g., fibrous tissue, lipid tissue, necrotic tissue, calcified tissue), characteristics of thrombus, characteristics of diffuse disease, presence of total or sub-total occlusion, presence of myocardial bridging (superficial and/or deep), congenital anomalies of coronary arteries (e.g., anomalous origin of a coronary artery from an abnormal sinus of Valsalva with an inter-arterial course between the great arteries, anomalous origin of one coronary artery from the pulmonary trunk, or others), aneurysmal dilatation and superimposed atherosclerosis, “high take off” coronary artery (e.g., the ostium is several millimeters above the sino-tubular junction (the artery may have a sharp downward angle and runs partially through the aortic wall)), myocardial bridging: superficial and deep, coronary fistula, coronary artery dissection, coronary vasculitis (e.g., rheumatoid arthritis, systemic lupus erythematosus (SLE), or Behçet's disease, Kawasaki disease, polyarteritis nodosa, and/or persisting (post) inflammatory aneurysms), fibromuscular dysplasia, coronary micro embolization, and/or left or right dominance. Additional, different, or fewer abnormality features may be used.--, in [0102], and, -- A local ischemia weight value is estimated independently for each root/branch/leaf segment using geometric features of the segment, such as the reference radius, length, tapering rate and other features…. An average value of healthy radiuses of the entire branch or a part of the branch, an average value of healthy radiuses obtained when excluding the largest x % and the smallest y % of the radius values of the entire branch or a part of the branch, or maximum or minimum value of healthy radii of the entire branch or part of the branch are computed.--, in [0110]-[0112]; and,
-- [0146] The in vitro model 23 and the flow conditions may be modified in numerous ways to generate a large number of setups. For example, the number, position and shape of the occluders is altered. As another example, the resistance at one or more locations is altered. In yet another example, the operation of the pump is altered. The number of side branches and any occlusions may be altered. Other alterations of combinations of different alterations are used to create different models with corresponding features and resulting flow characteristics.--, in [0145]-[0146], and, -- [0152] More than one classifier may be created. Since different types of branches and regions are present in a vessel tree, different classifiers may be machine trained for the different branches and/or regions. For example, different classifiers are trained for main and side branches, bifurcation regions and single branch regions, different types of pathologic regions such as different types of single branch stenotic regions (e.g., focal, long, diffuse, restenosis, or other), different types of bifurcation stenoses (e.g. a separate model for each bifurcation stenosis type in the medina classification), different types of aneurysms, different types of plaque, different types of total and/or sub-total occlusions, stenotic and regurgitant valves, various pathologies of the heart (e.g., past infarct or myopathies), or different types of branches (e.g. in case of coronary arterial trees: LM, LAD, LCx, RCA, Diagonal, OM, or other). Since the training is based on synthetic geometries, a large enough number of training instances may be generated for each of these different classifiers. Another possibility is to divide the geometry into separate segments (e.g. for coronary geometries: proximal LAD, mid LAD, and distal LAD) and to extract the features discussed in the previous sections separately for each segment. Afterwards these features may either be combined into cumulative features or used separately for a single or multiple machine learning algorithms for predicting a hemodynamic metric of interest.--, in [0152], and, -- In act 60, uncertainty is assigned to one or more features. The uncertainty is a distribution of possible values for the feature. For example, the radius may be measured as 0.25 cm, but the accuracy or tolerance in the measurement provides that the radius is between 0.20 cm and 0.30 cm with greater probability for the values closer to 0.25 cm. Any distribution of possible or probable values may be used, such as a normal distribution, a distribution from a study, or from another source…. Using the machine-learnt algorithm, the confidence of the estimated hemodynamic metric is provided. A confidence or probability is provided for one value of the metric. Alternatively, the predictions from the learnt model may also be ranges or confidence intervals within which the predicted quantity is expected. The predicted confidence interval for the patient could be either directly predicted from the model or estimated from a set of similar anatomies from a saved database of synthetic models. --, in [0178]-[0180]).
Re Claim 17, MIHALEF as modified by ITU and BYDLON further disclose wherein the list of multiple treatment devices include different types of aneurysm treatment devices (see MIHALEF: e.g., --the user may indicate the starting and ending points on an image of the vessel or vessel model (e.g., 3D mesh rendered to 2D display image). Any locations of packing to increase porosity may be indicated. Other user inputs of the device and/or placement may be used, such as configuration for multiple devices…[0044]… the decision support processor selects. A selection function may be used, such as relating one or more different patient- specific characteristics to characteristics of or specific ones of the available devices. In one embodiment, a machine-learned network selects the vascular implant and/or placement of the vascular implant….the machine learning model inputs may be real (from patients, or bench-scale measurements), virtual (synthetic or not representative of a given patient or physical bench arrangement), or a combination of the two. For patient data, the input samples may contain the patient images, blood biomarkers, demographics, measurements and/or genetic data. For the ground truth, the input data includes the therapy that was chosen for the patient. Other ground truths may alternatively or additionally be used, such as the recorded outcome at different temporal points (e.g., 30 days, 90 days, etc.) and/or the occurrence or lack thereof of adverse events.--, in [0043]-[0046]; and, --the implant model is bent to follow the curvature of the vessel centerline. In a last stage, the release of the crimping is simulated so the implant model expands and fills the lumen. Other non- clinical-based simulation may be used, such as outputting a deployment and corresponding deformation from a machine-learned network based on inputs of the scan data, vessel model, patient-specific information, and/or device model. …[0062] In act 23, the decision support processor calculates a porosity of the endovascular device from the deployment. For the flow diverter or other implant, the porosity, in part, controls the flow. In-situ implantation may produce variations in the metal coverage ratio (MCR) and porosity of the device, especially in the aneurysm and adjacent regions. Hemodynamics is consequently affected by such regions, so porosity is a consideration when performing the implantation. By simulating deployment, the porosity that results in general or by location may be determined for estimating outcome.--, in [0060]-[0063]; also see: -- for patients with aneurysms, the choices could be coiling the aneurysm or placement of any one of different models of flow diverters with multiple possible implantation configurations. The clinician decides if implantation of endovascular devices like flow diverters or stents would be beneficial to the patient, and if so, make an optimal choice of the device type and its parameters (e.g. diameter, length, porosity, metal coverage area, material mechanical properties, etc.). In the case of patients with aneurysms--, in [0001], and, -- The decision support for the treatment planning process is based on a systemic analysis of the above-mentioned factors along with simulated treatment of the patient’s aneurysm using a virtual implantation of a flow-diverting stent. Only the data acquired prior to the treatment planning (medical images, blood biomarkers, patient medical history and demographics) is used for further processing. The model may also utilize simulated outcomes using machine learning or physics-based simulators, such as computational fluid dynamics models. For such simulations, an anatomical model may be extracted using either manual, semi-automatic or automatic segmentation algorithms. The virtual deployment system is integrated into the decision support. [0022] Figure 1 shows a flow chart or outline of one embodiment of the decision support for endovascular procedure planning. The outline illustrates a workflow, including device modeling (fitting the device characteristics to a physics model), deployment of the device model in a representation of the vessel of the patient, and outcome prediction based on the deployment.--, in [0021]-[0022]; also see ITU: e.g., -- [0146] The in vitro model 23 and the flow conditions may be modified in numerous ways to generate a large number of setups. For example, the number, position and shape of the occluders is altered. As another example, the resistance at one or more locations is altered. In yet another example, the operation of the pump is altered. The number of side branches and any occlusions may be altered. Other alterations of combinations of different alterations are used to create different models with corresponding features and resulting flow characteristics.--, in [0145]-[0146], and, -- [0152] More than one classifier may be created. Since different types of branches and regions are present in a vessel tree, different classifiers may be machine trained for the different branches and/or regions. For example, different classifiers are trained for main and side branches, bifurcation regions and single branch regions, different types of pathologic regions such as different types of single branch stenotic regions (e.g., focal, long, diffuse, restenosis, or other), different types of bifurcation stenoses (e.g. a separate model for each bifurcation stenosis type in the medina classification), different types of aneurysms, different types of plaque, different types of total and/or sub-total occlusions, stenotic and regurgitant valves, various pathologies of the heart (e.g., past infarct or myopathies), or different types of branches (e.g. in case of coronary arterial trees: LM, LAD, LCx, RCA, Diagonal, OM, or other). Since the training is based on synthetic geometries, a large enough number of training instances may be generated for each of these different classifiers. Another possibility is to divide the geometry into separate segments (e.g. for coronary geometries: proximal LAD, mid LAD, and distal LAD) and to extract the features discussed in the previous sections separately for each segment. Afterwards these features may either be combined into cumulative features or used separately for a single or multiple machine learning algorithms for predicting a hemodynamic metric of interest.--, in [0152]).
Re Claim 18, MIHALEF as modified by ITU and BYDLON further disclose wherein the outcome prediction comprises a size for an intrasaccular device having a highest likelihood of complete occlusion (see MIHALEF: e.g., --[0021] Aneurysm treatment planning in clinical practice depends on acquiring suitable medical images describing the pathologic vascular region. Based on the imaging data in combination with patient demographics (age, gender, weight, body mass index, etc.), past medical history (diabetes, current list of medications, previous stroke, etc.) and/or blood biomarkers (e.g., blood counts, clotting factors, hematocrit, blood viscosity, glucose level, etc.), a patient-individualized treatment strategy is set up and performed. The decision support for the treatment planning process is based on a systemic analysis of the above-mentioned factors along with simulated treatment of the patient’s aneurysm using a virtual implantation of a flow-diverting stent. Only the data acquired prior to the treatment planning (medical images, blood biomarkers, patient medical history and demographics) is used for further processing. The model may also utilize simulated outcomes using machine learning or physics-based simulators, such as computational fluid dynamics models. For such simulations, an anatomical model may be extracted using either manual, semi-automatic or automatic segmentation algorithms. The virtual deployment system is integrated into the decision support. [0022] Figure 1 shows a flow chart or outline of one embodiment of the decision support for endovascular procedure planning. The outline illustrates a workflow, including device modeling (fitting the device characteristics to a physics model), deployment of the device model in a representation of the vessel of the patient, and outcome prediction based on the deployment.--, in [0021]-[0022], and, --[0026] In stage 16, a machine-learned network predicts outcome based on the simulated deployment, imaging data (e.g., segmented vessel), and/or other patient-specific information. The machine-learned network may be used to output implant information (e.g., type, size, placement, number of devices, configuration of device, etc.) used for the simulation in stage 14. Alternatively or additionally, the machine-learned network may be used to predict outcome or prognosis for a simulated deployment. The decision support system may be trained on outcome data, if available, to predict not only short-term physiological response, but also long-term outcomes and/or costs. The outcome or prognosis may be used to compare or select implant information, or the network directly outputs the selection.
[0027] In stage 18, the simulated deployment, predicted outcome, and/or a calculated hemodynamic characteristic based on the deployment are provided to the clinician. The information may be used to support decisions for planning an endovascular implantation.--, in [0026]-[0027]; also see ITU: e.g., -- FIG. 8 shows an example of the normal distribution of the root radius of a coronary left arterial tree. Known, estimated, or standard normal distributions may be used. The synthetic examples generated are assigned the value for the root radius of the coronary left arterial tree based on the distribution (e.g., probability of a given value per example assigned using the distribution).--, in [0084], and,
-- [0102] Other features extracted include parameters for one or more abnormalities of the vessel structure. Abnormal morphology may be characterized by characteristics of calcification, characteristics of the plaque (e.g., fibrous tissue, lipid tissue, necrotic tissue, calcified tissue), characteristics of thrombus, characteristics of diffuse disease, presence of total or sub-total occlusion, presence of myocardial bridging (superficial and/or deep), congenital anomalies of coronary arteries (e.g., anomalous origin of a coronary artery from an abnormal sinus of Valsalva with an inter-arterial course between the great arteries, anomalous origin of one coronary artery from the pulmonary trunk, or others), aneurysmal dilatation and superimposed atherosclerosis, “high take off” coronary artery (e.g., the ostium is several millimeters above the sino-tubular junction (the artery may have a sharp downward angle and runs partially through the aortic wall)), myocardial bridging: superficial and deep, coronary fistula, coronary artery dissection, coronary vasculitis (e.g., rheumatoid arthritis, systemic lupus erythematosus (SLE), or Behçet's disease, Kawasaki disease, polyarteritis nodosa, and/or persisting (post) inflammatory aneurysms), fibromuscular dysplasia, coronary micro embolization, and/or left or right dominance. Additional, different, or fewer abnormality features may be used.--, in [0102], and, -- A local ischemia weight value is estimated independently for each root/branch/leaf segment using geometric features of the segment, such as the reference radius, length, tapering rate and other features…. An average value of healthy radiuses of the entire branch or a part of the branch, an average value of healthy radiuses obtained when excluding the largest x % and the smallest y % of the radius values of the entire branch or a part of the branch, or maximum or minimum value of healthy radii of the entire branch or part of the branch are computed.--, in [0110]-[0112]; and,
-- [0146] The in vitro model 23 and the flow conditions may be modified in numerous ways to generate a large number of setups. For example, the number, position and shape of the occluders is altered. As another example, the resistance at one or more locations is altered. In yet another example, the operation of the pump is altered. The number of side branches and any occlusions may be altered. Other alterations of combinations of different alterations are used to create different models with corresponding features and resulting flow characteristics.--, in [0145]-[0146], and, -- [0152] More than one classifier may be created. Since different types of branches and regions are present in a vessel tree, different classifiers may be machine trained for the different branches and/or regions. For example, different classifiers are trained for main and side branches, bifurcation regions and single branch regions, different types of pathologic regions such as different types of single branch stenotic regions (e.g., focal, long, diffuse, restenosis, or other), different types of bifurcation stenoses (e.g. a separate model for each bifurcation stenosis type in the medina classification), different types of aneurysms, different types of plaque, different types of total and/or sub-total occlusions, stenotic and regurgitant valves, various pathologies of the heart (e.g., past infarct or myopathies), or different types of branches (e.g. in case of coronary arterial trees: LM, LAD, LCx, RCA, Diagonal, OM, or other). Since the training is based on synthetic geometries, a large enough number of training instances may be generated for each of these different classifiers. Another possibility is to divide the geometry into separate segments (e.g. for coronary geometries: proximal LAD, mid LAD, and distal LAD) and to extract the features discussed in the previous sections separately for each segment. Afterwards these features may either be combined into cumulative features or used separately for a single or multiple machine learning algorithms for predicting a hemodynamic metric of interest.--, in [0152], and, -- In act 60, uncertainty is assigned to one or more features. The uncertainty is a distribution of possible values for the feature. For example, the radius may be measured as 0.25 cm, but the accuracy or tolerance in the measurement provides that the radius is between 0.20 cm and 0.30 cm with greater probability for the values closer to 0.25 cm. Any distribution of possible or probable values may be used, such as a normal distribution, a distribution from a study, or from another source…. Using the machine-learnt algorithm, the confidence of the estimated hemodynamic metric is provided. A confidence or probability is provided for one value of the metric. Alternatively, the predictions from the learnt model may also be ranges or confidence intervals within which the predicted quantity is expected. The predicted confidence interval for the patient could be either directly predicted from the model or estimated from a set of similar anatomies from a saved database of synthetic models. --, in [0178]-[0180]).
Re Claim 19, MIHALEF as modified by ITU and BYDLON further disclose wherein the imaging information comprises one or more of: magnetic resonance imaging (MRI) information,
magnetic resonance angiography (MRA) information, a computed tomography (CT) scan information, a two-dimensional (2D) digital subtraction angiography information, or
a three-dimensional (3D) reconstruction from a sequence of 2D images (see MIHALEF: e.g., -- [0088] The scanner 1 10 is a medical diagnostic imaging CT system. A gantry supports a source of x-rays and a detector on opposite sides of a patient examination space. The gantry moves the source and detector about the patient to perform a coronary CT angiography scan. Various x-ray projections are acquired by the detector from different positions relative to the patient. Computed tomography solves for the two or three-dimensional distribution of the response from the projections. Ultrasound, x-ray, angiography, fluoroscopy, positron emission tomography, single photon emission computed tomography, and/or magnetic resonance scanners may additionally be used.--, in [0088]).
Re Claim 20, MIHALEF as modified by ITU and BYDLON further disclose wherein the outcome prediction indicates a predicted likelihood of a complete occlusion of the aneurysm sac imaged in the imaging information (see ITU: e.g., -- FIG. 8 shows an example of the normal distribution of the root radius of a coronary left arterial tree. Known, estimated, or standard normal distributions may be used. The synthetic examples generated are assigned the value for the root radius of the coronary left arterial tree based on the distribution (e.g., probability of a given value per example assigned using the distribution).--, in [0084], and,
-- [0102] Other features extracted include parameters for one or more abnormalities of the vessel structure. Abnormal morphology may be characterized by characteristics of calcification, characteristics of the plaque (e.g., fibrous tissue, lipid tissue, necrotic tissue, calcified tissue), characteristics of thrombus, characteristics of diffuse disease, presence of total or sub-total occlusion, presence of myocardial bridging (superficial and/or deep), congenital anomalies of coronary arteries (e.g., anomalous origin of a coronary artery from an abnormal sinus of Valsalva with an inter-arterial course between the great arteries, anomalous origin of one coronary artery from the pulmonary trunk, or others), aneurysmal dilatation and superimposed atherosclerosis, “high take off” coronary artery (e.g., the ostium is several millimeters above the sino-tubular junction (the artery may have a sharp downward angle and runs partially through the aortic wall)), myocardial bridging: superficial and deep, coronary fistula, coronary artery dissection, coronary vasculitis (e.g., rheumatoid arthritis, systemic lupus erythematosus (SLE), or Behçet's disease, Kawasaki disease, polyarteritis nodosa, and/or persisting (post) inflammatory aneurysms), fibromuscular dysplasia, coronary micro embolization, and/or left or right dominance. Additional, different, or fewer abnormality features may be used.--, in [0102], and, -- A local ischemia weight value is estimated independently for each root/branch/leaf segment using geometric features of the segment, such as the reference radius, length, tapering rate and other features…. An average value of healthy radiuses of the entire branch or a part of the branch, an average value of healthy radiuses obtained when excluding the largest x % and the smallest y % of the radius values of the entire branch or a part of the branch, or maximum or minimum value of healthy radii of the entire branch or part of the branch are computed.--, in [0110]-[0112]; and,
-- [0146] The in vitro model 23 and the flow conditions may be modified in numerous ways to generate a large number of setups. For example, the number, position and shape of the occluders is altered. As another example, the resistance at one or more locations is altered. In yet another example, the operation of the pump is altered. The number of side branches and any occlusions may be altered. Other alterations of combinations of different alterations are used to create different models with corresponding features and resulting flow characteristics.--, in [0145]-[0146], and, -- [0152] More than one classifier may be created. Since different types of branches and regions are present in a vessel tree, different classifiers may be machine trained for the different branches and/or regions. For example, different classifiers are trained for main and side branches, bifurcation regions and single branch regions, different types of pathologic regions such as different types of single branch stenotic regions (e.g., focal, long, diffuse, restenosis, or other), different types of bifurcation stenoses (e.g. a separate model for each bifurcation stenosis type in the medina classification), different types of aneurysms, different types of plaque, different types of total and/or sub-total occlusions, stenotic and regurgitant valves, various pathologies of the heart (e.g., past infarct or myopathies), or different types of branches (e.g. in case of coronary arterial trees: LM, LAD, LCx, RCA, Diagonal, OM, or other). Since the training is based on synthetic geometries, a large enough number of training instances may be generated for each of these different classifiers. Another possibility is to divide the geometry into separate segments (e.g. for coronary geometries: proximal LAD, mid LAD, and distal LAD) and to extract the features discussed in the previous sections separately for each segment. Afterwards these features may either be combined into cumulative features or used separately for a single or multiple machine learning algorithms for predicting a hemodynamic metric of interest.--, in [0152], and, -- In act 60, uncertainty is assigned to one or more features. The uncertainty is a distribution of possible values for the feature. For example, the radius may be measured as 0.25 cm, but the accuracy or tolerance in the measurement provides that the radius is between 0.20 cm and 0.30 cm with greater probability for the values closer to 0.25 cm. Any distribution of possible or probable values may be used, such as a normal distribution, a distribution from a study, or from another source…. Using the machine-learnt algorithm, the confidence of the estimated hemodynamic metric is provided. A confidence or probability is provided for one value of the metric. Alternatively, the predictions from the learnt model may also be ranges or confidence intervals within which the predicted quantity is expected. The predicted confidence interval for the patient could be either directly predicted from the model or estimated from a set of similar anatomies from a saved database of synthetic models. --, in [0178]-[0180]).
Re Claim 30, MIHALEF as modified by ITU and BYDLON further disclose wherein the neural network is pre-trained for the classification of at least one thousand categories of objects (see BYDLON: e.g., Fig. 8, --[0019] a training system to retrain in dynamic learning the model based on the shape measurements encountered during deployment… [0020] The model may be one of clustering, trained and built on training data including historic and/or synthetically generated shape measurements. The categories predicted or output are hence presented by clusters in the training data. Training may be done independent from imagery such as X-ray imagery….. If a clustering approach is used, the current measurement may be displayed concurrently with a graphical rendition of some or all clusters. The models envisaged herein for the logic are however not confined to cluster type models. Other (machine learning) models, such as artificial neural networks, encoder-autoencoder, and other still, are also envisaged herein in embodiments. In other words, the predicted or output category is not necessarily indicative of a cluster in the training data, but may relate instead to classifier results, etc. Graphical indications of the said classes may also be displayed.--, in [0019]-[0021], [0188]-[0190], [0197]-[0198], and, Fig. 9, and Fig. 11, -- [0213] The trained machine learning module M may be stored in one or more memories MEM or databases and can be made available as pre-trained machine learning models for use in system.--, in [0210]-[0213], and [0240]-[0241]).
Re Claim 31, MIHALEF as modified by ITU and BYDLON further disclose segment the aneurysm sac within digital imaging information using the encoder that includes the image classification neural network that is pre-trained for the classification of objects (see BYDLON: e.g., -- [0230] It is also possible that it is known before the procedure that the new patient is atypical or may have complex anatomy, thus increasing the likelihood of observing shapes that may require new clusters to be built on the fly. This information may be known from preoperative 3D imaging. In such cases, instead of forming new clusters on the fly, a different approach may be taken where shape data that is used to generate the pre-built clusters is simulated. The simulation can be performed by segmenting the vasculature that will be traversed and extracting a 3D mesh. Known mechanical properties of the devices ID to be used (e.g., thickness, flexibility, tip shape, etc.) may be fed into the training data generator to simulate the path from the access point to the target location within the 3D mesh. Simulated shapes can be saved and clustered as described above. The step of FIG. 11 may be used for this purpose. Since the data is simulated, several variations can be generated (e.g., by tweaking the parameters that control the device simulation) leading to a larger amount of data. During the procedure, shapes generated by devices in the procedure are assigned to clusters built from the simulated data, thus decreasing the likelihood for opening new clusters or categories on the fly as this may likely consume CPU resources.--, in [0230]; and, -- [0237] For example, 3D imaging data may be reconstructed in a way to represent a particular aspect of the anatomy such as the airways or vessels. Identifying one or more target locations, such as a tumor or portion of the vessel needing repair, a 3D path can be drawn from a starting point to the target. These 3D shapes, or paths, can be extracted from the imaging data. The whole path/shape or a segment of the path/shape may then define a single shape cluster respectively. 2D projection data, such as in X-ray radiography, may be used instead of 3D image data.
[0238] In more detail, a segmentation algorithm SEG may be used to define the vessel tree VT with which the user can interact through the user interface UI. For example, the user may use a free-hand drawing tool to draw shapes into the vessel tree representing an anatomical location of interest such as a certain part of the vessel. Alternatively, a discrete number of control points CP1-5 are set by the user through the interface in the vessel tree, and a spline algorithm is used to compute a curve passing through those control points and conforming in shape with the relevant part of the segmented vessel tree. Based on the shape, curvature or strain, stress values may be computed as desired, or alternatively the 3D (X,Y,Z) or 2D (X,Y) coordinates are used to define the shape s. In order to compute the stress and or strain, material properties of the arm A to be used may be required. The shapes may be automatically generated by the segmenter SEG using centerlines of the vessel tree.--, in [0237]-[0238], and, -- [0240] If a clustering algorithm is used for training, the so generated artificial shapes S+ with their respective category attached can be grouped according to their categories to define clusters which can then be immediately used during deployment to assign new shape measurements to the existing clusters. A linguistic analyzer component may be used that analyzes the user generated tags for grouping into the clusters. Alternatively, the training algorithm is re-run to adjust the parameters of the neural network or other classifier model thereby accounting for the newly added synesthetic training data specimens S.sup.+. The previous (old) training data already stored may be used in addition to newly generated ones to retrain the model. The generation of the synthetic shape measurements S.sup.+ may not necessarily rely on user support through the user interface UI as described above in FIG. 11 but may be done fully automatically. In this embodiment, the imagery is segmented as before, and portions of the vessel tree are labelled for the anatomical locations. Spline curves are then run across the anatomies to obtain the artificial shape measurements S.sup.+.
[0241] Another embodiment of generating training data may be based on biophysical modeling algorithms to define the most likely path the interventional device ID (such as a guidewire or catheter) would take, based on vessel shape and device mechanical properties. The path, from which the synthetically generated shape measurement may be generated as above, may be defined based on this prediction driven by said biophysical modeling algorithm. In addition, or instead, in embodiments past device ID data, such as provided by the shape sensing system or other tracking technology), is used to learn the most likely path taken by devices during successful navigation to the target. Variational autoencoders may be used for this ML task. Optionally, uncertainty values for each path are provided, such as standard deviation or other as obtainable by the encoder stage of the variational autoencoder.--, in [0240]-[0241]; also see ITU: e.g., -- FIG. 8 shows an example of the normal distribution of the root radius of a coronary left arterial tree. Known, estimated, or standard normal distributions may be used. The synthetic examples generated are assigned the value for the root radius of the coronary left arterial tree based on the distribution (e.g., probability of a given value per example assigned using the distribution).--, in [0084], and,
-- [0102] Other features extracted include parameters for one or more abnormalities of the vessel structure. Abnormal morphology may be characterized by characteristics of calcification, characteristics of the plaque (e.g., fibrous tissue, lipid tissue, necrotic tissue, calcified tissue), characteristics of thrombus, characteristics of diffuse disease, presence of total or sub-total occlusion, presence of myocardial bridging (superficial and/or deep), congenital anomalies of coronary arteries (e.g., anomalous origin of a coronary artery from an abnormal sinus of Valsalva with an inter-arterial course between the great arteries, anomalous origin of one coronary artery from the pulmonary trunk, or others), aneurysmal dilatation and superimposed atherosclerosis, “high take off” coronary artery (e.g., the ostium is several millimeters above the sino-tubular junction (the artery may have a sharp downward angle and runs partially through the aortic wall)), myocardial bridging: superficial and deep, coronary fistula, coronary artery dissection, coronary vasculitis (e.g., rheumatoid arthritis, systemic lupus erythematosus (SLE), or Behçet's disease, Kawasaki disease, polyarteritis nodosa, and/or persisting (post) inflammatory aneurysms), fibromuscular dysplasia, coronary micro embolization, and/or left or right dominance. Additional, different, or fewer abnormality features may be used.--, in [0102], and, -- A local ischemia weight value is estimated independently for each root/branch/leaf segment using geometric features of the segment, such as the reference radius, length, tapering rate and other features…. An average value of healthy radiuses of the entire branch or a part of the branch, an average value of healthy radiuses obtained when excluding the largest x % and the smallest y % of the radius values of the entire branch or a part of the branch, or maximum or minimum value of healthy radii of the entire branch or part of the branch are computed.--, in [0110]-[0112]; and,
-- [0146] The in vitro model 23 and the flow conditions may be modified in numerous ways to generate a large number of setups. For example, the number, position and shape of the occluders is altered. As another example, the resistance at one or more locations is altered. In yet another example, the operation of the pump is altered. The number of side branches and any occlusions may be altered. Other alterations of combinations of different alterations are used to create different models with corresponding features and resulting flow characteristics.--, in [0145]-[0146], and, -- [0152] More than one classifier may be created. Since different types of branches and regions are present in a vessel tree, different classifiers may be machine trained for the different branches and/or regions. For example, different classifiers are trained for main and side branches, bifurcation regions and single branch regions, different types of pathologic regions such as different types of single branch stenotic regions (e.g., focal, long, diffuse, restenosis, or other), different types of bifurcation stenoses (e.g. a separate model for each bifurcation stenosis type in the medina classification), different types of aneurysms, different types of plaque, different types of total and/or sub-total occlusions, stenotic and regurgitant valves, various pathologies of the heart (e.g., past infarct or myopathies), or different types of branches (e.g. in case of coronary arterial trees: LM, LAD, LCx, RCA, Diagonal, OM, or other). Since the training is based on synthetic geometries, a large enough number of training instances may be generated for each of these different classifiers. Another possibility is to divide the geometry into separate segments (e.g. for coronary geometries: proximal LAD, mid LAD, and distal LAD) and to extract the features discussed in the previous sections separately for each segment. Afterwards these features may either be combined into cumulative features or used separately for a single or multiple machine learning algorithms for predicting a hemodynamic metric of interest.--, in [0152], and, -- In act 60, uncertainty is assigned to one or more features. The uncertainty is a distribution of possible values for the feature. For example, the radius may be measured as 0.25 cm, but the accuracy or tolerance in the measurement provides that the radius is between 0.20 cm and 0.30 cm with greater probability for the values closer to 0.25 cm. Any distribution of possible or probable values may be used, such as a normal distribution, a distribution from a study, or from another source…. Using the machine-learnt algorithm, the confidence of the estimated hemodynamic metric is provided. A confidence or probability is provided for one value of the metric. Alternatively, the predictions from the learnt model may also be ranges or confidence intervals within which the predicted quantity is expected. The predicted confidence interval for the patient could be either directly predicted from the model or estimated from a set of similar anatomies from a saved database of synthetic models. --, in [0178]-[0180]),
wherein the decoder is trained to segment aneurysm sacs in images and output measurement information for the aneurysm sac after digital imaging information is processed at the encoder (see BYDLON: e.g., -- 0225] Further, the latent representation of a shape can be used by the decoder g.sub.d to reconstruct the new shape using only the low dimensional representation of the input shape. If the shape is not well represented by the training data, the difference between the original and reconstructed shape will be large. This difference, as above, can also be used to indicate whether a new cluster is required for the new shape. A simple thresholding of this.
[0226] difference may be used by the logic PL to decide whether or not to open a new category such as a cluster.--, in [0225]-[0226]; also see: -- A prediction algorithm is then used in deployment that identifies the relevant cluster for a shape measurement s, based on a distance measure. In more detail, a latent representation (vector) o.sub.i given a shape data as input using a pretrained encoder network g.sub.e is extracted. The decoder network can be excluded at this stage. This is done for all, or for a predefined number of, training data specimens s′. The latent space so obtained may then be clustered.--, in [0193]-[0194]).
Re Claim 32, MIHALEF as modified by ITU and BYDLON further disclose wherein the digital imaging information comprises one or more two-dimensional (2D) digital subtraction angiography (DSA) images of a wide-neck bifurcation aneurysm prior to implantation of an intrasaccular device (see MIHALEF: e.g., -- [0088] The scanner 1 10 is a medical diagnostic imaging CT system. A gantry supports a source of x-rays and a detector on opposite sides of a patient examination space. The gantry moves the source and detector about the patient to perform a coronary CT angiography scan. Various x-ray projections are acquired by the detector from different positions relative to the patient. Computed tomography solves for the two or three-dimensional distribution of the response from the projections. Ultrasound, x-ray, angiography, fluoroscopy, positron emission tomography, single photon emission computed tomography, and/or magnetic resonance scanners may additionally be used.--, in [0088]).
Re Claim 33, MIHALEF as modified by ITU and BYDLON further disclose wherein the digital imaging information includes one or more of a lateral and anterior-posterior (AP) views on a two-dimensional (2D) digital subtraction angiography (DSA) image or a three-dimensional (3D) axial slice stack reconstructed from a sequence of DSA images (see MIHALEF: e.g., --the user may indicate the starting and ending points on an image of the vessel or vessel model (e.g., 3D mesh rendered to 2D display image). Any locations of packing to increase porosity may be indicated. Other user inputs of the device and/or placement may be used, such as configuration for multiple devices…[0044]… the decision support processor selects. A selection function may be used, such as relating one or more different patient- specific characteristics to characteristics of or specific ones of the available devices. In one embodiment, a machine-learned network selects the vascular implant and/or placement of the vascular implant….the machine learning model inputs may be real (from patients, or bench-scale measurements), virtual (synthetic or not representative of a given patient or physical bench arrangement), or a combination of the two. For patient data, the input samples may contain the patient images, blood biomarkers, demographics, measurements and/or genetic data. For the ground truth, the input data includes the therapy that was chosen for the patient. Other ground truths may alternatively or additionally be used, such as the recorded outcome at different temporal points (e.g., 30 days, 90 days, etc.) and/or the occurrence or lack thereof of adverse events.--, in [0043]-[0046], and, -- [0088] The scanner 1 10 is a medical diagnostic imaging CT system. A gantry supports a source of x-rays and a detector on opposite sides of a patient examination space. The gantry moves the source and detector about the patient to perform a coronary CT angiography scan. Various x-ray projections are acquired by the detector from different positions relative to the patient. Computed tomography solves for the two or three-dimensional distribution of the response from the projections. Ultrasound, x-ray, angiography, fluoroscopy, positron emission tomography, single photon emission computed tomography, and/or magnetic resonance scanners may additionally be used.--, in [0088]).
Re Claim 34, MIHALEF as modified by ITU and BYDLON further disclose wherein the digital imaging information comprises a raw angiography image, and the at least one processor is further configured to:
identify a presence of the aneurysm sac in the raw angiography image using the image classification neural network prior to automatic segmentation of the aneurysm sac (see MIHALEF: e.g., --the user may indicate the starting and ending points on an image of the vessel or vessel model (e.g., 3D mesh rendered to 2D display image). Any locations of packing to increase porosity may be indicated. Other user inputs of the device and/or placement may be used, such as configuration for multiple devices…[0044]… the decision support processor selects. A selection function may be used, such as relating one or more different patient- specific characteristics to characteristics of or specific ones of the available devices. In one embodiment, a machine-learned network selects the vascular implant and/or placement of the vascular implant….the machine learning model inputs may be real (from patients, or bench-scale measurements), virtual (synthetic or not representative of a given patient or physical bench arrangement), or a combination of the two. For patient data, the input samples may contain the patient images, blood biomarkers, demographics, measurements and/or genetic data. For the ground truth, the input data includes the therapy that was chosen for the patient. Other ground truths may alternatively or additionally be used, such as the recorded outcome at different temporal points (e.g., 30 days, 90 days, etc.) and/or the occurrence or lack thereof of adverse events.--, in [0043]-[0046], and, -- [0088] The scanner 1 10 is a medical diagnostic imaging CT system. A gantry supports a source of x-rays and a detector on opposite sides of a patient examination space. The gantry moves the source and detector about the patient to perform a coronary CT angiography scan. Various x-ray projections are acquired by the detector from different positions relative to the patient. Computed tomography solves for the two or three-dimensional distribution of the response from the projections. Ultrasound, x-ray, angiography, fluoroscopy, positron emission tomography, single photon emission computed tomography, and/or magnetic resonance scanners may additionally be used.--, in [0088]; also see BYDLON: e.g., -- [0230] It is also possible that it is known before the procedure that the new patient is atypical or may have complex anatomy, thus increasing the likelihood of observing shapes that may require new clusters to be built on the fly. This information may be known from preoperative 3D imaging. In such cases, instead of forming new clusters on the fly, a different approach may be taken where shape data that is used to generate the pre-built clusters is simulated. The simulation can be performed by segmenting the vasculature that will be traversed and extracting a 3D mesh. Known mechanical properties of the devices ID to be used (e.g., thickness, flexibility, tip shape, etc.) may be fed into the training data generator to simulate the path from the access point to the target location within the 3D mesh. Simulated shapes can be saved and clustered as described above. The step of FIG. 11 may be used for this purpose. Since the data is simulated, several variations can be generated (e.g., by tweaking the parameters that control the device simulation) leading to a larger amount of data. During the procedure, shapes generated by devices in the procedure are assigned to clusters built from the simulated data, thus decreasing the likelihood for opening new clusters or categories on the fly as this may likely consume CPU resources.--, in [0230]; and, -- [0237] For example, 3D imaging data may be reconstructed in a way to represent a particular aspect of the anatomy such as the airways or vessels. Identifying one or more target locations, such as a tumor or portion of the vessel needing repair, a 3D path can be drawn from a starting point to the target. These 3D shapes, or paths, can be extracted from the imaging data. The whole path/shape or a segment of the path/shape may then define a single shape cluster respectively. 2D projection data, such as in X-ray radiography, may be used instead of 3D image data.
[0238] In more detail, a segmentation algorithm SEG may be used to define the vessel tree VT with which the user can interact through the user interface UI. For example, the user may use a free-hand drawing tool to draw shapes into the vessel tree representing an anatomical location of interest such as a certain part of the vessel. Alternatively, a discrete number of control points CP1-5 are set by the user through the interface in the vessel tree, and a spline algorithm is used to compute a curve passing through those control points and conforming in shape with the relevant part of the segmented vessel tree. Based on the shape, curvature or strain, stress values may be computed as desired, or alternatively the 3D (X,Y,Z) or 2D (X,Y) coordinates are used to define the shape s. In order to compute the stress and or strain, material properties of the arm A to be used may be required. The shapes may be automatically generated by the segmenter SEG using centerlines of the vessel tree.--, in [0237]-[0238], and, -- [0240] If a clustering algorithm is used for training, the so generated artificial shapes S+ with their respective category attached can be grouped according to their categories to define clusters which can then be immediately used during deployment to assign new shape measurements to the existing clusters. A linguistic analyzer component may be used that analyzes the user generated tags for grouping into the clusters. Alternatively, the training algorithm is re-run to adjust the parameters of the neural network or other classifier model thereby accounting for the newly added synesthetic training data specimens S.sup.+. The previous (old) training data already stored may be used in addition to newly generated ones to retrain the model. The generation of the synthetic shape measurements S.sup.+ may not necessarily rely on user support through the user interface UI as described above in FIG. 11 but may be done fully automatically. In this embodiment, the imagery is segmented as before, and portions of the vessel tree are labelled for the anatomical locations. Spline curves are then run across the anatomies to obtain the artificial shape measurements S.sup.+.
[0241] Another embodiment of generating training data may be based on biophysical modeling algorithms to define the most likely path the interventional device ID (such as a guidewire or catheter) would take, based on vessel shape and device mechanical properties. The path, from which the synthetically generated shape measurement may be generated as above, may be defined based on this prediction driven by said biophysical modeling algorithm. In addition, or instead, in embodiments past device ID data, such as provided by the shape sensing system or other tracking technology), is used to learn the most likely path taken by devices during successful navigation to the target. Variational autoencoders may be used for this ML task. Optionally, uncertainty values for each path are provided, such as standard deviation or other as obtainable by the encoder stage of the variational autoencoder.--, in [0240]-[0241]; also see ITU: e.g., -- FIG. 8 shows an example of the normal distribution of the root radius of a coronary left arterial tree. Known, estimated, or standard normal distributions may be used. The synthetic examples generated are assigned the value for the root radius of the coronary left arterial tree based on the distribution (e.g., probability of a given value per example assigned using the distribution).--, in [0084], and,
-- [0102] Other features extracted include parameters for one or more abnormalities of the vessel structure. Abnormal morphology may be characterized by characteristics of calcification, characteristics of the plaque (e.g., fibrous tissue, lipid tissue, necrotic tissue, calcified tissue), characteristics of thrombus, characteristics of diffuse disease, presence of total or sub-total occlusion, presence of myocardial bridging (superficial and/or deep), congenital anomalies of coronary arteries (e.g., anomalous origin of a coronary artery from an abnormal sinus of Valsalva with an inter-arterial course between the great arteries, anomalous origin of one coronary artery from the pulmonary trunk, or others), aneurysmal dilatation and superimposed atherosclerosis, “high take off” coronary artery (e.g., the ostium is several millimeters above the sino-tubular junction (the artery may have a sharp downward angle and runs partially through the aortic wall)), myocardial bridging: superficial and deep, coronary fistula, coronary artery dissection, coronary vasculitis (e.g., rheumatoid arthritis, systemic lupus erythematosus (SLE), or Behçet's disease, Kawasaki disease, polyarteritis nodosa, and/or persisting (post) inflammatory aneurysms), fibromuscular dysplasia, coronary micro embolization, and/or left or right dominance. Additional, different, or fewer abnormality features may be used.--, in [0102], and, -- A local ischemia weight value is estimated independently for each root/branch/leaf segment using geometric features of the segment, such as the reference radius, length, tapering rate and other features…. An average value of healthy radiuses of the entire branch or a part of the branch, an average value of healthy radiuses obtained when excluding the largest x % and the smallest y % of the radius values of the entire branch or a part of the branch, or maximum or minimum value of healthy radii of the entire branch or part of the branch are computed.--, in [0110]-[0112]; and,
-- [0146] The in vitro model 23 and the flow conditions may be modified in numerous ways to generate a large number of setups. For example, the number, position and shape of the occluders is altered. As another example, the resistance at one or more locations is altered. In yet another example, the operation of the pump is altered. The number of side branches and any occlusions may be altered. Other alterations of combinations of different alterations are used to create different models with corresponding features and resulting flow characteristics.--, in [0145]-[0146], and, -- [0152] More than one classifier may be created. Since different types of branches and regions are present in a vessel tree, different classifiers may be machine trained for the different branches and/or regions. For example, different classifiers are trained for main and side branches, bifurcation regions and single branch regions, different types of pathologic regions such as different types of single branch stenotic regions (e.g., focal, long, diffuse, restenosis, or other), different types of bifurcation stenoses (e.g. a separate model for each bifurcation stenosis type in the medina classification), different types of aneurysms, different types of plaque, different types of total and/or sub-total occlusions, stenotic and regurgitant valves, various pathologies of the heart (e.g., past infarct or myopathies), or different types of branches (e.g. in case of coronary arterial trees: LM, LAD, LCx, RCA, Diagonal, OM, or other). Since the training is based on synthetic geometries, a large enough number of training instances may be generated for each of these different classifiers. Another possibility is to divide the geometry into separate segments (e.g. for coronary geometries: proximal LAD, mid LAD, and distal LAD) and to extract the features discussed in the previous sections separately for each segment. Afterwards these features may either be combined into cumulative features or used separately for a single or multiple machine learning algorithms for predicting a hemodynamic metric of interest.--, in [0152], and, -- In act 60, uncertainty is assigned to one or more features. The uncertainty is a distribution of possible values for the feature. For example, the radius may be measured as 0.25 cm, but the accuracy or tolerance in the measurement provides that the radius is between 0.20 cm and 0.30 cm with greater probability for the values closer to 0.25 cm. Any distribution of possible or probable values may be used, such as a normal distribution, a distribution from a study, or from another source…. Using the machine-learnt algorithm, the confidence of the estimated hemodynamic metric is provided. A confidence or probability is provided for one value of the metric. Alternatively, the predictions from the learnt model may also be ranges or confidence intervals within which the predicted quantity is expected. The predicted confidence interval for the patient could be either directly predicted from the model or estimated from a set of similar anatomies from a saved database of synthetic models. --, in [0178]-[0180]).
Conclusion
Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any extension fee pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to WEI WEN YANG whose telephone number is (571)270-5670. The examiner can normally be reached on 8:00 - 5:00 pm.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Amandeep Saini can be reached on 571-272-3382. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of an application may be obtained from the Patent Application Information Retrieval (PAIR) system. Status information for published applications may be obtained from either Private PAIR or Public PAIR. Status information for unpublished applications is available through Private PAIR only. For more information about the PAIR system, see http://pair-direct.uspto.gov. Should you have questions on access to the Private PAIR system, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative or access to the automated information system, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/WEI WEN YANG/Primary Examiner, Art Unit 2662
Prosecution Insights