DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Applicant' s arguments, filed 6/4/2026, have been fully considered. The following rejections and/or objections are either reiterated or newly applied. They constitute the complete set presently being applied to the instant application.
Applicants have amended their claims, filed 6/4/2026, and therefore rejections newly made in the instant office action have been necessitated by amendment.
Claims 1-16 and 21-28 are the currently pending claims hereby under examination. Claims 17-20 have been previously canceled and claims 27-28 have been newly added.
Claim Rejections - 35 USC § 112
The following is a quotation of the first paragraph of 35 U.S.C. 112(a):
(a) IN GENERAL.—The specification shall contain a written description of the invention, and of the manner and process of making and using it, in such full, clear, concise, and exact terms as to enable any person skilled in the art to which it pertains, or with which it is most nearly connected, to make and use the same, and shall set forth the best mode contemplated by the inventor or joint inventor of carrying out the invention.
The following is a quotation of the first paragraph of pre-AIA 35 U.S.C. 112:
The specification shall contain a written description of the invention, and of the manner and process of making and using it, in such full, clear, concise, and exact terms as to enable any person skilled in the art to which it pertains, or with which it is most nearly connected, to make and use the same, and shall set forth the best mode contemplated by the inventor of carrying out his invention.
Claims 1-16 and 21-28 are rejected under 35 U.S.C. 112(a) or 35 U.S.C. 112 (pre-AIA ), first paragraph, as failing to comply with the written description requirement. The claim(s) contains subject matter which was not described in the specification in such a way as to reasonably convey to one skilled in the relevant art that the inventor or a joint inventor, or for applications subject to pre-AIA 35 U.S.C. 112, the inventor(s), at the time the application was filed, had possession of the claimed invention.
Independent claims 1, 8, 14, and 16 were amended to recite one or more sensors, or a sensor, configured to be worn or worn “at or near a sacrum or lumbar region” (lines 6, 8, 5-6, and 7-8 respectively) of the specific patient. The originally filed disclosure reasonably supports placement at or near the sacrum (Instant Application, [0012], [0038]-[0039]). The originally filed disclosure, however, does not describe positioning a wearable sensor at or near a lumbar region. Paragraph [0037] generally states that multiple sensors or accelerometers may be used at varying locations on a patient’s body. This general disclosure does not identify the lumbar region as a contemplated sensor location or otherwise reasonably convey to a person of ordinary skill in the art that the inventors possessed the specifically claimed lumbar placement. Applicant identifies paragraphs [0020], [0038], and [0039] as exemplary support for the amendments. Paragraph [0020] concerns PAD treatments and does not describe sensor placement. Paragraphs [0038] and [0039] describe the sacral position but do not describe a lumbar position.
Claims 2-7, 21-23 and 27-28 are rejected by virtue of their dependence from claim 1.
Claims 9-13 and 24-26 are rejected by virtue of their dependence from claim 8.
Claim 15 is rejected by virtue of its dependence from claim 14.
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 1-16, and 21-28 are rejected under 35 U.S.C. 101 because the claimed invention is directed to a judicial exception (i.e., a law of nature, a natural phenomenon, or an abstract idea) without significantly more. Claims 1-16, and 21-28 are directed to receiving acceleration data related to a patient, extracting or using gait characteristic data, using a trained machine learning model to identify gait features, and outputting a diagnosis of peripheral artery disease (PAD), which is an abstract idea. Claims 1-16, and 21-28 do not include additional elements that integrate the exception into a practical application or that are sufficient to amount to significantly more than the judicial exception for the reasons provided below which are in line with the 2014 Interim Guidance on Patent Subject Matter Eligibility (Federal Register, Vol. 79, No. 241, p 74618, December 16, 2014), the July 2015 Update on Subject Matter Eligibility (Federal Register, Vol. 80, No. 146, p. 45429, July 30, 2015), the May 2016 Subject Matter Eligibility Update (Federal Register, Vol. 81, No. 88, p. 27381, May 6, 2016), and the 2019 Revised Patent Subject Matter Eligibility Guidance (Federal Register, Vol. 84, No. 4, page 50, January 7, 2019).
The analysis of claim 1 is as follows:
Step 1: Claim 1 is directed to a process.
Step 2A, Prong One: Claim 1 recites the following limitation:
[A1] diagnosing the specific patient as having PAD or not having PAD based on the one or more identified gait features.
Limitation [A1] recites an abstract idea in the form of a mental process. In particular, evaluating previously identified gait information and reaching a conclusion as to whether a patient has PAD constitutes observation, evaluation, judgment, or opinion. See MPEP 2106.04(a)(2).
Claim 1 requires only “one or more” identified gait features and does not require any particular quantity or complexity of information to be considered in making the diagnostic determination. The claim therefore encompasses evaluating a single previously identified gait feature and reaching a diagnostic judgment based on that feature.
The Examiner does not determine that a human could practically process all of the underlying acceleration data or perform the separately recited machine learning operations in the human mind. Those limitations are evaluated as additional elements under Step 2A, Prong Two. Nevertheless, claim 1 recites a judicial exception because the diagnostic determination itself encompasses a mental evaluation based on previously identified information.
The recitation of a trained machine learning model is not separately characterized as a mathematical concept. Claim 1 does not recite any particular mathematical relationship, formula, equation, or calculation performed by the model.
Step 2A, Prong Two: Claim 1 recites the following additional elements beyond the judicial exception:
[A2] receiving acceleration data related to a specific patient from one or more sensors configured to be worn at or near a sacrum or lumbar region of the specific patient;
[B2] extracting gait characteristics data from the acceleration data related to the specific patient; and
[C2] feeding the extracted gait characteristics data to a trained machine learning model to identify one or more gait features for the specific patient.
The additional elements do not integrate the judicial exception into a practical application.
Limitation [A2] constitutes data gathering performed before the diagnostic evaluation. Although the claim specifies that the sensors are configured to be worn at or near the sacrum or lumbar region, this limitation identifies the anatomical source from which the acceleration data is gathered. Claim 1 does not recite that the placement changes or improves operation of the sensor, produces an improved form of acceleration data, improves signal quality, or causes the machine learning model to operate differently. The anatomical placement therefore does not apply the diagnostic evaluation through a particular machine in a manner that imposes a meaningful technological limitation. See MPEP 2106.05(b) and 2106.05(g).
Limitations [B2] and [C2] recite extracting information and applying a trained machine learning model according to their desired results. Claim 1 does not specify a particular signal processing procedure, feature extraction technique, machine learning architecture, training procedure, model parameter adjustment, or other technological mechanism by which those results are achieved. These limitations use computer processing and a trained machine learning model as tools to prepare and analyze the gathered information. See MPEP 2106.05(f).
Considered as an ordered combination, the additional elements collect acceleration data, extract gait information from the collected data, and apply a trained model to identify information used in reaching the diagnostic conclusion. The ordered combination does not improve the functioning of the sensor, the computer, or the machine learning model. Nor does the claim require administering a treatment or performing another technological operation based on the diagnostic conclusion.
Collecting information, analyzing that information, and using the resulting information to reach or report a conclusion does not integrate an abstract evaluation into a practical application merely because the information originates from physical sensors. See Electric Power Group, LLC v. Alstom S.A., 830 F.3d 1350, 1353-1356 (Fed. Cir. 2016).
Desjardins: Applicant’s reliance on Ex parte Desjardins, Appeal No. 2024 000567 (Appeals Review Panel Sept. 26, 2025), is not persuasive. Desjardins involved claimed limitations that reflected a particular improvement in machine learning operation. The present specification discusses possible benefits from concurrently collected biomechanics data, transfer learning, and particular signal processing procedures. See, for example, paragraphs [0022], [0035], and [0039]. Claim 1 does not require those procedures or the limitations described as producing their stated benefits. The specification also does not identify placement at the sacrum or lumbar region as producing an improvement in sensor operation, signal processing, or machine learning operation.
Accordingly, the additional elements, considered individually and as an ordered combination, do not integrate the recited mental process into a practical application. Claim 1 is therefore directed to a judicial exception.
Step 2B: Claim 1 does not recite additional elements that amount to significantly more than the judicial exception.
The sensors recited in limitation [A2] perform their ordinary function of collecting motion data. Paragraph [0038] of the specification explains that acceleration ultimately can be measured directly by a wearable accelerometer. Vining et al., US 2017/0205789 A1, paragraph [0018], describes a commercial off the shelf accelerometer, supporting that accelerometer hardware for detecting motion was well understood, routine, and conventional.
The record also supports that placement of inertial sensors at the claimed anatomical regions for gait analysis was commonly known and used before the effective filing date.
Lim et al., “Prediction of Lower Limb Kinetics and Kinematics during Walking by a Single IMU on the Lower Back Using Machine Learning,” Sensors, volume 20, article 130 (2020), p. 2, Introduction, explains that the market for wearable motion-monitoring systems had rapidly grown, identifies established commercial wearable products, and reviews prior work using a single IMU at the sacrum. Lim further describes an EBIMU-9DOFV4 inertial sensor worn near the sacrum (Lim, p. 1, Abstract; p. 5, Sec. 2.2). Lim therefore supports that wearable inertial sensing at the claimed sacral region was known in gait analysis.
Teufl et al., “Towards Inertial Sensor Based Mobile Gait Analysis: Event Detection and Spatio Temporal Parameters,” Sensors, volume 19, article 38, p. 1-2 (2019), states that numerous gait analysis systems based on one or two inertial sensors already existed. Teufl further identifies prior systems using a single inertial measurement unit attached to the sacrum and a single inertial measurement unit installed on the lower back.
These publications do more than identify an isolated prior use of a sensor. They describe multiple prior gait analysis systems employing inertial sensors at the sacrum, lumbar region, lower back, pelvis, or waist. Together with Vining, they support that both the wearable acceleration sensor and its use at the claimed anatomical region for gathering gait data were well understood, routine, and conventional.
Limitations [B2] and [C2] likewise are recited at a high level of generality. Paragraph [0026] of the specification states that the model may include any useful machine learning or artificial intelligence algorithm apparent to a person skilled in the art and identifies established algorithm classes, including neural networks, nearest neighbor algorithms, random forest algorithms, support vector machines, and Logit algorithms. Claim 1 does not require a particular architecture, training method, parameter adjustment, or feature extraction process that departs from the ordinary use of a trained model.
The generic application of established machine learning techniques to a particular field or type of data does not itself provide an inventive concept. See Recentive Analytics, Inc. v. Fox Corp., 134 F.4th 1205, 1212-1214 (Fed. Cir. 2025).
Considered as an ordered combination, the sensors collect acceleration data, computer processing extracts information from that data, and a generic trained model identifies gait information used in making the diagnostic judgment. Each component performs its ordinary function, and the claim does not recite a nonconventional arrangement or interaction among the components. Applying these components to PAD diagnosis does not itself supply an inventive concept. See Alice Corp. Pty. Ltd. v. CLS Bank International, 573 U.S. 208, 223-226 (2014); Electric Power Group, 830 F.3d at 1355-1356.
Accordingly, claim 1 does not recite significantly more than the judicial exception.
Dependent claims: Claims 2 through 7, 21 through 23, 27, and 28 depend directly or indirectly from claim 1 and therefore retain the mental process identified for claim 1. Their additional limitations do not integrate that mental process into a practical application and do not amount to significantly more.
Claim 2 specifies that the trained machine learning model is a neural network, random forest, support vector machine, or Logit algorithm. This limitation selects among established classes of machine learning models without reciting a particular architecture, training procedure, parameter adjustment, or improvement in model operation. Paragraph [0026] identifies these algorithm classes as examples of useful algorithms apparent to a person skilled in the art. Claim 2 therefore does not recite a technological improvement or an inventive concept.
Claims 3, 5, and 7 further specify how input data is obtained or the content of that input data. Claim 3 recites sensors worn by the patient, claim 5 identifies particular acceleration and temporal data, and claim 7 adds biometric data as an input to the model. These limitations constitute data gathering, data selection, or identification of the information being analyzed. They do not improve sensor operation, computer functionality, feature extraction, or machine learning operation. The conventional nature of wearable motion sensors is supported by Vining, Lim, and Teufl as discussed for claim 1.
Claims 4, 6, and 27 specify the gait features or biomechanics information identified or evaluated. These claims identify temporal gait measurements, asymmetry and variability measurements, stance to swing ratio, gait symmetry, ground reaction force information, and joint related measurements. The claims do not recite a particular technological procedure for obtaining or calculating those features. Merely limiting the diagnostic evaluation to particular categories of information does not integrate the mental process into a practical application or provide significantly more.
Claim 21 further recites determining PAD severity or a change in PAD severity based on gait characteristics data. Determining disease severity from previously obtained information constitutes further evaluation and judgment. This limitation therefore remains within the mental process category and does not integrate the exception into a practical application.
Claims 22 and 28 recite selecting or modifying a treatment course based on the PAD severity determination. Claim 28 limits the available choices to supervised exercise therapy, surgical intervention, an ankle foot orthosis, or exoskeleton footwear. Neither claim requires administering or otherwise carrying out the selected treatment. Merely selecting or recommending a treatment does not apply the exception through a particular treatment or prophylaxis. See MPEP 2106.04(d)(2).
Claim 23 recites providing an alert on a display device based on the diagnosis, severity, or change in severity. This limitation outputs the result after the diagnostic evaluation has been completed and therefore constitutes insignificant activity following the exception. See MPEP 2106.05(g); Electric Power Group, LLC v. Alstom S.A., 830 F.3d 1350, 1354 (Fed. Cir. 2016).
Considered individually and in combination with claim 1, the dependent claim limitations further specify the model, input information, evaluated information, subsequent clinical judgments, or manner in which results are reported. They do not recite an improvement in computer, sensor, signal processing, or machine learning technology; an unconventional arrangement of components; or an actually administered treatment. Accordingly, claims 2 through 7, 21 through 23, 27, and 28 do not integrate the judicial exception into a practical application and do not amount to significantly more than the judicial exception.
The analysis of claim 8 is as follows:
Step 1: Claim 8 is directed to a machine.
Step 2A, Prong One: Claim 8 recites the following limitation:
[A1] diagnosing the specific patient as having PAD or not having PAD based on the one or more identified gait features.
Limitation [A1] recites an abstract idea in the form of a mental process. Evaluating previously identified gait information and reaching a diagnostic conclusion constitutes observation, evaluation, judgment, or opinion. See MPEP 2106.04(a)(2).
Although claim 8 configures a processor to perform the diagnostic determination, using a generic processor to perform an evaluation that otherwise can be performed mentally does not remove the underlying evaluation from the mental process category. The identified mental process is the diagnostic determination based on previously identified gait features, not the machine learning operations or the processing of all underlying acceleration data.
Step 2A, Prong Two: Claim 8 recites the following additional elements beyond the judicial exception:
[A2] a processor and memory storing instructions that cause the processor to access or execute a trained machine learning model, where the model has been trained using gait characteristic data from patients known to have PAD and patients that do not have PAD;
[B2] one or more sensors configured to be worn at or near a sacrum or lumbar region of the specific patient; and
[C2] configuring the processor to receive acceleration data from the sensors, extract gait characteristics data, and feed the extracted data to the trained machine learning model to identify one or more gait features.
The additional elements do not integrate the mental process into a practical application.
The processor and memory perform generic computer functions of executing instructions, accessing a model, receiving data, and processing data. The limitation describing the information used to train the model identifies the model’s training data but does not recite a particular model architecture, training procedure, parameter adjustment, or improvement in model operation.
The sensors and receiving operation gather acceleration data from the patient. Specifying placement at or near the sacrum or lumbar region identifies the anatomical source of the gathered data but does not recite an improvement in sensor operation, signal quality, feature extraction, or machine learning operation.
The extraction and machine learning limitations are recited according to the results to be achieved. Claim 8 does not require a particular signal processing procedure, feature extraction technique, or technological mechanism by which the trained model identifies the gait features.
Desjardins: Applicant’s reliance on Ex parte Desjardins, Appeal No. 2024 000567 (Appeals Review Panel Sept. 26, 2025), has been considered but is not persuasive as to claim 8. Claim 8 recites a trained machine learning model that was trained using gait characteristic data from patients known to have PAD and patients that do not have PAD. The claim does not recite a particular training architecture, objective function, parameter adjustment, transfer learning procedure, or other mechanism that improves operation of the model. Although the specification discusses concurrent biomechanics training, transfer learning, and particular signal processing procedures in paragraphs [0022], [0035], and [0039], claim 8 does not reflect those procedures or the improvements attributed to them. The claimed sensor placement also is not described as improving sensor, computer, or model operation. Claim 8 therefore does not reflect the type of technological improvement addressed in Desjardins.
Considered as an ordered combination, claim 8 recites a processor and memory that receive sensor data, extract information, apply a trained model, and make a diagnostic determination. The components perform their ordinary functions and do not interact in a manner that improves the processor, memory, sensor, signal processing, or machine learning model. Accordingly, the additional elements do not integrate the mental process into a practical application.
Step 2B: The additional elements do not amount to significantly more than the judicial exception.
Paragraph [0025] describes implementing the method using a computer system having a processor and associated memory. Paragraph [0026] permits any useful machine learning or artificial intelligence algorithm apparent to a person skilled in the art. These disclosures treat the processor, memory, and machine learning model as ordinary tools for implementing the diagnostic evaluation.
Vining et al., US 2017/0205789 A1, paragraph [0018], describes a commercial off the shelf accelerometer and supports the conventional nature of accelerometer hardware.
Lim et al., “Prediction of Lower Limb Kinetics and Kinematics during Walking by a Single IMU on the Lower Back Using Machine Learning,” Sensors, volume 20, article 130 (2020), p. 2, Introduction, describes prior single-IMU gait-analysis work at the sacrum. Lim also describes an inertial sensor actually worn near the sacrum for collecting gait acceleration data (Lim, p. 1, Abstract; p. 5, Sec. 2.2).
Teufl et al., “Towards Inertial Sensor Based Mobile Gait Analysis: Event Detection and Spatio Temporal Parameters,” Sensors, volume 19, article 38, p. 1-2 (2019), describes numerous inertial sensor gait analysis systems and identifies prior systems using single sensors at the sacrum and lower back.
These publications support that wearable acceleration sensors and their placement at the claimed anatomical regions for gathering gait information were well understood, routine, and conventional.
Considered individually and as an ordered combination, the processor, memory, trained model, and sensors perform their ordinary functions of storing instructions, collecting data, processing information, and applying a model. Claim 8 does not recite a nonconventional arrangement or interaction among those components. Accordingly, claim 8 does not recite significantly more than the judicial exception.
Dependent claims: Claims 9 through 13 and 24 through 26 depend from claim 8 and retain the mental process identified for claim 8. Their additional limitations do not integrate that mental process into a practical application and do not amount to significantly more.
Claims 9 and 10 specify implementation details. Claim 9 stores code and data associated with the model in memory. Claim 10 selects among established machine learning model classes. Neither claim recites an improvement in memory operation, model architecture, model training, or computer functionality. Paragraphs [0025] and [0026] describe generic memory and established machine learning algorithm classes.
Claims 11 through 13 specify the information identified, evaluated, or supplied to the model. Claim 11 identifies temporal gait features, claim 12 identifies acceleration, temporal, ground reaction force, and joint related information, and claim 13 adds biometric training and patient data. These limitations further specify the content of the information being processed without reciting an improved technological procedure for obtaining or processing that information.
Claim 24 further recites determining PAD severity or a change in PAD severity based on gait characteristics data. Determining disease severity from previously obtained information constitutes further evaluation and judgment and therefore remains within the mental process category.
Claim 25 further recites selecting or modifying a treatment course based on the severity determination. The claim does not require administering or otherwise carrying out the selected treatment. It therefore does not apply the mental process through a particular treatment or prophylaxis. See MPEP 2106.04(d)(2).
Claim 26 recites providing an alert on a display device based on the diagnosis or severity determination. This limitation outputs the result after the evaluation and constitutes insignificant activity following the exception. See MPEP 2106.05(g); Electric Power Group, LLC v. Alstom S.A., 830 F.3d 1350, 1354 (Fed. Cir. 2016).
Considered individually and in combination with claim 8, these limitations further specify data storage, model type, information content, subsequent clinical judgments, or output of results. They do not recite an improvement in computer, sensor, signal processing, or machine learning technology; an unconventional arrangement of components; or an actually administered treatment. Accordingly, claims 9 through 13 and 24 through 26 do not integrate the judicial exception into a practical application and do not amount to significantly more than the judicial exception.
The analysis of claim 14 is as follows:
Step 1: Claim 14 is directed to a process.
Step 2A, Prong One: Claim 14 recites the following limitation:
[A1] diagnosing the specific patient as having PAD or not having PAD based on the one or more identified gait features.
For the reasons explained for claim 1, limitation [A1] recites a mental process involving evaluation and diagnostic judgment. Claim 14 requires only one or more identified gait features and does not require any particular quantity or complexity of information to be considered in making the diagnostic determination.
The Examiner does not characterize the model training or model application limitations as mental processes or mathematical concepts. Those limitations are evaluated as additional elements under Step 2A, Prong Two.
Step 2A, Prong Two: Claim 14 recites the following additional elements:
[A2] training a machine learning model using acceleration or accelerometer data for patients known to have PAD and patients that do not have PAD;
[B2] receiving acceleration data related to a specific patient from a sensor worn at or near a sacrum or lumbar region of the specific patient; and
[C2] feeding the acceleration data to the trained machine learning model to identify one or more gait features for the specific patient.
The additional elements do not integrate the mental process into a practical application.
Limitation [A2] recites model training at a high level of generality. It identifies labeled training data but does not specify a particular model architecture, objective function, parameter adjustment, training calculation, or other procedure that improves model operation.
Limitation [B2] constitutes data gathering. The anatomical placement specifies where the data is collected without reciting an improvement in sensor operation, signal quality, or the form of acceleration data produced.
Limitation [C2] applies the trained model according to the result to be achieved. It does not recite a particular model operation or technological procedure for identifying the gait features.
Desjardins: Applicant’s reliance on Ex parte Desjardins has been considered but is not persuasive as to claim 14. Although claim 14 expressly recites training a machine learning model, it merely identifies acceleration data from patients known to have PAD and patients that do not have PAD as labeled training data. The claim does not recite how the model is trained, how model parameters are adjusted, or how the training improves model operation. Claim 14 does not require the concurrent biomechanics training discussed in paragraph [0022], the transfer learning discussed in paragraph [0035], or the particular signal processing discussed in paragraph [0039]. The claim therefore does not reflect a particular technological improvement in machine learning operation of the type addressed in Desjardins.
Considered as an ordered combination, the additional elements generically train a model, gather patient data, apply the trained model, and use the resulting information in making a diagnostic judgment. They do not improve sensor, computer, signal processing, or machine learning operation and therefore do not integrate the mental process into a practical application.
Step 2B: The additional elements do not amount to significantly more than the judicial exception.
Paragraph [0026] states that any useful machine learning or artificial intelligence algorithm apparent to a person skilled in the art may be used. Claim 14 does not recite any particular training technique or model configuration beyond training an unspecified model using labeled acceleration data.
The conventional nature of the sensor hardware and anatomical placement is supported by Vining, Lim, and Teufl (as shown above). Vining describes commercial off the shelf accelerometer hardware. Lim explains that the wearable motion-monitoring market had rapidly grown, identifies established commercial wearable products, reviews prior single-IMU work at the sacrum, and uses an inertial sensor worn near the sacrum. Teufl describes numerous inertial-sensor gait-analysis systems and prior single-sensor placement at the sacrum and lower back.
Applying established machine learning techniques to a particular field or type of data does not itself provide an inventive concept. See Recentive Analytics, Inc. v. Fox Corp., 134 F.4th 1205, 1212-1214 (Fed. Cir. 2025).
Considered individually and as an ordered combination, the training, data gathering, and model application limitations perform their ordinary functions and do not recite a nonconventional interaction or technological improvement. Accordingly, claim 14 does not amount to significantly more than the judicial exception.
Dependent Claims: Claim 15 depends from claim 14 and specifies that the machine learning model is a recurrent neural network or long short term memory model. This limitation selects between established model classes without reciting a particular architecture, training procedure, parameter adjustment, or improvement in model operation. Paragraph [0009] identifies recurrent neural networks as an available model type, while paragraph [0026] describes the use of established machine learning algorithms generally. Claim 15 therefore does not integrate the exception into a practical application or provide significantly more.
The analysis of claim 16 is as follows:
Step 1: Claim 16 is directed to a manufacture, namely, a nontransitory computer readable medium storing instructions.
Step 2A, Prong One: Claim 16 recites the following limitation:
[A1] diagnosing the specific patient as having PAD or not having PAD based on the one or more identified gait features.
For the reasons explained for claims 1 and 14, limitation [A1] recites a mental process involving evaluation and diagnostic judgment. The identified mental process is the diagnostic determination based on previously identified gait features, not the model training, model application, or processing of all underlying acceleration data.
Step 2A, Prong Two: Claim 16 recites the following additional elements:
[A2] a nontransitory computer readable medium storing instructions that, when executed by one or more processors, cause the processors to perform the method;
[B2] training a machine learning model using acceleration or accelerometer data for patients known to have PAD and patients that do not have PAD;
[C2] receiving acceleration data related to a specific patient from a sensor worn at or near a sacrum or lumbar region of the specific patient; and
[D2] feeding the acceleration data to the trained machine learning model to identify one or more gait features for the specific patient.
The additional elements do not integrate the mental process into a practical application.
The computer readable medium and processors provide generic computer implementation. They do not recite an improvement in data storage, processor operation, memory operation, or computer functionality.
The model training limitation identifies labeled training data but does not specify a particular model architecture, training objective, parameter adjustment, or other procedure that improves model operation. The sensor limitation gathers acceleration data from a specified anatomical location without improving sensor operation or signal quality. The model application limitation is recited according to the result to be achieved and does not specify a particular technological procedure for identifying gait features.
Desjardins: Applicant’s reliance on Ex parte Desjardins has been considered but is not persuasive as to claim 16. Like claim 14, claim 16 recites training a machine learning model using labeled acceleration data without reciting a particular training procedure, parameter adjustment, model architecture, or improvement in model operation. Storing instructions for performing that generic training on a nontransitory computer readable medium does not cause the claim to reflect the concurrent biomechanics training, transfer learning, or particular signal processing procedures discussed in paragraphs [0022], [0035], and [0039]. Claim 16 therefore does not reflect the type of technological improvement addressed in Desjardins.
Considered as an ordered combination, claim 16 stores instructions for generically training and applying a machine learning model to sensor data and producing a diagnostic conclusion. Storing the instructions on a computer readable medium does not meaningfully limit the mental process or produce a technological improvement. The additional elements therefore do not integrate the mental process into a practical application.
Step 2B: The additional elements do not amount to significantly more than the judicial exception.
Paragraph [0025] describes generic processors and associated memory, and paragraph [0026] describes using any useful machine learning algorithm apparent to a person skilled in the art. The sensor hardware and anatomical placement are well understood, routine, and conventional for the reasons and evidence discussed for claims 1, 8, and 14.
The recitation of a computer readable medium storing instructions for generic computer implementation does not provide an inventive concept. See Alice Corp. Pty. Ltd. v. CLS Bank International, 573 U.S. 208, 223-226 (2014); SAP America, Inc. v. InvestPic, LLC, 898 F.3d 1161, 1169-1170 (Fed. Cir. 2018).
Considered individually and as an ordered combination, the computer readable medium, processors, sensor, model training, and model application limitations perform ordinary computing and data processing functions. Claim 16 does not recite a nonconventional arrangement or interaction among these elements. Accordingly, claim 16 does not recite significantly more than the judicial exception.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
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 1-6, 8-12, 14-16, 21, 23-24, 26-27 are rejected under 35 U.S.C. 103 as being unpatentable over Huang et al. (US 2018/0279915 A1), hereinafter referred to as Huang, and further in view of Lim et al. (Lim, Hyerim et al. “Prediction of Lower Limb Kinetics and Kinematics during Walking by a Single IMU on the Lower Back Using Machine Learning.” Sensors (Basel, Switzerland) vol. 20,1 130. 24 Dec. 2019), hereinafter referred to as Lim, and further in view of Arshad et al. (Arshad, Muhammad Zeeshan, et al. “Gait Events Prediction Using Hybrid CNN-RNN-Based Deep Learning Models through a Single Waist-Worn Wearable Sensor.” Sensors [Basel], vol. 22, no. 21, October 2022), hereinafter referred to as Arshad, and further in view of Chidean et al. (Chidean, Mihaela I. et al. “Full Band Spectra Analysis of Gait Acceleration Signals for Peripheral Arterial Disease Patients.” Frontiers in Physiology 9 (2018)), hereinafter referred to as Chidean, and further in view of Rahman et al. (Rahman, Hafizur et al. “Gait Variability Is Affected More by Peripheral Artery Disease than by Vascular Occlusion.” PLoS ONE 16.3 (2021)), hereinafter referred to as Rahman.
Regarding claim 1, Huang teaches the method comprising: receiving acceleration data related to a specific patient from one or more sensors (Huang, ¶[0006]: “The method includes receiving a signal comprising sensor data,” wherein the sensors include “a three-axis accelerometer”; ¶[0015]: the wearable device detects data related to a subject’s gait, balance, or posture, thereby receiving patient-specific acceleration data from wearable sensors); and extracting gait characteristics data from the acceleration data related to the specific patient (Huang, claim 18: “analyzing, by the system, the sensor data to identify a pattern related to gait, balance or posture within the sensor data”; ¶[0029]: the analyzer identifies patterns associated with gait, including walking speed, stride height, stride length, speed difference between the legs, cadence, and swing time, thereby extracting gait characteristics from the acceleration data).
Also regarding claim 1, Huang receives gait acceleration data from wearable sensors but does not teach that the sensors are configured to be worn at or near a sacrum or lumbar region of the specific patient.
Lim teaches collecting walking data from a single IMU worn near the sacrum and using the sacral motion as a measurable approximation of the center of mass. Lim explains that the center-of-mass location permits lower-limb dynamic data to be predicted while addressing the tradeoff between data quantity and wearable convenience (Lim, p. 1, Abstract; p. 2, Introduction). Lim further describes an EBIMU-9DOFV4 inertial sensor positioned at the lower back near the sacrum and collecting sagittal-plane acceleration data during walking (Lim, p. 5, Sec. 2.2).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Huang in view of Lim to configure Huang’s acceleration sensor to be worn near the patient’s sacrum. The modification would have been implemented by mounting Lim’s inertial sensor at the lower back near the sacrum and supplying its acceleration data to Huang’s analyzer and machine learning system. One of ordinary skill would have had a reasonable expectation of success because Huang and Lim both use wearable accelerometers to acquire gait data, and Lim actually collected gait acceleration data from a single IMU near the sacrum and used the data to predict lower-limb dynamics. The benefit would have been a minimally obstructive single-sensor arrangement located near the body’s center of mass, with reduced sensor burden and improved wearable convenience.
Also regarding claim 1, the modified Huang teaches extracting gait-related information from acceleration data and feeding extracted features to a trained machine learning classifier, but does not expressly teach feeding the extracted gait characteristics data to a trained machine learning model to identify one or more gait features for the specific patient, particularly that the trained model itself identifies the gait features from the extracted gait characteristics data (Huang, ¶[0026]; ¶¶[0029]-[0032]).
Arshad teaches filtering pelvis IMU data and providing anteroposterior, mediolateral, and vertical acceleration signals in moving windows as inputs to trained deep-learning models. The model outputs are right and left stance and swing phase signals, and transitions in those signals identify heel-strike and toe-off events (Arshad, p. 3-5, Sec. 2.1 and Figs. 2-3). Arshad trained CNN, RNN, LSTM, GRU, and hybrid CNN-RNN models using the pelvis IMU signals as inputs and the stance and swing phase signals as outputs (Arshad, p. 5-7, Sec. 2.2).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the modified Huang in view of Arshad so that the trained model identifies gait features from the gait characteristics data Huang already extracts. This modification would preserve Huang’s extraction step. Huang’s analyzer would continue to extract and organize directional or temporal gait-characteristic data from the patient’s acceleration data, and that extracted data would be supplied in moving windows to Arshad’s trained neural-network architecture. The model would identify heel-strike and toe-off events and corresponding stance and swing phases for the specific patient. Arshad is analogous art because it concerns wearable-pelvis IMU gait analysis using trained neural networks and is reasonably pertinent to identifying clinically meaningful gait features from acceleration data. One of ordinary skill would have had a reasonable expectation of success because Huang and Arshad process time-series gait acceleration data using trained machine learning models, Arshad expressly uses filtered and windowed pelvis IMU inputs, and Arshad demonstrates accurate model-based identification of gait events and phases. The benefit would have been accurate, automated identification of temporal patient gait features from wearable acceleration data.
Also regarding claim 1, the modified Huang applies a trained classifier to extracted gait features to select an outcome class representing a clinical parameter and teaches using the clinical parameter in diagnosing a medical condition, but does not expressly teach diagnosing the specific patient as having PAD or not having PAD based on the one or more identified gait features (Huang, ¶[0031]; ¶¶[0035]-[0036]).
Chidean teaches analyzing acceleration gait signals from PAD patients and control subjects and determining gait characteristics that discriminate between those groups. Chidean states that full spectral analysis “allowed to better discriminate PAD patients and control subjects” and “could be used for clinical early diagnosis” of PAD (Chidean, Abstract; p. 5-7, Sec. 6).
Rahman teaches using gait variability features to diagnose an individual as having PAD or not having PAD. Rahman performs logistic regression to determine whether each gait variable is associated with the presence or absence of PAD, calculates the probability of PAD for an individual observation, and classifies the observation as PAD or healthy based on an optimal cutoff value (Rahman, p. 5-6, Sec. 2.3.3). Rahman reports that multiple gait variability features provided acceptable to excellent discrimination between PAD patients and healthy controls (Rahman, p. 9, Sec. 3.3).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the modified Huang in view of Chidean and Rahman to diagnose a specific patient as having PAD or not having PAD based on the identified gait features. The modification would have been implemented by defining the classifier output classes as PAD and non-PAD, training the classifier using labeled gait features from PAD patients and controls, and applying the classifier to the specific patient’s identified gait features. One of ordinary skill would have had a reasonable expectation of success because the modified Huang already identifies and classifies patient gait features, Chidean identifies acceleration-derived gait characteristics that discriminate PAD patients from controls, and Rahman demonstrates individual PAD classification using gait features. The benefit would have been objective and earlier PAD diagnosis using wearable acceleration data. Taken together, the modifications would have advanced a common design objective for the modified Huang: objective, wearable, accelerometer-based PAD screening. Lim supplies acquisition near the sacrum using a minimally obstructive single-sensor arrangement, Arshad supplies trained-model identification of gait events and phases, and Chidean and Rahman provide the PAD-specific discrimination and diagnosis framework.
Regarding claim 2, the modified Huang teaches that the trained machine learning model comprises one of a neural network algorithm, a random forest algorithm, a support vector machine (SVM) algorithm or a Logit algorithm (as established above regarding claim 1, the modified Huang uses Arshad’s trained neural-network architecture as its machine learning model; Arshad trained CNN, RNN, LSTM, GRU, and hybrid CNN-RNN models using pelvis IMU signals (Arshad, p. 5-7, Sec. 2.2); those architectures are neural-network algorithms and therefore satisfy at least one of the alternatively recited machine learning models).
Regarding claim 3, the modified Huang teaches that the acceleration data related to the specific patient is obtained from one or more sensors worn by the specific patient (as established above regarding claim 1, Huang receives acceleration data from sensors worn by the subject, and Lim actually collects walking acceleration data from an inertial sensor worn at the lower back near the sacrum (Huang, ¶[0041]; ¶[0045]; Lim, p. 1, Abstract; p. 5, Sec. 2.2). The claim 1 combination mounts Lim’s inertial sensor near the sacrum and supplies the resulting acceleration data to the modified Huang for analysis, thereby obtaining the patient’s acceleration data from a sensor worn by the patient at the claimed location).
Regarding claim 4, the modified Huang teaches that the one or more gait features include one or more of step time asymmetry, step time variability, step time, stance time, stride time, and swing time (Arshad trains the model with pelvis IMU inputs and right and left stance and swing phase signals as outputs. Arshad defines stance as the period from heel strike to toe off and swing as the period from toe off to heel strike, and reports step, stance, and stride times (Arshad, p. 3, Table 1; p. 4-5, Sec. 2.1 and Figs. 2-3). Arshad therefore identifies at least stance time, stride time, or swing time, satisfying at least one of the alternatively recited gait features.)
Regarding claim 5, the modified Huang teaches the method of claim 1 as discussed above. Huang records three-dimensional accelerometer data in X, Y, and Z axes and analyzes that data to identify gait-related characteristics, but Huang does not expressly teach the gait characteristics data includes one or more of vertical acceleration data, anterior acceleration data, a number of steps, and a period of time between steps (Huang, ¶[0029]; ¶[0045]).
Chidean teaches analyzing vertical and anterior acceleration components as gait characteristics. In particular, Chidean evaluates “X-axis (V acceleration)” and “Y-axis (AP acceleration),” and identifies statistically significant differences in those acceleration components between PAD patients and controls (Chidean, p. 5, Sec. 5 and Fig. 5).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the modified Huang to include vertical acceleration data or anterior acceleration data as gait characteristics, as taught by Chidean. The modification would have been implemented by orienting or transforming Huang’s three-dimensional accelerometer data into vertical and anterior acceleration components and providing those components for gait analysis. One of ordinary skill would have had a reasonable expectation of success because Huang already records three-dimensional acceleration data and Chidean demonstrates analysis of vertical and anterior acceleration components from tri-axial gait sensors. The benefit would have been improved characterization of PAD-related gait differences using directional acceleration components shown by Chidean to distinguish PAD patients from controls.
Regarding claim 6, the modified Huang teaches the method of claim 1 as discussed above. Huang collects multi-axis accelerometer data and force-related underfoot data using wearable sensors, including three-axis accelerometers and insole-based pressure sensors, and further provides a Joint Angular and EMG Sensing Unit that records leg IMU data concurrently with gait measurements. However, Huang does not expressly teach wherein the gait characteristics data includes biomechanics data including one or more of braking impulse, braking peak, propulsive peak, propulsive impulse, other forces derived from ground reaction forces (GRF) data, and joint torques and powers, the joint torques and powers including hip, knee and/or ankle angles, torques and powers (Huang, ¶[0006]; ¶¶[0045]-[0046]).
Lim teaches estimating lower-limb kinetic quantities from acceleration data collected by a single IMU worn near the sacrum using an artificial neural network. Lim reports that “three joint torques, and two GRFs were estimated” from the kinematics measured by the single IMU (Lim, p. 1, Abstract). Lim further teaches feeding processed sacral IMU data to an artificial neural network having output nodes for the “joint torques of the hip, knee, and ankle” and the horizontal and vertical ground reaction forces (Lim, p. 3, Methods and Fig. 1).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the modified Huang in view of Lim so that the gait characteristics data includes hip, knee, or ankle joint torques. The modification would have been implemented by processing the acceleration data obtained from the sacral IMU, supplying the processed data to Lim’s artificial neural network to estimate hip, knee, and ankle joint torques, and including the resulting torque values in the gait characteristics data analyzed by the modified Huang. One of ordinary skill would have been motivated to make the modification because Lim identifies joint torques and ground reaction forces as quantitative indicators of human-motion performance and rehabilitation and demonstrates that those kinetic quantities can be estimated from data obtained by a single IMU worn near the sacrum (Lim, p. 1, Abstract and Introduction; p. 3, Methods and Fig. 1). One of ordinary skill would have had a reasonable expectation of success because the modified Huang already obtains and processes acceleration data from that anatomical region, and Lim successfully uses such data to estimate hip, knee, and ankle joint torques. The benefit would have been a more complete objective characterization of gait by supplementing acceleration-derived temporal and spatial information with complementary lower-limb kinetic information while retaining a minimally obstructive single-sensor configuration.
Regarding claim 8, Huang teaches that the system comprises: a processor and a memory storing instructions, which when executed by the processor causes the processor to access or execute a trained machine learning model (Huang, ¶[0027]: the mobile computing device includes a non-transitory memory and a processing resource; ¶[0028]: the memory stores machine-readable instructions executed by the processing resource to perform the operations of the analyzer and classifier; ¶[0031]: “Each classifier is trained on a plurality of training patterns representing various classes of interest”); wherein the processor is further configured to receive acceleration data related to the specific patient generated by the one or more sensors (Huang, ¶[0004]: the wearable device includes sensors comprising a three-axis accelerometer; ¶[0017]: the wearable device collects subject gait data and sends the data to the mobile computing device for analysis; ¶[0027]: the input/output unit receives the signal streamed by the wearable device); and extract gait characteristics data from the acceleration data related to the specific patient (Huang, claim 18: “analyzing, by the system, the sensor data to identify a pattern related to gait, balance or posture within the sensor data”; ¶[0029]: the analyzer identifies gait-related patterns and features including walking speed, stride height, stride length, cadence, and swing time).
Also regarding claim 8, Huang teaches executing a trained machine learning classifier and training that classifier using training patterns representing classes of interest, but does not expressly teach that the trained machine learning model has been trained with gait characteristic data extracted from acceleration data for patients known to have PAD and patients that do not have PAD (Huang, ¶¶[0031]-[0032]).
Chidean teaches recording acceleration gait signals from PAD patients and control subjects and extracting gait characteristics from those signals using spectral analysis. Chidean analyzes the acceleration signals for each subject and extracts gait characteristics including fundamental gait frequency, periodicity, variability, and spectral-envelope information (Chidean, p. 3, Sec. 3.2; p. 4, Sec. 4.2). Chidean further confirms that the control subjects did not suffer from cardiovascular disease and had an ABI greater than 1, while the PAD patients had a documented PAD diagnosis and an ABI below 0.9 (Chidean, p. 4, Sec. 4.1). Chidean reports that the resulting acceleration-derived gait characteristics discriminated PAD patients from control subjects (Chidean, Abstract).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Huang in view of Chidean to train Huang’s machine learning classifier using gait characteristic data extracted from acceleration signals collected from diagnosed PAD patients and confirmed non-PAD controls. The modification would have been implemented by extracting Chidean’s spectral gait characteristics from the acceleration signals and using those characteristics as Huang’s training patterns, with PAD and non-PAD designated as the classes of interest. One of ordinary skill would have had a reasonable expectation of success because Huang expressly trains its classifier using patterns representing classes of interest, and Chidean provides acceleration-derived gait characteristics that statistically discriminate diagnosed PAD patients from confirmed controls. The benefit would have been a PAD-specific trained classifier using objectively measured gait characteristics.
Also regarding claim 8, the modified Huang includes wearable sensors that generate acceleration data for a patient, but does not teach one or more sensors configured to be worn at or near a sacrum or lumbar region of a specific patient.
Lim teaches collecting walking data from a single IMU worn near the sacrum and using the sacral motion as a measurable approximation of the center of mass. Lim explains that the center-of-mass location permits lower-limb dynamic data to be predicted while addressing the tradeoff between data quantity and wearable convenience (Lim, p. 1, Abstract; p. 2, Introduction). Lim further describes an EBIMU-9DOFV4 inertial sensor positioned at the lower back near the sacrum and collecting sagittal-plane acceleration data during walking (Lim, p. 5, Sec. 2.2).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the modified Huang in view of Lim to configure the acceleration sensor to be worn near the patient’s sacrum. The modification would have been implemented by mounting Lim’s inertial sensor at the lower back near the sacrum to acquire corresponding acceleration data from the training subjects and the specific patient and supplying that data to the modified Huang’s machine learning system. One of ordinary skill would have had a reasonable expectation of success because the modified Huang and Lim use wearable accelerometers to acquire gait data, Lim actually collected gait acceleration data from a single IMU near the sacrum, and using the same sensor configuration for the training subjects and the specific patient would have provided corresponding training and inference inputs. The benefit would have been a minimally obstructive single-sensor arrangement near the body’s center of mass, with reduced sensor burden and improved wearable convenience.
Also regarding claim 8, the modified Huang extracts gait-related information from acceleration data and feeds extracted features to a trained machine learning classifier, but does not expressly teach that the processor is configured to feed the extracted gait characteristics data to the trained machine learning model to identify one or more gait features for the specific patient, particularly that the trained model itself identifies the gait features from the extracted gait characteristics data (Huang, ¶[0026]; ¶¶[0029]-[0032]).
Arshad teaches filtering pelvis IMU data and providing anteroposterior, mediolateral, and vertical acceleration signals in moving windows as inputs to trained deep-learning models. The model outputs are right and left stance and swing phase signals, and transitions in those signals identify heel-strike and toe-off events (Arshad, p. 3-5, Sec. 2.1 and Figs. 2-3). Arshad trained CNN, RNN, LSTM, GRU, and hybrid CNN-RNN models using the pelvis IMU signals as inputs and the stance and swing phase signals as outputs (Arshad, p. 5-7, Sec. 2.2).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the modified Huang in view of Arshad so that the trained model identifies gait features from the gait characteristics data Huang already extracts. The modification would preserve Huang’s extraction step. The processor would first extract and organize directional or temporal gait-characteristic data as Huang teaches, then supply that extracted data in moving windows to Arshad’s trained neural-network architecture, trained using the PAD and non-PAD gait-characteristic data established above. The model would identify heel-strike and toe-off events and corresponding stance and swing phases for the specific patient. Arshad is analogous art because it concerns wearable-pelvis IMU gait analysis using trained neural networks and is reasonably pertinent to identifying clinically meaningful gait features from acceleration data. One of ordinary skill would have had a reasonable expectation of success because the modified Huang and Arshad process time-series gait acceleration data using trained machine learning models, Arshad expressly uses filtered and windowed pelvis IMU inputs, and Arshad demonstrates accurate model-based identification of gait events and phases. The benefit would have been accurate, automated identification of temporal patient gait features from wearable acceleration data.
Also regarding claim 8, the modified Huang trains a classifier using acceleration-derived gait characteristics from PAD patients and confirmed controls and applies the trained classifier to gait features identified for the specific patient, but does not expressly teach configuring the processor to diagnose the specific patient as having PAD or not having PAD based on the one or more identified gait features.
Rahman teaches using gait variability features to diagnose an individual as having PAD or not having PAD. Rahman performs logistic regression to determine whether each gait variable is associated with the presence or absence of PAD, calculates the probability of PAD for an individual observation, and classifies the observation as PAD or healthy based on an optimal cutoff value (Rahman, p. 5-6, Sec. 2.3.3). Rahman reports that multiple gait variability features provided acceptable to excellent discrimination between PAD patients and healthy controls (Rahman, p. 9, Sec. 3.3).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the modified Huang in view of Rahman to configure the processor to diagnose a specific patient as having PAD or not having PAD based on the identified gait features. The modification would have been implemented by applying the trained PAD classifier to the specific patient’s identified gait features and treating the resulting PAD or non-PAD classification as the patient’s diagnosis, consistent with Rahman’s individual probability and cutoff-based classification. One of ordinary skill would have had a reasonable expectation of success because the modified Huang already trains and applies a classifier using acceleration-derived gait characteristics from PAD patients and controls, and Rahman demonstrates individual PAD classification using gait features. The benefit would have been objective and earlier PAD diagnosis using wearable acceleration data. Taken together, the modifications would have advanced a common design objective for the modified Huang: objective, wearable, accelerometer-based PAD screening. Lim supplies acquisition near the sacrum using a minimally obstructive single-sensor arrangement, Arshad supplies trained-model identification of gait events and phases, and Chidean and Rahman provide the PAD-specific discrimination and diagnosis framework.
Regarding claim 9, the modified Huang teaches the system of claim 8 as discussed above. Huang stores machine-readable instructions that cause the processor to perform the classifier operations in non-transitory memory, and the modified Huang uses Arshad’s trained neural-network architecture. However, the references do not expressly teach wherein code and data associated with the trained machine learning model is stored in the memory (Huang, ¶[0028]; Arshad, p. 5-7, Sec. 2.2).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have stored the executable neural-network code and the trained weights or other learned model parameters in Huang’s non-transitory memory. The modification would have been implemented by storing the executable model instructions and learned parameters in Huang’s RAM, flash memory, solid-state drive, or hard-disk drive. One of ordinary skill would have had a reasonable expectation of success because Huang’s processor already accesses the memory to execute the classifier and Arshad’s trained neural-network models use learned parameters to perform their operations. The benefit would have been efficient and reliable local execution of the trained machine learning model without repeatedly retrieving its code or learned parameters from an external source.
Regarding claim 10, the modified Huang teaches that the trained machine learning model comprises one of a neural network algorithm, a random forest algorithm, a support vector machine (SVM) algorithm or a Logit algorithm (as established above regarding claim 8, the modified Huang uses Arshad’s trained neural-network architecture as its machine learning model; Arshad trained CNN, RNN, LSTM, GRU, and hybrid CNN-RNN models using pelvis IMU signals (Arshad, p. 5-7, Sec. 2.2); those architectures are neural-network algorithms and therefore satisfy at least one of the alternatively recited machine learning models).
Regarding claim 11, the modified Huang teaches that the one or more gait features include one or more of step time asymmetry, step time variability, step time, stance time, stride time, and swing time (Arshad trains the model with pelvis IMU inputs and right and left stance and swing phase signals as outputs. Arshad defines stance as the period from heel strike to toe off and swing as the period from toe off to heel strike, and reports step, stance, and stride times (Arshad, p. 3, Table 1; p. 4-5, Sec. 2.1 and Figs. 2-3). Arshad therefore identifies at least stance time, stride time, or swing time, satisfying at least one of the alternatively recited gait features.)
Regarding claim 12, the modified Huang teaches the system of claim 8 as discussed above. Huang records three-dimensional accelerometer data and analyzes that data to identify gait-related characteristics, but Huang does not expressly teach wherein the gait characteristics data includes one or more of vertical acceleration, anterior acceleration, a number of steps, a period of time between steps, braking impulse, braking peak, propulsive peak, propulsive impulse, other forces derived from ground reaction forces (GRF) data, and joint torques and powers, the joint torques and powers including hip, knee and/or ankle angles, torques and powers (Huang, ¶[0029]; ¶[0045]).
Chidean teaches analyzing vertical and anterior acceleration components as gait characteristics. In particular, Chidean evaluates “X-axis (V acceleration)” and “Y-axis (AP acceleration)” and identifies statistically significant differences in those acceleration components between PAD patients and control subjects (Chidean, p. 5, Sec. 5 and Fig. 5).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the modified Huang to include vertical acceleration or anterior acceleration as gait characteristics data, as taught by Chidean. The modification would have been implemented by orienting or transforming Huang’s three-dimensional accelerometer data into vertical and anterior acceleration components and providing those components for gait analysis. One of ordinary skill would have had a reasonable expectation of success because Huang already records three-dimensional acceleration data and Chidean demonstrates analysis of vertical and anterior acceleration components from tri-axial gait sensors. The benefit would have been improved characterization of PAD-related gait differences using directional acceleration components shown by Chidean to distinguish PAD patients from control subjects. Because claim 12 recites these items in the alternative, Chidean’s vertical or anterior acceleration teaching satisfies the limitation without requiring a separate showing for each remaining listed item.
Regarding claim 14, Huang teaches training a machine learning model (Huang, ¶[0031]: “Each classifier is trained on a plurality of training patterns representing various classes of interest”; ¶[0032]: an SVM classifier and a convolutional neural network classifier each process training data); and receiving acceleration data related to a specific patient from a sensor (Huang, ¶[0006]: “The method includes receiving a signal comprising sensor data,” wherein the sensors include “a three-axis accelerometer”; ¶[0015]: the wearable device detects data related to a subject’s gait, balance, or posture and sends the data to a mobile computing device for analysis).
Also regarding claim 14, Huang trains a machine learning classifier using training patterns representing classes of interest, but does not expressly teach training the model with acceleration or accelerometer data for patients known to have PAD and patients that do not have PAD (Huang, ¶¶[0031]-[0032]).
Chidean teaches recording acceleration gait signals from PAD patients and control subjects using tri-axial wireless sensor nodes positioned at ankle and hip height on both sides of each subject. Chidean confirms that the control subjects did not suffer from cardiovascular disease and had an ABI greater than 1, while the PAD patients had a documented PAD diagnosis and an ABI below 0.9 (Chidean, p. 4, Sec. 4.1). Chidean further teaches analyzing the acceleration signals for each subject and reports that full spectral analysis discriminated PAD patients from control subjects (Chidean, Abstract; p. 4, Sec. 4.2).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Huang in view of Chidean to train Huang’s machine learning model with acceleration data for patients known to have PAD and patients that do not have PAD. The modification would have been implemented by collecting acceleration gait signals from diagnosed PAD patients and confirmed non-PAD controls and using those acceleration signals as Huang’s training patterns, with PAD and non-PAD designated as the classes of interest, consistent with Chidean’s cohort and analysis framework. One of ordinary skill would have had a reasonable expectation of success because Huang expressly trains its classifier using training patterns representing classes of interest, and Chidean provides acceleration gait signals from diagnosed PAD patients and confirmed controls that exhibit statistically distinguishable gait characteristics. The benefit would have been a PAD-specific trained classifier using objectively measured acceleration data.
Also regarding claim 14, the modified Huang receives acceleration data from wearable sensors but does not teach that the sensor is worn at or near a sacrum or lumbar region of the specific patient.
Lim teaches collecting walking data from a single IMU worn near the sacrum and using the sacral motion as a measurable approximation of the center of mass. Lim explains that the center-of-mass location permits lower-limb dynamic data to be predicted while addressing the tradeoff between data quantity and wearable convenience (Lim, p. 1, Abstract; p. 2, Introduction). Lim further describes an EBIMU-9DOFV4 inertial sensor positioned at the lower back near the sacrum and collecting sagittal-plane acceleration data during walking (Lim, p. 5, Sec. 2.2).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the modified Huang in view of Lim to configure the acceleration sensor to be worn near the patient’s sacrum. The modification would have been implemented by mounting Lim’s inertial sensor at the lower back near the sacrum to acquire corresponding acceleration data from the training subjects and the specific patient and supplying that data to the modified Huang’s machine learning system. One of ordinary skill would have had a reasonable expectation of success because the modified Huang and Lim use wearable accelerometers to acquire gait data, Lim actually collected gait acceleration data from a single IMU near the sacrum, and using the same sensor configuration for the training subjects and the specific patient would have provided corresponding training and inference inputs. The benefit would have been a minimally obstructive single-sensor arrangement near the body’s center of mass, with reduced sensor burden and improved wearable convenience.
Also regarding claim 14, the modified Huang receives acceleration data and uses a trained machine learning classifier, but Huang extracts gait features before supplying those features to its classifier and therefore does not expressly teach feeding the acceleration data to the trained machine learning model to identify one or more gait features for the specific patient (Huang, ¶[0026]; ¶¶[0029]-[0032]).
Arshad teaches feeding pelvis IMU signals, including anteroposterior, mediolateral, and vertical acceleration, in moving windows to trained deep-learning models. The trained models output right and left stance and swing phase signals, and transitions in those signals identify heel-strike and toe-off events (Arshad, p. 3-5, Sec. 2.1 and Figs. 2-3). Arshad trained CNN, RNN, LSTM, GRU, and hybrid CNN-RNN models using the pelvis IMU signals as inputs and the stance and swing phase signals as outputs (Arshad, p. 5-7, Sec. 2.2).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the modified Huang in view of Arshad to feed acceleration data to a trained machine learning model that identifies gait features for the specific patient. The modification would have been implemented by using Arshad’s trained neural-network architecture as the machine learning architecture of the modified Huang, training that architecture using the PAD and non-PAD acceleration data established above, and feeding corresponding directional acceleration signals from the specific patient to the trained model in moving windows. The model would identify heel-strike and toe-off events and corresponding stance and swing phases. Arshad is analogous art because it concerns wearable-pelvis IMU gait analysis using trained neural networks and is reasonably pertinent to identifying clinically meaningful gait features from acceleration data. One of ordinary skill would have had a reasonable expectation of success because the modified Huang and Arshad process time-series gait acceleration data using trained machine learning models, and Arshad demonstrates accurate model-based identification of gait events and phases from directly supplied pelvis IMU signals. The benefit would have been accurate, automated identification of temporal patient gait features from wearable acceleration data.
Also regarding claim 14, the modified Huang trains a classifier using acceleration data from PAD patients and confirmed controls and automatically identifies gait features from the specific patient’s acceleration data, but does not expressly teach diagnosing the specific patient as having PAD or not having PAD based on the one or more identified gait features.
Rahman teaches using gait variability features to diagnose an individual as having PAD or not having PAD. Rahman performs logistic regression to determine whether each gait variable is associated with the presence or absence of PAD, calculates the probability of PAD for an individual observation, and classifies the observation as PAD or healthy based on an optimal cutoff value (Rahman, p. 5-6, Sec. 2.3.3). Rahman reports that multiple gait variability features provided acceptable to excellent discrimination between PAD patients and healthy controls (Rahman, p. 9, Sec. 3.3).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the modified Huang in view of Rahman to diagnose a specific patient as having PAD or not having PAD based on the identified gait features. The modification would have been implemented by applying the trained PAD classifier to the specific patient’s identified gait features and treating the resulting PAD or non-PAD classification as the patient’s diagnosis, consistent with Rahman’s individual probability and cutoff-based classification. One of ordinary skill would have had a reasonable expectation of success because the modified Huang already trains and applies a classifier using acceleration data from PAD patients and controls, and Rahman demonstrates individual PAD classification using gait features. The benefit would have been objective and earlier PAD diagnosis using wearable acceleration data. Taken together, the modifications would have advanced a common design objective for the modified Huang: objective, wearable, accelerometer-based PAD screening. Lim supplies acquisition near the sacrum using a minimally obstructive single-sensor arrangement, Arshad supplies trained-model identification of gait events and phases, and Chidean and Rahman provide the PAD-specific discrimination and diagnosis framework.
Regarding claim 15, the modified Huang teaches that the machine learning model comprises a recurrent neural network or a long short-term memory (LSTM) model (as established above regarding claim 14, the modified Huang uses Arshad’s trained neural-network architecture, and Arshad expressly trained RNN, LSTM, bidirectional LSTM, stacked LSTM, CNN-LSTM, and other recurrent architectures using pelvis IMU signals (Arshad, p. 5-7, Sec. 2.2); Arshad’s express disclosure satisfies the alternatively recited machine learning models).
Regarding claim 16, Huang teaches a non-transitory computer-readable medium storing instructions, which when executed by one or more processors, cause the one or more processors to implement a method (Huang, ¶[0004]: a non-transitory computer-readable medium comprises computer-executable instructions that, when executed by a processing resource, perform operations including receiving wearable sensor data, analyzing the data to identify a gait-related pattern, and applying a machine learning-based classification; ¶[0039]: “The method can be stored in one or more non-transitory computer-readable media and executed by one or more processing resources”; ¶[0040]: the non-transitory memory stores machine-executable instructions that the processing resource executes to perform the method); training a machine learning model (Huang, ¶[0031]: “Each classifier is trained on a plurality of training patterns representing various classes of interest”; ¶[0032]: an SVM classifier and a convolutional neural network classifier each process training data); and receiving acceleration data related to a specific patient from a sensor (Huang, ¶[0006]: “The method includes receiving a signal comprising sensor data,” wherein the sensors include “a three-axis accelerometer”; ¶[0015]: the wearable device detects data related to a subject’s gait, balance, or posture and sends the data to a mobile computing device for analysis).
Also regarding claim 16, Huang trains a machine learning classifier using training patterns representing classes of interest, but does not expressly teach training the model with acceleration or accelerometer data for patients known to have PAD and patients that do not have PAD (Huang, ¶¶[0031]-[0032]).
Chidean teaches recording acceleration gait signals from PAD patients and control subjects using tri-axial wireless sensor nodes positioned at ankle and hip height on both sides of each subject. Chidean confirms that the control subjects did not suffer from cardiovascular disease and had an ABI greater than 1, while the PAD patients had a documented PAD diagnosis and an ABI below 0.9 (Chidean, p. 4, Sec. 4.1). Chidean further teaches analyzing the acceleration signals for each subject and reports that full spectral analysis discriminated PAD patients from control subjects (Chidean, Abstract; p. 4, Sec. 4.2).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Huang in view of Chidean to train Huang’s machine learning model with acceleration data for patients known to have PAD and patients that do not have PAD. The modification would have been implemented by collecting acceleration gait signals from diagnosed PAD patients and confirmed non-PAD controls and using those acceleration signals as Huang’s training patterns, with PAD and non-PAD designated as the classes of interest, consistent with Chidean’s cohort and analysis framework. One of ordinary skill would have had a reasonable expectation of success because Huang expressly trains its classifier using training patterns representing classes of interest, and Chidean provides acceleration gait signals from diagnosed PAD patients and confirmed controls that exhibit statistically distinguishable gait characteristics. The benefit would have been a PAD-specific trained classifier using objectively measured acceleration data.
Also regarding claim 16, the modified Huang receives acceleration data from wearable sensors but does not teach that the sensor is worn at or near a sacrum or lumbar region of the specific patient.
Lim teaches collecting walking data from a single IMU worn near the sacrum and using the sacral motion as a measurable approximation of the center of mass. Lim explains that the center-of-mass location permits lower-limb dynamic data to be predicted while addressing the tradeoff between data quantity and wearable convenience (Lim, p. 1, Abstract; p. 2, Introduction). Lim further describes an EBIMU-9DOFV4 inertial sensor positioned at the lower back near the sacrum and collecting sagittal-plane acceleration data during walking (Lim, p. 5, Sec. 2.2).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the modified Huang in view of Lim to configure the acceleration sensor to be worn near the patient’s sacrum. The modification would have been implemented by mounting Lim’s inertial sensor at the lower back near the sacrum to acquire corresponding acceleration data from the training subjects and the specific patient and supplying that data to the modified Huang’s machine learning system. One of ordinary skill would have had a reasonable expectation of success because the modified Huang and Lim use wearable accelerometers to acquire gait data, Lim actually collected gait acceleration data from a single IMU near the sacrum, and using the same sensor configuration for the training subjects and the specific patient would have provided corresponding training and inference inputs. The benefit would have been a minimally obstructive single-sensor arrangement near the body’s center of mass, with reduced sensor burden and improved wearable convenience.
Also regarding claim 16, the modified Huang receives acceleration data and uses a trained machine learning classifier, but Huang extracts gait features before supplying those features to its classifier and therefore does not expressly teach feeding the acceleration data to the trained machine learning model to identify one or more gait features for the specific patient (Huang, ¶[0026]; ¶¶[0029]-[0032]).
Arshad teaches feeding pelvis IMU signals, including anteroposterior, mediolateral, and vertical acceleration, in moving windows to trained deep-learning models. The trained models output right and left stance and swing phase signals, and transitions in those signals identify heel-strike and toe-off events (Arshad, p. 3-5, Sec. 2.1 and Figs. 2-3). Arshad trained CNN, RNN, LSTM, GRU, and hybrid CNN-RNN models using the pelvis IMU signals as inputs and the stance and swing phase signals as outputs (Arshad, p. 5-7, Sec. 2.2).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the modified Huang in view of Arshad to feed acceleration data to a trained machine learning model that identifies gait features for the specific patient. The modification would have been implemented by using Arshad’s trained neural-network architecture as the machine learning architecture of the modified Huang, training that architecture using the PAD and non-PAD acceleration data established above, and feeding corresponding directional acceleration signals from the specific patient to the trained model in moving windows. The model would identify heel-strike and toe-off events and corresponding stance and swing phases. Arshad is analogous art because it concerns wearable-pelvis IMU gait analysis using trained neural networks and is reasonably pertinent to identifying clinically meaningful gait features from acceleration data. One of ordinary skill would have had a reasonable expectation of success because the modified Huang and Arshad process time-series gait acceleration data using trained machine learning models, and Arshad demonstrates accurate model-based identification of gait events and phases from directly supplied pelvis IMU signals. The benefit would have been accurate, automated identification of temporal patient gait features from wearable acceleration data.
Also regarding claim 16, the modified Huang trains a classifier using acceleration data from PAD patients and confirmed controls and automatically identifies gait features from the specific patient’s acceleration data, but does not expressly teach diagnosing the specific patient as having PAD or not having PAD based on the one or more identified gait features.
Rahman teaches using gait variability features to diagnose an individual as having PAD or not having PAD. Rahman performs logistic regression to determine whether each gait variable is associated with the presence or absence of PAD, calculates the probability of PAD for an individual observation, and classifies the observation as PAD or healthy based on an optimal cutoff value (Rahman, p. 5-6, Sec. 2.3.3). Rahman reports that multiple gait variability features provided acceptable to excellent discrimination between PAD patients and healthy controls (Rahman, p. 9, Sec. 3.3).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the modified Huang in view of Rahman to diagnose a specific patient as having PAD or not having PAD based on the identified gait features. The modification would have been implemented by applying the trained PAD classifier to the specific patient’s identified gait features and treating the resulting PAD or non-PAD classification as the patient’s diagnosis, consistent with Rahman’s individual probability and cutoff-based classification. One of ordinary skill would have had a reasonable expectation of success because the modified Huang already trains and applies a classifier using acceleration data from PAD patients and controls, and Rahman demonstrates individual PAD classification using gait features. The benefit would have been objective and earlier PAD diagnosis using wearable acceleration data. Taken together, the modifications would have advanced a common design objective for the modified Huang: objective, wearable, accelerometer-based PAD screening. Lim supplies acquisition near the sacrum using a minimally obstructive single-sensor arrangement, Arshad supplies trained-model identification of gait events and phases, and Chidean and Rahman provide the PAD-specific discrimination and diagnosis framework.
Regarding claim 21, the modified Huang teaches the method of claim 1 as discussed above. Huang applies a trained classifier to gait-related features to produce a clinical parameter, teaches using a regression model to calculate a parameter representing the likelihood that the subject exhibits the clinical parameter, and teaches using the clinical parameter for diagnosis, stratification, or monitoring of a medical condition. However, Huang does not expressly teach determining a severity of PAD in the specific patient based on the gait characteristics data or determining a change in the severity of PAD in the specific patient based on the gait characteristics data (Huang, ¶¶[0035]-[0036]).
Chidean teaches using gait characteristics extracted from acceleration data to assess PAD severity and disease evolution. Chidean reports, as a preliminary result, that patients having intermittent claudication in both legs may exhibit a lower fundamental gait frequency than patients having intermittent claudication in one leg and explains that bilateral claudication can be considered a more severe condition. Chidean further states that full spectral analysis of acceleration gait signals could be used “to monitor the disease evolution in PAD patients” and showed promising results for assessing PAD severity (Chidean, p. 5-7, Sec. 6).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the modified Huang in view of Chidean to determine PAD severity or a change in PAD severity based on the specific patient’s gait characteristics data. The modification would have been implemented by using Chidean’s acceleration-derived fundamental gait frequency or related spectral gait characteristics to produce Huang’s clinical stratification or monitoring parameter, with gait characteristics associated with unilateral and bilateral claudication representing different levels of PAD severity. Changes in those gait characteristics relative to the patient’s earlier measurements would have been used to determine a change in severity. One of ordinary skill would have had a reasonable expectation of success because Huang already produces clinical parameters from gait features for stratification and monitoring, and Chidean identifies acceleration-derived gait characteristics associated with PAD severity and disease evolution. The benefit would have been objective assessment and longitudinal monitoring of PAD severity using wearable acceleration data.
Regarding claim 23, the modified Huang teaches providing an alert on a display device based on the specific patient being diagnosed as having PAD, or based on the severity of PAD in the specific patient, or based on the change in severity of PAD in the specific patient (Huang, ¶[0027]: the user interface displays visual information and can comprise a touchscreen; ¶[0037]: the alert generator provides an alert based on the clinical parameter, and the alert can comprise a visual signal; as established above regarding claim 1, the modified Huang configures the clinical parameter to represent the specific patient’s PAD or non-PAD diagnosis, such that Huang’s visual alert based on that clinical parameter is an alert based on the patient being diagnosed as having PAD; this satisfies at least one of the alternatively recited bases for providing the alert).
Regarding claim 24, the modified Huang teaches the system of claim 8 as discussed above. Huang applies a trained classifier to gait-related features to produce a clinical parameter, teaches using a regression model to calculate a parameter representing the likelihood that the subject exhibits the clinical parameter, and teaches using the clinical parameter for diagnosis, stratification, or monitoring of a medical condition. However, Huang does not expressly teach that the processor is further configured to determine a severity of PAD in the specific patient based on the gait characteristics data or determine a change in the severity of PAD in the specific patient based on the gait characteristics data (Huang, ¶¶[0035]-[0036]).
Chidean teaches using gait characteristics extracted from acceleration data to assess PAD severity and disease evolution. Chidean reports, as a preliminary result, that patients having intermittent claudication in both legs may exhibit a lower fundamental gait frequency than patients having intermittent claudication in one leg and explains that bilateral claudication can be considered a more severe condition. Chidean further states that full spectral analysis of acceleration gait signals could be used “to monitor the disease evolution in PAD patients” and showed promising results for assessing PAD severity (Chidean, p. 5-7, Sec. 6).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the modified Huang in view of Chidean to configure the processor to determine PAD severity or a change in PAD severity based on the specific patient’s gait characteristics data. The modification would have been implemented by using Chidean’s acceleration-derived fundamental gait frequency or related spectral gait characteristics to produce Huang’s clinical stratification or monitoring parameter, with gait characteristics associated with unilateral and bilateral claudication representing different levels of PAD severity. Changes in those gait characteristics relative to the patient’s earlier measurements would have been used to determine a change in severity. One of ordinary skill would have had a reasonable expectation of success because Huang already produces clinical parameters from gait features for stratification and monitoring, and Chidean identifies acceleration-derived gait characteristics associated with PAD severity and disease evolution. The benefit would have been objective assessment and longitudinal monitoring of PAD severity using wearable acceleration data.
Regarding claim 26, the modified Huang teaches that the processor is further configured to provide an alert on a display device based on the specific patient being diagnosed as having PAD, or based on the severity of PAD in the specific patient, or based on the change in severity of PAD in the specific patient (Huang, ¶[0027]: the user interface displays visual information and can comprise a touchscreen; ¶[0037]: the alert generator provides an alert based on the clinical parameter, and the alert can comprise a visual signal; as established above regarding claim 8, the modified Huang configures the clinical parameter to represent the specific patient’s PAD or non-PAD diagnosis, such that Huang’s visual alert based on that clinical parameter is an alert based on the patient being diagnosed as having PAD; this satisfies at least one of the alternatively recited bases for providing the alert).
Regarding claim 27, the modified Huang teaches that the one or more identified gait features are selected from the group consisting of vertical acceleration, anterior acceleration, step time, stance time, stride time, swing time, step time asymmetry, step time variability, stance-to-swing ratio, gait symmetry, ground reaction force features, and joint torques and powers, the joint torques and powers including hip, knee, or ankle angles, torques, or powers (Arshad trains the model with pelvis IMU inputs and right and left stance and swing phase signals as outputs, defines stance and swing by heel-strike and toe-off transitions, and reports step, stance, and stride times (Arshad, p. 3, Table 1; p. 4-5, Sec. 2.1 and Figs. 2-3). Arshad’s stance time, stride time, or swing time satisfies at least one member of the recited group).
Claims 7, 13, 22, 25, and 28 are rejected under 35 U.S.C. 103 as being unpatentable over Huang et al. (US 2018/0279915 A1), hereinafter referred to as Huang, and further in view of Lim et al. (Lim, Hyerim et al. “Prediction of Lower Limb Kinetics and Kinematics during Walking by a Single IMU on the Lower Back Using Machine Learning.” Sensors (Basel, Switzerland) vol. 20,1 130. 24 Dec. 2019), hereinafter referred to as Lim, and further in view of Arshad et al. (Arshad, Muhammad Zeeshan, et al. “Gait Events Prediction Using Hybrid CNN-RNN-Based Deep Learning Models through a Single Waist-Worn Wearable Sensor.” Sensors [Basel], vol. 22, no. 21, October 2022), hereinafter referred to as Arshad, and further in view of Chidean et al. (Chidean, Mihaela I. et al. “Full Band Spectra Analysis of Gait Acceleration Signals for Peripheral Arterial Disease Patients.” Frontiers in Physiology 9 (2018)), hereinafter referred to as Chidean, and further in view of Rahman et al. (Rahman, Hafizur et al. “Gait Variability Is Affected More by Peripheral Artery Disease than by Vascular Occlusion.” PLoS ONE 16.3 (2021)), hereinafter referred to as Rahman, and further in view of Flores et al. (Flores, Alyssa M. et al. “Leveraging Machine Learning and Artificial Intelligence to Improve Peripheral Artery Disease Detection, Treatment, and Outcomes.” Circulation Research 128.12 (2021)), hereinafter referred to as Flores.
The modified Huang teaches claim 1 as shown above.
The modified Huang teaches claim 8 as shown above.
Regarding claim 7, the modified Huang teaches the method of claim 1 as discussed above and feeds extracted patient gait features to a trained machine learning classifier. However, the modified Huang does not expressly teach wherein the step of feeding the trained machine learning model further includes feeding biometric data of the specific patient to the trained machine learning model (Huang, ¶¶[0031]-[0033]).
The instant specification uses ‘biometric data’ as additional patient data supplied to the machine learning model but provides no special definition restricting the term to identity-authentication data (Spec., ¶[0005]; ¶[0006]; ¶[0009]). Under the broadest reasonable interpretation, measurable physical or physiological characteristics of the patient fall within biometric data. Flores teaches using patient-specific physical and physiological data as inputs to machine learning analysis for PAD detection and identifies physical measures and physiological data obtained from wearables as available inputs (Flores, p. 1834, Data Types for ML and AI). Flores further describes a trained PAD classification model that integrated more than 120 patient baseline characteristics, including medical, genetic, and coronary angiography findings (Flores, p. 1835-1836, Vascular Disease Diagnosis, Disease Detection Using Supervised Learning Approaches). The model achieved an area under the receiver operating curve of 0.84 and a sensitivity of 76% for PAD classification.
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the modified Huang in view of Flores to feed biometric data of the specific patient to the trained machine learning model. The modification would have been implemented by adding measurable physical or physiological patient variables to the multidimensional feature vector supplied to the existing classifier. One of ordinary skill would have had a reasonable expectation of success because Huang’s classifier operates on multidimensional feature vectors and Flores demonstrates successful PAD classification using numerous patient-specific variables, including physical or physiological data, achieving an area under the receiver operating curve of 0.84 and a sensitivity of 76%. The benefit would have been a more individualized and clinically informative PAD assessment based on both gait characteristics and patient biometric data.
Regarding claim 13, the modified Huang teaches the system of claim 8, including a machine learning model trained with acceleration-derived gait characteristics from PAD patients and patients without PAD and feeding the specific patient’s gait characteristics to the trained model. However, the modified Huang does not expressly teach the trained machine learning model having been further trained with biometric data, wherein the instructions to feed the extracted gait characteristics data to the trained machine learning model further include instructions to feed biometric data of the specific patient to the trained machine learning model.
The instant specification uses ‘biometric data’ as additional patient data supplied to or used to train the machine learning model but provides no special definition restricting the term to identity-authentication data (Spec., ¶[0005]; ¶[0006]; ¶[0009]). Under the broadest reasonable interpretation, measurable physical or physiological characteristics of the patient fall within biometric data. Flores teaches training a PAD classification model using more than 120 patient baseline characteristics, including medical, genetic, and coronary angiography findings (Flores, p. 1835, Vascular Disease Diagnosis, Disease Detection Using Supervised Learning Approaches), and identifies physical measures and physiological data from wearables as patient data available for machine learning analysis (Flores, p. 1834, Data Types for ML and AI). Rahman provides concrete examples of measurable physical patient data collected for PAD patients and control subjects, including body mass, height, and anthropometric measurements (Rahman, p. 4, Secs. 2.1-2.2).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the modified Huang in view of Flores to train the machine learning model further with biometric data and to feed corresponding biometric data of the specific patient to the trained model. The modification would have been implemented by adding measurable physical or physiological patient variables, such as height, body mass, or other anthropometric measurements, to the training feature vectors for PAD patients and patients without PAD and adding the corresponding values for the specific patient to the feature vector supplied to the trained model. One of ordinary skill would have had a reasonable expectation of success because Huang’s classifier already processes multidimensional feature vectors, Flores demonstrates training a PAD classifier using numerous patient-specific characteristics, and Rahman demonstrates that physical and anthropometric measurements are routinely collected for PAD patients and control subjects (Huang, ¶¶[0031]-[0033]). The benefit would have been a more individualized and clinically informative PAD assessment by allowing the trained model to consider both gait characteristics and patient biometric data.
Regarding claim 22, the modified Huang teaches the method of claim 21, including determining PAD severity or a change in PAD severity from the specific patient’s gait characteristics data. Huang further teaches using a gait-derived clinical parameter to determine, track, or ensure compliance with an exercise or rehabilitation program. However, the modified Huang does not expressly teach selecting or modifying a treatment course for the specific patient based on the severity of PAD in the specific patient or the change in severity of PAD in the specific patient (Huang, ¶[0036]).
Flores teaches selecting treatment strategies for PAD patients and using machine learning to support individualized PAD treatment planning. Flores explains that the costs and risks of available PAD treatments should be considered when “selecting an appropriate treatment strategy,” that machine learning models can match preventive efforts to patients most likely to benefit from new or existing drugs, and that algorithms trained using PAD patient data can develop plans that synchronize therapy and limit adverse interactions (Flores, p. 1840, Identifying Appropriate Medical Treatment for Vascular Disease).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the modified Huang in view of Flores to select or modify a treatment course for the specific patient based on the determined PAD severity or change in severity. The modification would have been implemented by providing the PAD severity or severity-change parameter produced by the modified Huang as an input to a treatment-planning model or decision rule that selects or modifies an exercise, rehabilitation, preventive, or medical treatment course according to the patient’s current disease severity. One of ordinary skill would have had a reasonable expectation of success because Huang already uses gait-derived clinical parameters in connection with exercise and rehabilitation programs, the modified Huang determines PAD severity from those parameters, and Flores teaches using machine learning outputs and patient risk information to select and coordinate PAD treatment. The benefit would have been more personalized and responsive PAD management based on objectively determined disease severity.
Regarding claim 25, the modified Huang teaches the system of claim 24, including configuring the processor to determine PAD severity or a change in PAD severity from the specific patient’s gait characteristics data. Huang further teaches using a gait-derived clinical parameter to determine, track, or ensure compliance with an exercise or rehabilitation program. However, the modified Huang does not expressly teach that the processor is further configured to select or modify a treatment course for the specific patient based on the severity of PAD in the specific patient or the change in severity of PAD in the specific patient (Huang, ¶[0036]).
Flores teaches selecting treatment strategies for PAD patients and using machine learning to support individualized PAD treatment planning. Flores explains that the costs and risks of available PAD treatments should be considered when “selecting an appropriate treatment strategy,” that machine learning models can match preventive efforts to patients most likely to benefit from new or existing drugs, and that algorithms trained using PAD patient data can develop plans that synchronize therapy and limit adverse interactions (Flores, p. 1840, Identifying Appropriate Medical Treatment for Vascular Disease).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the modified Huang in view of Flores to configure the processor to select or modify a treatment course for the specific patient based on the determined PAD severity or change in severity. The modification would have been implemented by configuring the processor to provide the PAD severity or severity-change parameter as an input to a treatment-planning model or decision rule that selects or modifies an exercise, rehabilitation, preventive, or medical treatment course according to the patient’s current disease severity. One of ordinary skill would have had a reasonable expectation of success because Huang already uses gait-derived clinical parameters in connection with exercise and rehabilitation programs, the modified Huang determines PAD severity from those parameters, and Flores teaches using machine learning outputs and patient risk information to select and coordinate PAD treatment. The benefit would have been more personalized and responsive PAD management based on objectively determined disease severity.
Regarding claim 28, the modified Huang teaches the method of claim 22, including selecting or modifying a treatment course for the specific patient based on PAD severity or a change in PAD severity. However, the modified Huang does not expressly teach wherein the treatment course is selected from the group consisting of supervised exercise therapy, surgical intervention, an ankle-foot orthosis, and exoskeleton footwear.
Flores teaches supervised exercise therapy as a treatment for patients with PAD. Flores explains that reinforcement-learning systems could promote supervised exercise therapy for PAD patients, that supervised exercise therapy increasingly is delivered using mobile-health technologies, and that reinforcement-learning algorithms could be combined with the Society for Vascular Surgery supervised exercise therapy mobile application (Flores, p. 1845, Optimizing Behavior and Lifestyle Modifications).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the modified Huang in view of Flores to select supervised exercise therapy as the treatment course based on the determined PAD severity or change in severity. The modification would have been implemented by using the PAD severity parameter to select or modify a provider-prescribed supervised exercise program, including its frequency, duration, or intensity, and using Huang’s monitoring functionality to track the patient’s participation and response. One of ordinary skill would have had a reasonable expectation of success because the modified Huang already selects treatment based on PAD severity, Huang uses gait-derived clinical parameters in connection with exercise and rehabilitation programs, and Flores expressly teaches supervised exercise therapy for PAD patients. The benefit would have been individualized, noninvasive PAD treatment and objective monitoring of the patient’s response. Supervised exercise therapy satisfies at least one member of the recited group.
Response to Arguments
Objections to the Drawings
Applicant's arguments filed 6/4/2026, page 8, regarding the previous Objections to the Drawings have been fully considered and are persuasive. The previous Objections to the Drawings have been withdrawn.
Objections
Applicant's arguments filed 6/4/2026, page 8, regarding the previous Objections of claims 6, 8, 13, 21, and 25 have been fully considered and are persuasive. The previous Objections have been withdrawn.
35 U.S.C. §112(b)
Applicant's arguments filed 6/4/2026, page 9, regarding the previous 112(b) Rejections of claims 6 and 12 have been fully considered and are persuasive. The previous 112(b) rejections have been withdrawn.
35 U.S.C. §101
Applicant's arguments filed 6/4/2026, pages 9-10, regarding the previous 101 Rejections of claims 1-16 and 21-26 have been fully considered and are not persuasive for at least the reasons outlined below.
Applicant’s Argument: Applicant argues that the amendments to independent claims 1, 8, 14, and 16 overcome the rejection because the claims now require a sensor configured to be worn at or near the sacrum or lumbar region, processing acceleration data using a trained machine learning model, and diagnosing whether a specific patient has PAD. Applicant argues that this combination is not an abstract idea and instead provides a technical improvement and a more robust PAD diagnosis.
Examiner’s Response: The argument is not persuasive. Under Step 2A, Prong One, the rejection does not characterize the claims merely as directed to a mathematical concept or disregard the claims as a whole. The claims recite obtaining patient information, identifying gait information from that information, and using the identified information to determine whether the patient has PAD. The diagnostic determination is a mental process because a person can observe or review an identified gait feature and determine whether that feature indicates PAD. The claims require only “one or more” gait features and therefore encompass making the determination from a single feature. The possibility that certain embodiments may process a large quantity of data or employ a complicated model does not remove from the claim scope embodiments in which the recited evaluation can practically be performed in the human mind.
Under Step 2A, Prong Two, the recited sensor location does not integrate the judicial exception into a practical application. The sensor obtains the acceleration data used in the diagnostic evaluation and therefore constitutes data gathering for the recited analysis. Specifying the anatomical source from which the data are collected does not, without more, improve the operation of the sensor, computer, or machine learning model.
Similarly, the claims do not recite a particular signal-processing technique, feature-extraction algorithm, machine learning architecture, training procedure, model parameter, or other technological mechanism that produces the asserted improvement. The claims recite the desired functional results of extracting data, providing data to a trained model, identifying features, and diagnosing PAD. Unclaimed implementation details appearing in the specification cannot be imported into the claims to establish eligibility. Applicant’s assertion that the claimed combination produces a more robust PAD diagnosis identifies a desired result, but the claims do not recite a particular technological mechanism that produces that result.
Applicant’s Argument: Applicant argues that the claims are analogous to the claims found eligible in CardioNet, LLC v. InfoBionic, Inc., because they recite a physical configuration tied to a specific diagnostic task.
Examiner’s Response: The argument is not persuasive. CardioNet concerned claims directed to a specific improved cardiac monitoring device and a particular manner of distinguishing cardiac conditions. The present claims use a wearable acceleration sensor, data processing, and a trained machine learning model according to their ordinary functions to obtain and evaluate gait information. The claims do not recite a comparable technological mechanism that changes how the sensor or computer operates. Restricting the information analysis to PAD diagnosis and specifying the anatomical source of the data do not provide the type of technological improvement present in CardioNet.
Applicant’s Argument: Applicant argues that the claims are analogous to those in McRO, Inc. v. Bandai Namco Games America Inc., because they recite a particular ordered combination of operations for diagnosing PAD.
Examiner’s Response: The argument is not persuasive. The claims in McRO recited a particular ordered set of unconventional rules that automated a process in a manner different from the process previously performed by human animators. The present claims do not recite specific rules governing the extraction of gait data, the identification of gait features, or the PAD determination. Instead, the claims recite those operations primarily according to their desired results. The ordering of data collection, analysis, and diagnosis does not provide the particularized technological rules that were material in McRO.
Applicant’s Argument: Applicant argues that Electric Power Group, LLC v. Alstom S.A. is distinguishable because the present claims require a sensor having a particular physical configuration and placement.
Examiner’s Response: The argument is not persuasive. The recited sensor remains a source of information collected for subsequent analysis, and the claimed analysis remains focused on identifying information and drawing a diagnostic conclusion from that information. Adding a physical sensor to collect the information, or specifying the anatomical location from which the information is collected, does not by itself transform the information-focused process into an eligible technological improvement.
Applicant’s Argument: Applicant argues that the claimed combination amounts to significantly more than any alleged abstract idea because it requires a physically positioned sensor, extraction of gait-characteristics data, a trained machine learning model, identification of patient-specific gait features, and diagnosis of PAD.
Examiner’s Response: The argument is not persuasive. Under Step 2B, the additional elements, individually and as an ordered combination, do not amount to significantly more than the judicial exception. Vining expressly describes the employed sensors as commercially available or off-the-shelf devices. Lim and Teufl further demonstrate the established use of wearable inertial sensors at or near the sacrum or lower back for gait analysis. Applicant’s specification describes the processor, memory, communication components, and machine learning techniques at a general functional level rather than as newly developed technological components. The record therefore supports that the additional technological elements are employed according to their established functions.
The ordered combination likewise does not recite an inventive concept. It uses a wearable sensor to collect acceleration information, conventional computing components to process that information, and a trained machine learning model to identify information used in reaching a diagnostic conclusion. Applying generic machine learning to a particular type of data or field of use, without claiming a technological improvement to the machine learning system itself, does not supply an inventive concept. Accordingly, Applicant’s arguments do not overcome the rejection of claims 1-16 and 21-28 under 35 U.S.C. 101.
35 U.S.C. §103
Applicant's arguments filed 6/4/2026, pages 10-13, regarding the previous 103 Rejections of claims 1-16 and 21-26 have been fully considered but are moot because the new ground of rejection does not rely on any reference applied in the prior rejection of record for any teaching or matter specifically challenged in the argument. That is, there are new grounds of rejection. Applicant's amendment requiring a single sensor at the sacrum changed the sensing configuration of the claimed invention as a whole and necessitated re-evaluation of the entire combination, including reliance on Lim for the sacral location and on Arshad for identifying gait features from a single trunk-worn sensor, rendering this action final under MPEP 706.07(a). Additionally:
Applicant’s Argument: Applicant argues that Huang, Chidean, and Rahman do not teach or suggest configuring the sensor to be worn at or near the sacrum or lumbar region. Applicant notes that Huang places sensors in a shoe insole, Chidean uses ankle and hip sensors, and Rahman obtains measurements in a laboratory setting.
Examiner’s Response: While the argument is persuasive regarding the references applied in the previous office action (2/18/2026), it does not overcome the rejection presently set forth. The present rejection relies on Lim (combined with Huang, Arshad, Chidean, and Rahman), rather than Huang, Chidean, or Rahman, for the amended sensor-location limitation. Lim expressly teaches obtaining gait data using a single inertial measurement unit located near the sacrum. Lim therefore supplies the anatomical sensor placement that Applicant correctly observes is not expressly disclosed by Huang, Chidean, or Rahman.
Applicant’s Argument: Applicant argues that the previous rejection did not articulate a reason why one of ordinary skill would have relocated the sensors disclosed by the cited references to the sacrum or lumbar region.
Examiner’s Response: The argument is persuasive regarding the previous rejection, but it does not overcome the rejection presently set forth. The present rejection relies on Lim (combined with Huang, Arshad, Chidean, and Rahman) for both the sacral sensor location and the reason for using that location. Lim explains that using multiple body-mounted IMUs limits the applicability of gait-monitoring systems as wearable devices and identifies a need to increase the quantity and quality of measured gait information with minimal increase in system complexity. Lim teaches that motion measured near the sacrum approximates center-of-mass motion, which is dynamically related to lower-limb gait mechanics, and demonstrates that a single sacral IMU can provide information concerning multiple lower-limb movements and forces. Lim further identifies improved wearable convenience as a benefit of obtaining this information from a single sensor near the sacrum (Lim, p. 1, Abstract; p. 2, Introduction; p. 3, Sec. 2.1).
Accordingly, one of ordinary skill would have been motivated to configure at least one of Huang’s wearable acceleration sensors to be worn near the sacrum so that Huang’s gait-analysis system could obtain centrally measured information representative of overall lower-limb gait while reducing the number of body-mounted sensors, limiting system complexity, and improving wearable convenience. One of ordinary skill would have had a reasonable expectation of success because Huang and Lim both obtain gait information from wearable acceleration sensors, and Lim demonstrates successful collection and analysis of gait acceleration data using a single sensor near the sacrum. Thus, the modification is supported by Lim’s express design objective and disclosed advantages, rather than by Applicant’s disclosure.
Applicant’s Argument: Applicant argues that the references fail to teach the complete claimed sequence of extracting gait-characteristics data, providing the extracted data to a trained machine learning model, identifying one or more gait features, and diagnosing the patient as having or not having PAD.
Examiner’s Response: The argument is not persuasive because it addresses the references individually rather than the collective teachings of the proposed combination. Huang supplies the wearable acquisition and extraction of patient gait characteristics. For claims 1 and 8, Huang’s extraction step is preserved, and the resulting extracted gait-characteristics data are provided to the trained model. Arshad supplies the use of a trained neural network to identify gait events and gait phases from gait sensor data. Chidean teaches that acceleration-derived gait characteristics discriminate PAD patients from control subjects and may be used for early PAD diagnosis. Rahman teaches determining the presence or absence of PAD for an individual using gait variables and a trained classification technique.
Each reference is applied for the teaching identified in the rejection, and the articulated modifications collectively produce the claimed sequence. The combination is directed to the common purpose of obtaining objective gait information from a wearable acceleration sensor, identifying clinically meaningful gait features, and using those features for PAD screening. Lim supplies a suitable centrally positioned wearable measurement arrangement, Arshad supplies trained-model identification of gait events and phases, and Chidean and Rahman supply the relationship between gait features and PAD diagnosis. These modifications represent predictable uses of known gait-sensing and data-analysis techniques for their established purposes.
Applicant’s Argument: Applicant argues that independent claims 8, 14, and 16 are patentable for substantially the same reasons presented regarding claim 1.
Examiner’s Response: The argument is not persuasive for the same reasons discussed regarding claim 1. The individual rejections account for the differences in the respective claim language. In particular, claims 1 and 8 preserve Huang’s initial extraction of gait-characteristics data before the extracted data are provided to Arshad’s trained model, whereas claims 14 and 16 permit the acceleration data to be provided to the trained model more directly. Lim supplies the amended sacral sensor placement, and Chidean and Rahman supply the PAD-specific teachings for each independent claim.
Applicant’s Argument: Applicant argues that the dependent claims are patentable because they depend from independent claims that are allegedly patentable.
Examiner’s Response: The argument is not persuasive. Because Applicant’s arguments do not establish the patentability of the independent claims, dependency alone does not overcome the rejections of the dependent claims. Applicant has not presented separate substantive arguments identifying an error in the particular prior-art mappings or obviousness rationales applied to the additional limitations of the dependent claims. Those additional limitations are separately addressed in the rejections above.
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. 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 nonprovisional extension fee (37 CFR 1.17(a)) 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 mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the
examiner should be directed to AARON MERRIAM whose telephone number is (703) 756-
5938. The examiner can normally be reached M-F 8:00 am - 5:00 pm.
Examiner interviews are available via telephone, in-person, and video conferencing
using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is
encouraged to use the USPTO Automated Interview Request (AIR) at
http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s
supervisor, Jason Sims can be reached on (571)272-4867. The fax phone number for the
organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained
from Patent Center. Unpublished application information in Patent Center is available to
registered users. To file and manage patent submissions in Patent Center, visit:
https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for
more information about Patent Center and https://www.uspto.gov/patents/docx for
information about filing in DOCX format. For additional questions, contact the Electronic
Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO
Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/AARON MERRIAM/Examiner, Art Unit 3791
/JASON M SIMS/Supervisory Patent Examiner, Art Unit 3791
Prosecution Insights