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 3/17/2026, 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 3/17/2026, and therefore rejections newly made in the instant office action have been necessitated by amendment.
Claims 1-4, 7, 9-11, 13-15, and 17-21 are the currently pending claims hereby under examination. Claims 1, 11, 15, 19, and 21 have been amended.
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 1-4, 7, 9, 10-11, 13-14, and 21 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 1-4, 7, 9, 10-11, 13-14, and 21 are directed to a method of diagnosing diabetic retinopathy using a computational algorithm, which is an abstract idea. Claims 1-4, 7, 9, 10-11, 13-14, and 21 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 drawn to a process.
Step 2A – Prong One: Claim 1 recites an abstract idea. In particular, claim 1 recites the following limitations:
A1: Receiving image data including an image of the retina of a subject
B1: Processing the image data to segment the image of the retina
C1: Extracting at least one feature from the segmented image of the retina
D1: Receiving demographic data and clinical data associated with the subject
E1: Generating, using a machine learning classifier, a diagnosis based at least in part on the at least one feature, the demographic data, and the clinical data
These elements A1-E1 of claim 1 are drawn to an abstract idea since (1) they involve mathematical concepts in the form of mathematical relationships, mathematical formulas or equations, and/or mathematical calculations; and/or (2) they involve a mental process that can be practically performed in the human mind including observation, evaluation, judgment, and opinion and using pen and paper.
Step 2A – Prong Two: Claim 1 recites the following limitations that are beyond the judicial exception:
A2: a computer
The element A2 of claim 1 does not integrate the exception into a practical application of the exception. In particular, the element A2 is merely an instruction to implement an abstract idea on a computer, as it does not provide a technological improvement to machine learning models beyond generic classification - see MPEP 2106.04(d) and MPEP 2106.05(f).
Under Step 2A, Prong 2, it is determined whether the claim as a whole integrates the judicial exception into a practical application. In view of the updated guidance and Desjardins, this analysis considers the description in the specification to determine whether the claim reflects an improvement in the functioning of a computer, an improvement to another technology or technical field in the Enfish / McRO / Desjardins sense, or merely uses generic computer components as tools.
The specification describes using a conventional OCT/OCTA imaging system to acquire retinal images of a subject, employing known image processing and segmentation techniques to partition OCT images into multiple retinal layers and OCTA images into vascular regions, extracting numerical image-based and clinical features (including thickness, reflectivity, vessel density, foveal avascular zone metrics, and counts of vascular bifurcations or crossovers), and then applying a supervised learning classifier, such as a random forest, to classify subjects as normal or having various grades of diabetic retinopathy. The asserted advantages are improved accuracy and objectivity of diabetic-retinopathy diagnosis and grading compared to manual clinical evaluation.
However:
The claim does not recite any improvement to the functioning of the computer itself (e.g., a new memory structure, processor operation, or data storage scheme), nor does the specification describe such an improvement.
The claim does not recite a new or improved structure or operation of the OCT or OCTA hardware (e.g., no new scanning protocol, hardware configuration, or acquisition technique). The imaging system is treated as a conventional data source.
The machine-learning classifier is recited at a high level as a “random forest” operating on extracted features. The specification does not describe a new random-forest architecture, training procedure, or representation that solves a specific technical problem in machine learning or computer technology (such as storage reduction, mitigation of catastrophic forgetting, or improved computational efficiency), in contrast to the type of improvements discussed in Desjardins. Instead, the classifier is used as a conventional tool to map input features to a DR diagnosis or grade.
The additional elements beyond the abstract idea (e.g., “one or more processors”, a “computing system”, an OCT/OCTA imaging device, and storage of data) are recited at a generic, functional level and perform their basic, expected functions of data acquisition, data manipulation, and execution of mathematical/ML algorithms. They merely apply the abstract diagnostic and mathematical concepts using conventional imaging equipment and general-purpose computing components.
Thus, when considered in light of the specification and as a whole, the claim does not reflect an improvement in computer functionality or an improvement to another technology in the Enfish / McRO / Desjardins sense. Instead, it uses generic imaging and computing technology as tools to implement the abstract diagnostic idea. The claim therefore does not integrate the judicial exception into a practical application under Step 2A, Prong 2. See MPEP §§ 2106.04(d), 2106.05(a)–(c), (f).
Step 2B: Claim 1 does not recite additional elements that amount to significantly more than the judicial exception itself. In particular, the element A2 requires no more than a generic computer to perform routine classification tasks that are well-understood, routine, and conventional activities previously known in the industry (see Electric Power Group, 830 F.3d 1350 (Fed. Cir. 2016); Alice Corp. v. CLS Bank Int’l, 110 USPQ2d 1976 (2014)) and/or a claim to an abstract idea requiring no more than being stored on a computer readable medium which is a well-understood, routine and conventional activity previously known in the industry (see Electric Power Group, 830 F.3d 1350 (Fed. Cir. 2016); Alice Corp. v. CLS Bank Int’l, 110 USPQ2d 1976 (2014); SAP Am. v. InvestPic, 890 F.3d 1016 (Fed. Circ. 2018)).
Claims 2-4, 7, 9, 10-11, 13-14, and 21 depend from claim 1 and recite the same abstract idea as claim 1. Furthermore, these claims only contain recitations that further limit the abstract idea (that is, the claims only refine the classification and data processing without adding a specific technological improvement).
Claims 15 and 17-18 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 15 and 17-18 are directed to a method of diagnosing diabetic retinopathy using a computational algorithm, which is an abstract idea. Claims 15 and 17-18 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 15 is as follows:
Step 1: Claim 15 is drawn to a process.
Step 2A – Prong One: Claim 15 recites an abstract idea. In particular, claim 15 recites the following limitations:
A1: Processing OCT/OCTA image data including an image of the subject retina to segment the image of the subject retina
B1: Extracting at least one feature from the segmented image of the retina
C1: Receiving demographic data and clinical data associated with the subject retina
D1: Classifying, using a machine learning classifier, the subject retina as normal or indicative of diabetic retinopathy based at least in part on the at least one feature, the demographic data, and the clinical data
These elements A1-D1 of claim 15 are drawn to an abstract idea since (1) they involve mathematical concepts in the form of mathematical relationships, mathematical formulas or equations, and/or mathematical calculations; and/or (2) they involve a mental process that can be practically performed in the human mind including observation, evaluation, judgment, and opinion and using pen and paper.
Step 2A – Prong Two: Claim 15 recites the following limitations that are beyond the judicial exception:
A2: a computer.
The element A2 of claim 15 does not integrate the exception into a practical application of the exception. In particular, the element A2 is merely an instruction to implement an abstract idea on a computer, as it does not provide a technological improvement to machine learning models beyond generic classification - see MPEP 2106.04(d) and MPEP 2106.05(f).
Under Step 2A, Prong 2, it is determined whether the claim as a whole integrates the judicial exception into a practical application. In view of the updated guidance and Desjardins, this analysis considers the description in the specification to determine whether the claim reflects an improvement in the functioning of a computer, an improvement to another technology or technical field in the Enfish / McRO / Desjardins sense, or merely uses generic computer components as tools.
The specification describes using a conventional OCT/OCTA imaging system to acquire retinal images of a subject, employing known image processing and segmentation techniques to partition OCT images into multiple retinal layers and OCTA images into vascular regions, extracting numerical image-based and clinical features (including thickness, reflectivity, vessel density, foveal avascular zone metrics, and counts of vascular bifurcations or crossovers), and then applying a supervised learning classifier, such as a random forest, to classify subjects as normal or having various grades of diabetic retinopathy. The asserted advantages are improved accuracy and objectivity of diabetic-retinopathy diagnosis and grading compared to manual clinical evaluation.
However:
The claim does not recite any improvement to the functioning of the computer itself (e.g., a new memory structure, processor operation, or data storage scheme), nor does the specification describe such an improvement.
The claim does not recite a new or improved structure or operation of the OCT or OCTA hardware (e.g., no new scanning protocol, hardware configuration, or acquisition technique). The imaging system is treated as a conventional data source.
The machine-learning classifier is recited at a high level as a “random forest” operating on extracted features. The specification does not describe a new random-forest architecture, training procedure, or representation that solves a specific technical problem in machine learning or computer technology (such as storage reduction, mitigation of catastrophic forgetting, or improved computational efficiency), in contrast to the type of improvements discussed in Desjardins. Instead, the classifier is used as a conventional tool to map input features to a DR diagnosis or grade.
The additional elements beyond the abstract idea (e.g., “one or more processors”, a “computing system”, an OCT/OCTA imaging device, and storage of data) are recited at a generic, functional level and perform their basic, expected functions of data acquisition, data manipulation, and execution of mathematical/ML algorithms. They merely apply the abstract diagnostic and mathematical concepts using conventional imaging equipment and general-purpose computing components.
Thus, when considered in light of the specification and as a whole, the claim does not reflect an improvement in computer functionality or an improvement to another technology in the Enfish / McRO / Desjardins sense. Instead, it uses generic imaging and computing technology as tools to implement the abstract diagnostic idea. The claim therefore does not integrate the judicial exception into a practical application under Step 2A, Prong 2. See MPEP §§ 2106.04(d), 2106.05(a)–(c), (f).
Step 2B: Claim 15 does not recite additional elements that amount to significantly more than the judicial exception itself. In particular, the element A2 does not qualify as significantly more because this limitation requires no more than a generic computer to perform routine classification tasks that are well-understood, routine, and conventional activities previously known in the industry (see Electric Power Group, 830 F.3d 1350 (Fed. Cir. 2016); Alice Corp. v. CLS Bank Int’l, 110 USPQ2d 1976 (2014)) and/or a claim to an abstract idea requiring no more than being stored on a computer readable medium which is a well-understood, routine and conventional activity previously known in the industry (see Electric Power Group, 830 F.3d 1350 (Fed. Cir. 2016); Alice Corp. v. CLS Bank Int’l, 110 USPQ2d 1976 (2014); SAP Am. v. InvestPic, 890 F.3d 1016 (Fed. Circ. 2018)).
Claims 17-18 depend from claim 15 and recite the same abstract idea as claim 1. Furthermore, these claims only contain recitations that further limit the abstract idea (that is, the claims only refine the classification and data processing without adding a specific technological improvement).
Claims 19-20 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 19-20 are directed to a method of diagnosing diabetic retinopathy using a computational algorithm, which is an abstract idea. Claims 19-20 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 19 is as follows:
Step 1: Claim 19 is drawn to a machine.
Step 2A – Prong One: Claim 19 recites an abstract idea. In particular, claim 19 recites the following limitations:
A1: receiving at least one feature extracted from OCA image data of a subject retina
B1: receiving at least one feature extracted from OCTA image data of the subject retina
C1: receiving demographic data and clinical data associated with the subject retina
D1: classifying the subject retina as normal or indicative of diabetic retinopathy based at least in part on the at least one feature, the demographic data, and the clinical data
These elements A1-D1 of claim 19 are drawn to an abstract idea since (1) they involve mathematical concepts in the form of mathematical relationships, mathematical formulas or equations, and/or mathematical calculations; and/or (2) they involve a mental process that can be practically performed in the human mind including observation, evaluation, judgment, and opinion and using pen and paper.
Step 2A – Prong Two: Claim 19 recites the following limitations that are beyond the judicial exception:
A2: A non-transitory computer readable storage medium having computer program instructions stored thereon that, when executed by a processor, cause the processor to perform the following instructions
The element A2 of claim 19 do not integrate the exception into a practical application of the exception. In particular, the element A2 is merely an instruction to implement an abstract idea on a computer, as it does not provide a technological improvement to machine learning models beyond generic classification - see MPEP 2106.04(d) and MPEP 2106.05(f).
Under Step 2A, Prong 2, it is determined whether the claim as a whole integrates the judicial exception into a practical application. In view of the updated guidance and Desjardins, this analysis considers the description in the specification to determine whether the claim reflects an improvement in the functioning of a computer, an improvement to another technology or technical field in the Enfish / McRO / Desjardins sense, or merely uses generic computer components as tools.
The specification describes using a conventional OCT/OCTA imaging system to acquire retinal images of a subject, employing known image processing and segmentation techniques to partition OCT images into multiple retinal layers and OCTA images into vascular regions, extracting numerical image-based and clinical features (including thickness, reflectivity, vessel density, foveal avascular zone metrics, and counts of vascular bifurcations or crossovers), and then applying a supervised learning classifier, such as a random forest, to classify subjects as normal or having various grades of diabetic retinopathy. The asserted advantages are improved accuracy and objectivity of diabetic-retinopathy diagnosis and grading compared to manual clinical evaluation.
However:
The claim does not recite any improvement to the functioning of the computer itself (e.g., a new memory structure, processor operation, or data storage scheme), nor does the specification describe such an improvement.
The claim does not recite a new or improved structure or operation of the OCT or OCTA hardware (e.g., no new scanning protocol, hardware configuration, or acquisition technique). The imaging system is treated as a conventional data source.
The machine-learning classifier is recited at a high level as a “random forest” operating on extracted features. The specification does not describe a new random-forest architecture, training procedure, or representation that solves a specific technical problem in machine learning or computer technology (such as storage reduction, mitigation of catastrophic forgetting, or improved computational efficiency), in contrast to the type of improvements discussed in Desjardins. Instead, the classifier is used as a conventional tool to map input features to a DR diagnosis or grade.
The additional elements beyond the abstract idea (e.g., “one or more processors”, a “computing system”, an OCT/OCTA imaging device, and storage of data) are recited at a generic, functional level and perform their basic, expected functions of data acquisition, data manipulation, and execution of mathematical/ML algorithms. They merely apply the abstract diagnostic and mathematical concepts using conventional imaging equipment and general-purpose computing components.
Thus, when considered in light of the specification and as a whole, the claim does not reflect an improvement in computer functionality or an improvement to another technology in the Enfish / McRO / Desjardins sense. Instead, it uses generic imaging and computing technology as tools to implement the abstract diagnostic idea. The claim therefore does not integrate the judicial exception into a practical application under Step 2A, Prong 2. See MPEP §§ 2106.04(d), 2106.05(a)–(c), (f).
Step 2B: Claim 19 does not recite additional elements that amount to significantly more than the judicial exception itself. In particular, the element A2 requires no more than a generic computer to perform routine classification tasks that are well-understood, routine, and conventional activities previously known in the industry (see Electric Power Group, 830 F.3d 1350 (Fed. Cir. 2016); Alice Corp. v. CLS Bank Int’l, 110 USPQ2d 1976 (2014)) and/or a claim to an abstract idea requiring no more than being stored on a computer readable medium which is a well-understood, routine and conventional activity previously known in the industry (see Electric Power Group, 830 F.3d 1350 (Fed. Cir. 2016); Alice Corp. v. CLS Bank Int’l, 110 USPQ2d 1976 (2014); SAP Am. v. InvestPic, 890 F.3d 1016 (Fed. Circ. 2018)).
Claim 20 depends from claim 19 and recite the same abstract idea as claim 1. Furthermore, these claims only contain recitations that further limit the abstract idea (that is, the claims only refine the classification and data processing without adding a specific technological improvement).
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 1-4, 7, 9-10, 13-14, and 21 are rejected under 35 U.S.C. 103 as being unpatentable over Sandhu et al. (Sandhu, Harpal Singh et al. “Automated Diagnosis and Grading of Diabetic Retinopathy Using Optical Coherence Tomography.” Investigative ophthalmology & visual science 59.7 (2018): 3155–3160. Web.), hereto referred as Sandhu, and further in view of Choi et al. (Choi, WooJhon et al. ”Ultrahigh Speed OCT Angiography of Retinal and Choriocapillaris Alterations in Diabetic Patients with and without Retinopathy Using Swept Source Optical Coherence Tomography”. Retina 37(1):p 11-21, January 2017), hereto referred as Choi, and further in view of Huang et al. (US 20160284085 A1), hereto referred as Huang, and further in view of Bhuiya et al. (Bhuiyan et al. “Detection and Classification of Bifurcation and Branch Points on Retinal Vascular Network.” 2012 International Conference on Digital Image Computing: Techniques and Applications. IEEE, 2012. 1–8. Web.), hereto referred as Bhuiyan, and further in view of Eleuteri et al. (Eleuteri, Antonio et al. “Individualised Variable-Interval Risk-Based Screening for Sight-Threatening Diabetic Retinopathy: The Liverpool Risk Calculation Engine.” Diabetologia 60.11 (2017)), hereto referred as Eleuteri.
Regarding claim 1, Sandhu teaches that a computer-implemented method for diagnosing diabetic retinopathy(Sandhu, Abstract: "A computer-assisted diagnostic (CAD) system to diagnose and grade nonproliferative, diabetic retinopathy (NPDR) from optical coherence tomography (OCT) images", outlines a computer-implemented method for diagnosing diabetic retinopathy, including differentiation and grading) comprises: receiving image data including an image of a retina of a subject (Sandhu, Abstract: "All patients underwent a... spectral-domain OCT of a 6 x 6 mm area of the macula in both eyes. These images then were analyzed by a novel CAD system", describing the acquisition of image data, specifically OCT scans of the retina); processing the image data to segment the image of the retina(Sandhu, Abstract: "these images then were analyzed by a novel CAD system that segments the retina into 12 layers", describing processing of image data to segment the retina); said processing includes processing the OCT image data to segment the image of the retina into a plurality of retinal layers (Sandhu, "The CAD System" 11, Page 3156: "A novel noninvasive framework for early diagnosis of DR using OCT images then was developed"; "Methods", 13, Page 3156:....First, 12 distinct retinal layers are segmented using previously described methods", describing segmentation of OCT image data into multiple retinal layers for diagnostic purposes); extracting at least one feature from the segmented image of the retina (Sandhu, "Abstract: "These images then were analyzed by a novel CAD system that... quantifies the reflectivity, curvature, and thickness of each layer", describing the extraction of features from segmented retina layers for analysis); receiving demographic data and clinical data associated with the subject (Sandhu, "The CAD System", 12, Page 3156: "First, a set of 12 normal OCTs (from six males and six females) was used as the template for automated segmentation. Second, OCTs from 200 normal patients, aged 18 to 75, were segmented by four different retinal specialists and used as a gold standard, or "ground truth, "for normal OCT images", and "Methods", 111, Page 3156: "All patients were diagnosed clinically as having no DR or DR based on dilated fundus exam", describing the receipt of demographic data, including patient age, and clinical data, as the segmentation by specialists provides an expert-validated standard for diagnosing retinal conditions); generating, using a machine learning classifier, a diagnosis for the subject based at least in part on the at least one feature, the demographic data, and the clinical data (Sandhu, "The CAD System", 15, Page 3157: "In the final step, after segmenting the 12 retinal layers and extracting the three key features, the CAD system classified normal and DR subjects... we used a form of a neural network called a deep learning network that had the ability to learn these features and fuse them together. To learn characteristics of normal and DR subjects, CDFs were calculated for each feature and fed into the proposed network", in conjunction with the previously established dataset "The CAD System", 12, Page 3156: "First, a set of 12 normal OCTs (from six males and six females) was used as the template for automated segmentation. Second, OCTs from 200 normal patients, aged 18 to 75, were segmented by four different retinal specialists and used as a gold standard, or 'ground truth, 'for normal OCT images", and "Methods", 11, Page 3156: "All patients were diagnosed clinically as having no DR or DR based on dilated fundus exam", describing the use of a machine learning classifier to generate a diagnosis based on extracted OCT features, demographic data (age and gender), and clinical data (diabetes status, prior diagnosis, and treatment history); and the machine learning classifier is a random forest classifier (Sandhu, “Results”, Table 1, Page 3158: “The proposed deep fusion classification network (DFCN) was compared to other commonly used classifiers, including Random Forest, Support Vector Machine (SVM), and k-Nearest Neighbors (k-NN)”, describing Random Forest as a standard classifier for diabetic retinopathy classification; it would have been prima facie obvious before the effective filing date of the claimed invention to substitute the explicitly-taught Random Forest classifier for the deep learning classifier in Sandhu’s CAD system as a known, conventional alternative performing the same supervised classification function).
Also regarding claim 1, Sandhu teaches that the image data includes optical coherence tomography (OCT) image data (Sandhu, Abstract: "...to diagnose and grade nonproliferative diabetic retinopathy (NPDR) from optical coherence tomography (OCT) images", describing OCT as a relevant imaging modality for DR analysis), but does not teach that the image data includes optical coherence tomography angiography (OCTA) image data. Sandhu recognizes the need to combine OCT and OCTA data to improve the accuracy and robustness of DR classification systems (Sandhu, Discussion, 16, Page 3159). However, Sandhu does not provide a working example that integrates both OCT and OCTA data for diagnosis and grading.
Choi, however, teaches the acquisition and use of both OCT and OCTA image data together for visualizing and analyzing retinal and choriocapillaris microvascular alterations in diabetic patients (Choi, Abstract, 'Introduction', 12, p. 2-3: "since OCTA is an extension of standard OCT imaging, structural information is obtained at the same time, and is intrinsically co-registered to angiographic information". 'Nonproliferative diabetic retinopathy', 12, p. 6). Choi specifically describes OCTA imaging and analysis techniques applied to the study of diabetic retinopathy combined with OCT image data, thus filling the gap left by Sandhu.
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 Sandhu in view of Choi to use both OCT and OCTA image data for diagnosing and grading diabetic retinopathy. The integration of OCTA data with OCT imaging would be possible as both imaging modalities can be acquired together, and Sandhu explicitly motivates combining the two for improved diagnostic accuracy (Sandhu, "Discussion", 16, Page 3159). The benefit of the combination would be improved diagnostic accuracy and more robust grading of diabetic retinopathy by integrating structural and vascular imaging data (Choi, Introduction).
Also regarding claim 1, the modified Sandhu does not fully teach that said processing includes processing the OCTA image data to segment a vasculature of the image of the retina. Rather, it discusses the use of OCT angiography (OCTA) to assess microvascular changes in diabetic retinopathy, particularly through the analysis of capillary density in different plexuses (Sandhu, “Discussion”, 16, Page 3159). However, Sandhu does not explicitly disclose segmenting vasculature from OCTA image data.
Huang, who investigates choroidal neovascularization (CNV) and retinal vasculature using OCT angiography, describes OCTA processing for generating en face angiograms to quantify retinal vasculature, including processing and segmentation techniques to extract blood vessel data (Huang, ¶[0028]: “Once generated, the en face angiogram image may be used to quantify various features of the retinal vasculature”, describing vessel-level extraction from OCTA images). This disclosure directly relates to processing OCTA image data to segment a vasculature of the image of the retina, filling the gap in Sandhu’s disclosure.
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 the Sandhu in view of Huang to process the OCTA image data to segment a vasculature of the retina as recited, because Sandhu already teaches the desirability of using OCTA to analyze microvascular changes, and Huang provides explicit OCTA vasculature segmentation techniques. The benefit of this modification would be improved diabetic retinopathy diagnosis via direct segmentation of retinal vasculature from OCTA images, enabling detailed characterization of capillary density and vascular alterations as taught by Huang.
Also regarding claim 1, the modified Sandhu does not teach that said extracting includes extracting from the segmented vasculature at least one of bifurcation points and crossover points.
Bhuiyan, however, teaches detecting and classifying retinal vascular bifurcation, branch, and crossover points from segmented vessel images. In particular, Bhuiyan describes that “identifying the vascular bifurcation, branch and crossover points in the retinal image is helpful for predicting many cardiovascular diseases” and proposes “a novel method to detect and classify the vascular bifurcation, branch and crossover points (landmarks) based on the vessel geometrical features. We utilize the vessel’s centerline and width information to detect and classify these landmarks, which can be used for image matching in medical diagnosis…” (Bhuiyan, Abstract); further, the method “segment[s] the blood vessels” and then performs “vessel segmentation, skeletonization and vessel geometrical features in the junction area” to detect these landmarks (Bhuiyan, p. 2-3; Sec. 2). This disclosure directly relates to processing a segmented retinal vasculature image to extract vascular bifurcation and crossover points as quantitative features.
It would have been prima facie obvious before the effective filing date of the claimed invention to modify the combined Sandhu, Choi, and Huang system in view of Bhuiyan so that said extracting includes extracting from the segmented vasculature at least one of bifurcation points and crossover points as recited. Sandhu, Choi, and Huang already teach using OCT and OCTA imaging to obtain segmented retinal vasculature for diabetic retinopathy analysis, while Bhuiyan teaches that bifurcation and crossover points are important retinal vascular landmarks whose locations can be automatically detected from segmented vessel images using vessel centerlines and geometrical features. Because both sets of references operate on segmented images of the retinal vasculature for medical analysis, one of ordinary skill in the art would have found it obvious to apply Bhuiyan’s bifurcation and crossover detection techniques to the segmented OCTA vasculature in Sandhu/Choi/Huang and to use the resulting bifurcation and crossover points as additional extracted vascular features for classification, thereby improving characterization of retinal vascular architecture.
Also regarding claim 1, the modified Sandhu does not teach that the clinical data includes at least one of HbA1C and dyslipidemia.
Eleuteri teaches a predictive model for diabetic retinopathy that incorporates retinal screening information and clinical risk factors as model inputs, including HbA1C and lipid-related parameters. Specifically, Eleuteri teaches that “covariates with sufficient predictive value were: duration of known disease, HbA1c, age, systolic BP and total cholesterol” (Eleuteri, Abstract), demonstrating that HbA1C and total cholesterol are used as inputs to a predictive model for diabetic retinopathy. Eleuteri further discloses that “values of clinical risk factors prior and nearest to the screen episode date were used” (Eleuteri, p. 2175, 'Methods'), demonstrating that such clinical variables are obtained and used as inputs to the predictive model. Total cholesterol, as a lipid-related clinical parameter, represents a lipid measurement corresponding to dyslipidemia. Total cholesterol is a primary lipid measurement used in the clinical assessment of dyslipidemia, and one of ordinary skill in the art would understand that inclusion of total cholesterol as a model input corresponds to the use of dyslipidemia-related clinical data.
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 Sandhu in view of Eleuteri to include HbA1C and dyslipidemia as part of the clinical data received and used by the machine learning classifier to generate a diabetic retinopathy diagnosis. Eleuteri demonstrates that HbA1C and lipid-related clinical parameters such as total cholesterol are used as inputs to a predictive model for diabetic retinopathy, and thus one of ordinary skill in the art would have incorporated such established clinical variables into the diagnostic framework of Sandhu. HbA1C is a well-established clinical biomarker of glycemic control and a recognized risk factor for diabetic retinopathy, and thus the inclusion of HbA1C alone satisfies the claimed limitation of “at least one of HbA1C and dyslipidemia.” Furthermore, Eleuteri teaches that “retinopathy grade informed baseline risk” (Eleuteri, Abstract), demonstrating that retinal condition derived from imaging is used in combination with clinical variables within the predictive model, thereby reinforcing the compatibility of Eleuteri’s model with Sandhu’s imaging-based classifier framework. Incorporating such clinical inputs alongside imaging-derived features in a predictive or classification model represents the predictable use of known elements according to their established functions. One of ordinary skill in the art would have been motivated to combine these teachings to improve diagnostic accuracy and provide a more comprehensive evaluation of the subject’s condition, and would have had a reasonable expectation of success in implementing this combination.
Regarding claim 2, the modified Sandhu teaches that the diagnosis is one of normal and diabetic retinopathy (Sandhu, "The CAD System", ¶5, Page 3157: "In the final step, after segmenting the 12 retinal layers and extracting the three key features, the CAD system classified normal and DR subjects...", describing the classification (i.e. diagnosis) of the subject as normal or having diabetic retinopathy (DR)).
Regarding claim 3, the modified Sandhu teaches that the diagnosis is one of normal, mild nonproliferative diabetic retinopathy, moderate nonproliferative diabetic retinopathy, severe nonproliferative diabetic retinopathy, and proliferative diabetic retinopathy (Sandhu, ‘The CAD System’, ¶5, Page 3157: “the CAD system classified normal and DR subjects”; ‘The CAD System’, ¶1, Page 3156: “used to classify the test subject as normal, or having subclinical or mild/moderate DR”, discloses a system for classifying diabetic retinopathy using a deep fusion classification network that distinguishes between normal and mild or moderate NPDR cases).
Regarding claim 4, the modified Sandhu teaches that the diagnosis is one of normal, mild nonproliferative diabetic retinopathy, and moderate nonproliferative diabetic retinopathy (Sandhu, ‘The CAD System’, ¶5, Page 3157: “the CAD system classified normal and DR subjects”; ‘The CAD System’, ¶1, Page 3156: “used to classify the test subject as normal, or having subclinical or mild/moderate DR”, discloses a system for classifying diabetic retinopathy using a deep fusion classification network that distinguishes between normal and mild or moderate NPDR cases).
Regarding claim 7, the modified Sandhu teaches that the at least one feature further includes at least one of retinal layer thickness, reflectivity, and curvature (Sandhu, "The CAD System", ¶5, Page 3157: "Second, three global features are measured based on curvature, reflectivity, and thickness of the segmented retinal layers", describing the extraction of retinal layer thickness, reflectivity, and curvature as features for classification).
Regarding claim 9, the modified Sandhu does not teach that the at least one feature further includes at least one of a distance map of a foveal avascular zone, blood vessel density, and blood vessel caliber. Sandhu discusses the importance of analyzing microvascular changes in diabetic retinopathy using OCT angiography (OCTA) and notes that factors such as blood vessel density and the foveal avascular zone (FAZ) are correlated with DR severity (Sandhu, "Discussion", ¶6, Page 3159), but does not teach or extract features such as, a distance map of a foveal avascular zone, blood vessel density, or blood vessel caliber. Choi teaches the extraction and assessment of such OCTA features by visualizing and analyzing vascular abnormalities including blood vessel density, blood vessel caliber, and foveal avascular zone enlargement (Choi, Abstract; 'Nonproliferative diabetic retinopathy', ¶2, p. 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 modified the combined Sandhu, Choi, Huang, and Bhuiyan in view of Choi to use least one feature selected from a distance map of a foveal avascular zone, blood vessel density, and blood vessel caliber from OCTA image data. The combination would be possible because both references discuss the importance of microvascular analysis and feature extraction for improved diabetic retinopathy classification, and the benefit would be enhanced diagnostic accuracy and earlier detection by utilizing additional vascular biomarkers (Sandhu, "Discussion", 16, Page 3159; Choi, Abstract; Choi, Discussion, ¶8, p. 8–9).
Regarding claim 10, the modified Sandhu teaches that the demographic data includes at least one of sex and age (Sandhu, "The CAD System", ¶2, Page 3156: "First, a set of 12 normal OCTs (from six males and six females) was used as the template for automated segmentation. Second, OCTs from 200 normal patients, aged 18 to 75, were segmented by four different retinal specialists and used as a gold standard, or ‘ground truth,’ for normal OCT images", describing the receipt of demographic data, including patient sex and age, as part of the dataset used in the analysis).
Regarding claim 13, the modified Sandhu teaches that the classifier is a two-stage classifier (Sandhu, "The CAD System", Page 3157: "To build the classification model, a deep neural network with two stages of autoencoders was used. The first stage consisted of several deep networks built with the encoders for each input feature, one autoencoder for each of the three features per each segmented layer for a total of 36 (12 × 3 = 36). In the second stage, detailed classification of subjects with DR was performed to determine the grade of DR using the deep fusion classification network", describing a two-stage classification approach where the first stage extracts feature representations via autoencoders, and the second stage applies the deep fusion classification network to categorize the level of diabetic retinopathy).
Regarding claim 14, the modified Sandhu teaches that the two-stage classifier includes a first stage which generates the diagnosis, the diagnosis being one of normal or diabetic retinopathy (Sandhu, "Results", ¶3, Page 3157: "For the first stage (normal vs. DR), the proposed CAD system showed a total diagnostic accuracy of 93.8% (150/160 subjects, describing a first-stage classification where the first stage distinguishes between normal and diabetic retinopathy); and a second stage which, if the first stage diagnoses diabetic retinopathy, differentiates between grades of diabetic retinopathy (Sandhu, "The CAD System", Page 3157: "In the second stage, detailed classification of subjects with DR was performed to determine the grade of DR using the deep fusion classification network", describing a second-stage classification where it grades the diabetic retinopathy).
Regarding claim 21, the modified Sandhu teaches that the machine learning classifier is trained using (i) OCT image data of retina from individuals with different grades of diabetic retinopathy and from individuals without a diagnosis of diabetic retinopathy (Sandhu, "The CAD System", 15, Page 3157: “a set of 12 normal OCTs (from six males and six females) was used as the template for automated segmentation. Second, OCTs from 200 normal patients, aged 18 to 75, were segmented... All patients were diagnosed clinically as having no DR or DR based on dilated fundus exam”; “the CAD system classified normal and DR subjects... To learn characteristics of normal and DR subjects, CDFs were calculated for each feature and fed into the proposed network”, showing the classifier was trained on OCT image data from both DR and non-DR individuals).
Also regarding claim 21, Sandhu does not teach (ii) OCTA image data of retina from individuals with different grades of diabetic retinopathy and from individuals without a diagnosis of diabetic retinopathy. While Sandhu discusses the importance of OCTA in the analysis of microvascular changes and suggests future integration of OCT and OCTA data (Sandhu, "Discussion", 16, Page 3159), Sandhu does not disclose training the classifier with OCTA image data or with a dataset that includes OCTA images from DR and non-DR patients. Choi, however, teaches the acquisition and analysis of OCTA image data from individuals with different grades of diabetic retinopathy and from individuals without a diagnosis of diabetic retinopathy (Choi, 'Methods', ¶2, p. 3). Choi’s study included three distinct clinical subject groups: (1) normal controls, (2) diabetic patients without retinopathy, and (3) diabetic patients with nonproliferative or proliferative diabetic retinopathy, showing that OCTA images were acquired and analyzed from different grades of disease and from normal subjects (Choi, 'Methods', ¶2, p. 3). Choi compares the retinal vasculature of these groups using OCTA, describing the observed microvascular abnormalities in both normal and diabetic retinopathy subjects, but does not itself perform automated grading or classification of the images. Choi also states, “OCTA can be rapidly acquired and ultimately can be automatically graded without reader intervention” (Choi, 'Discussion', p. 8–9), indicating that automated grading and classification using such datasets is anticipated as a logical next step. A person of ordinary skill in the art would recognize that Choi’s acquisition of OCTA data for well-defined clinical grades provides labeled datasets suitable for use in supervised machine learning approaches, such as those employed by Sandhu.
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 the Sandhu in view of Choi to train the machine learning classifier using OCTA image data from individuals with different grades of diabetic retinopathy and from individuals without a diagnosis of diabetic retinopathy. The combination would be possible because Sandhu and Choi both discuss acquiring imaging data from both patient and control populations for the analysis of diabetic retinopathy, and Choi demonstrates the feasibility and clinical value of including OCTA data from both groups. The benefit would be a more robust and accurate machine learning classifier for diabetic retinopathy diagnosis and grading by integrating both structural and vascular imaging data (Sandhu, 'Discussion, ¶6, Page 3159; Choi, Abstract; 'Introduction', ¶2, p. 2–3).
Also regarding claim 21, Sandhu partially teaches (iii) demographic data from the individuals in (i) and (ii), and (iv) clinical data from the individuals in (i) and (ii). Specifically, Sandhu teaches including demographic and clinical data from individuals with and without diabetic retinopathy for OCT images (Sandhu, 'The CAD System', ¶2, Page 3156). However, Sandhu does not disclose collecting demographic or clinical data for individuals who received OCTA imaging, since OCTA was not performed. Choi teaches the acquisition, via a specialist, and use of clinical data, complete ophthalmic examination, for all individuals undergoing OCTA imaging, including both diabetic patients and normal subjects (Choi, Methods, ¶2, p. 3). Choi also reports the ages of the subjects as demographic data (Choi, 'Results', p. 5-6), but does not explicitly use age as an input for assessment or classification. While the collection of demographic data in Choi is limited to age, a person of ordinary skill in the art would recognize that in a combined system using both OCT and OCTA imaging modalities, the demographic and clinical data would naturally be included for all subjects, since both imaging types would typically be acquired from the same patient cohorts and the machine learning classifier in Sandhu is already designed to incorporate such data. 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 the Sandhu in view of Choi to collect and use demographic and clinical data for all individuals undergoing OCTA imaging—both those with and without diabetic retinopathy—when training the machine learning classifier. The combination would be possible because both Sandhu and Choi disclose the clinical value of using comprehensive patient datasets, and the classifier is already structured to accept such data. The benefit would be a more accurate, clinically relevant classifier for diabetic retinopathy diagnosis, reflecting the true spectrum of patient presentations (Sandhu, "The CAD System", 12, Page 3156; Choi, Methods, p. 3–4).
Claims 15, 17, and 19-20 are rejected under 35 U.S.C. 103 as being unpatentable over Sandhu et al. (Sandhu, Harpal Singh et al. “Automated Diagnosis and Grading of Diabetic Retinopathy Using Optical Coherence Tomography.” Investigative ophthalmology & visual science 59.7 (2018): 3155–3160. Web.), hereto referred as Sandhu, and further in view of Choi et al. (Choi, WooJhon et al. ”Ultrahigh Speed OCT Angiography of Retinal and Choriocapillaris Alterations in Diabetic Patients with and without Retinopathy Using Swept Source Optical Coherence Tomography”. Retina 37(1):p 11-21, January 2017), hereto referred as Choi, and further in view of Bhuiya et al. (Bhuiyan et al. “Detection and Classification of Bifurcation and Branch Points on Retinal Vascular Network.” 2012 International Conference on Digital Image Computing: Techniques and Applications. IEEE, 2012. 1–8. Web.), hereto referred as Bhuiyan, and further in view of Eleuteri et al. (Eleuteri, Antonio et al. “Individualised Variable-Interval Risk-Based Screening for Sight-Threatening Diabetic Retinopathy: The Liverpool Risk Calculation Engine.” Diabetologia 60.11 (2017)), hereto referred as Eleuteri.
Regarding claim 15, Sandhu teaches that a computer-implemented method for classifying a retina (Sandhu, Abstract, Page 3155: "We determine the feasibility and accuracy of a computer-assisted diagnostic (CAD) system to diagnose and grade nonproliferative diabetic retinopathy", describing a computer-implemented method for classifying a retina), comprises: processing OCT image data including an OCT image of a subject retina to segment the OCT image of the subject retina (Sandhu, Abstract: "these images then were analyzed by a novel CAD system that segments the retina into 12 layers", describing processing of image data to segment the retina); extracting at least one OCT feature from the segmented OCT image of the subject retina (Sandhu, "Abstract: "These images then were analyzed by a novel CAD system that... quantifies the reflectivity, curvature, and thickness of each layer", describing the extraction of features from segmented retina layers for analysis); and receiving demographic data and clinical data associated with the subject retina (Sandhu, "The CAD System", ¶2, Page 3156: "First, a set of 12 normal OCTs (from six males and six females) was used as the template for automated segmentation. Second, OCTs from 200 normal patients, aged 18 to 75, were segmented by four different retinal specialists and used as a gold standard, or ‘‘ground truth,’’ for normal OCT images", and "Methods", ¶1, Page 3156: "All patients were diagnosed clinically as having no DR or DR based on dilated fundus exam", describing the receipt of demographic data, including patient age, and clinical data, as the segmentation by specialists provides an expert-validated standard for diagnosing retinal conditions).
Also regarding claim 15, Sandhu does not teach processing OCTA image data including an OCTA image of the subject retina to segment the OCTA image of the subject retina; extracting at least one OCTA feature from the segmented OCTA image of the subject retina. Rather, Sandhu contemplates combining OCT and OCTA data but does not expressly perform such actions (Sandhu, 'Discussion', 116, p. 3159). Rather, Sandhu teaches the use of OCT image data and segmentation and feature extraction for classifying diabetic retinopathy, as shown above, but does not teach processing OCTA image data including an OCTA image of the subject retina to segment the OCTA image or to extract features from the segmented OCTA image. Sandhu discusses the potential value of combining OCT with OCTA to improve diagnostic accuracy but does not describe a system that implements segmentation and feature extraction from OCTA images (Sandhu, "Discussion", 16, Page 3159). Choi, however, teaches the acquisition (Choi, Abstract, 'Introduction', ¶2, p. 2-3, 'Nonproliferative diabetic retinopathy', ¶2, p. 6), segmentation (Choi, 'Methods', ¶5-6, p. 4), and feature extraction (Choi, Abstract, 'Nonproliferative diabetic retinopathy', ¶2, p. 6, 'Discussion', ¶8, p. 8-9) from OCTA images for analyzing diabetic retinopathy. Choi teaches segmentation of OCTA images by generating three-dimensional OCTA volumes from hundreds of spatially resolved B-scan locations and by forming en face images projected through individual vascular plexuses or slabs of specific retinal layers (Choi, Methods, ¶5-6, p. 4: “OCTA images were generated by calculating a decorrelation signal on a pixel-by-pixel basis... performed at all 500 B-scan locations in order to obtain a three-dimensional OCTA volume”; “the OCTA volume was either mean projected through individual vascular plexuses... En face OCTA CC slabs were generated by selecting a single en face plane, ~4.4 µm in thickness, immediately below Bruch’s membrane”). Choi further teaches feature extraction from segmented OCTA images by identifying and assessing microvascular abnormalities such as “clustered capillaries, dilated capillary segments, tortuous capillaries, regions of capillary dropout, reduced capillary density, abnormal capillary loops, and foveal avascular zone enlargement” (Choi, Abstract; Choi, Nonproliferative diabetic retinopathy, ¶2, p. 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 modified Sandhu in view of Choi to process OCTA image data including an OCTA image of the subject retina to segment the OCTA image and extract at least one OCTA feature. The integration of OCTA segmentation and feature extraction would be possible as both imaging modalities and analytic steps are used together in Choi, and Sandhu explicitly motivates combining them for improved diagnostic accuracy (Sandhu, "Discussion", 16, Page 3159). The benefit of the combination would be improved and earlier detection and classification of diabetic retinopathy by utilizing both structural and vascular imaging features (Choi, Introduction).
Also regarding claim 15, the modified Sandhu does not teach that said extracting includes extracting from the segmented vasculature at least one of bifurcation points and crossover points.
Bhuiyan, however, teaches detecting and classifying retinal vascular bifurcation, branch, and crossover points from segmented vessel images. In particular, Bhuiyan describes that “identifying the vascular bifurcation, branch and crossover points in the retinal image is helpful for predicting many cardiovascular diseases” and proposes “a novel method to detect and classify the vascular bifurcation, branch and crossover points (landmarks) based on the vessel geometrical features. We utilize the vessel’s centerline and width information to detect and classify these landmarks, which can be used for image matching in medical diagnosis…” (Bhuiyan, Abstract); further, the method “segment[s] the blood vessels” and then performs “vessel segmentation, skeletonization and vessel geometrical features in the junction area” to detect these landmarks (Bhuiyan, p. 2-3; Sec. 2). This disclosure directly relates to processing a segmented retinal vasculature image to extract vascular bifurcation and crossover points as quantitative features.
It would have been prima facie obvious before the effective filing date of the claimed invention to have further modified the modified the Sandhu in view of Bhuiyan so that said extracting includes extracting from the segmented vasculature at least one of bifurcation points and crossover points as recited. Sandhu and Choi already teach using OCT and OCTA imaging to obtain segmented retinal vasculature for diabetic retinopathy analysis, while Bhuiyan teaches that bifurcation and crossover points are important retinal vascular landmarks whose locations can be automatically detected from segmented vessel images using vessel centerlines and geometrical features. Because both sets of references operate on segmented images of the retinal vasculature for medical analysis, one of ordinary skill in the art would have found it obvious to apply Bhuiyan’s bifurcation and crossover detection techniques to the segmented OCTA vasculature in Sandhu/Choi and to use the resulting bifurcation and crossover points as additional extracted vascular features for classification, thereby improving characterization of retinal vascular architecture.
Also regarding claim 15, the modified Sandhu teaches classifying, using a machine learning classifier, the subject retina as normal or indicative of diabetic retinopathy based at least in part on the at least one OCT feature, the demographic data, and the clinical data (Sandhu, "The CAD System", 15, Page 3157: "In the final step, after segmenting the 12 retinal layers and extracting the three key features, the CAD system classified normal and DR subjects... we used a form of a neural network called a deep learning network that had the ability to learn these features and fuse them together. To learn characteristics of normal and DR subjects, CDFs were calculated for each feature and fed into the proposed network", in conjunction with the previously established dataset "The CAD System", 12, Page 3156: "First, a set of 12 normal OCTs (from six males and six females) was used as the template for automated segmentation. Second, OCTs from 200 normal patients, aged 18 to 75, were segmented by four different retinal specialists and used as a gold standard, or 'ground truth,' for normal OCT images", and "Methods", 11, Page 3156: "All patients were diagnosed clinically as having no DR or DR based on dilated fundus exam", describing the use of a machine learning classifier to generate a diagnosis based on extracted OCT features, demographic data (age and gender), and clinical data (diabetes status, prior diagnosis, and treatment history)), but does not teach at least one OCTA feature being used as part of the classifying. Rather, Sandhu teaches classifying the retina using a machine learning classifier based on extracted OCT features, demographic, and clinical data, but does not teach using at least one OCTA feature as part of the classification process (Sandhu, 'The CAD System', 15, Page 3157). Choi, however, teaches the extraction and assessment of OCTA features that reflect microvascular abnormalities in diabetic retinopathy (Choi, Abstract; Choi, Nonproliferative diabetic retinopathy, ¶2, p. 6), and discusses the future potential for these features to be used in automated, algorithmic, or quantitative classification, stating that "OCTA can be rapidly acquired and ultimately can be automatically graded without reader intervention" and emphasizing the need for image processing algorithms for automatic quantitative assessment (Choi, Discussion, ¶8, p. 8–9). Sandhu’s machine learning framework is expressly feature-based and is designed to process and integrate multiple types of extracted data, “a neural network that had the ability to learn these features and fuse them together” (Sandhu, 'The CAD System', ¶5, Page 3157). While Sandhu specifically discloses fusing OCT-derived, demographic, and clinical features, Sandhu also explicitly motivates combining OCT information with OCTA data for improved diagnostic accuracy (Sandhu, 'Discussion', 16, Page 3159). A person of ordinary skill in the art would recognize that such a classifier could readily accept additional relevant imaging features, such as those derived from OCTA images as taught by Choi, by simply incorporating these as new inputs into the classification model. Both references highlight the importance of integrating structural (OCT) and vascular (OCTA) biomarkers for improved diabetic retinopathy diagnosis and grading (Sandhu, "Discussion", 16, Page 3159; Choi, Introduction, p. 2-3). 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 modified the combined Sandhu, Choi, and Bhuiyan in view of Choi to use at least one OCTA feature as part of the machine learning-based classification of diabetic retinopathy. The combination would be possible because both references discuss using imaging features (structural and vascular) to improve diagnostic accuracy, and Sandhu’s machine learning framework is readily adaptable to accept additional inputs such as OCTA-derived features. The benefit of the combination would be improved accuracy, robustness, and clinical utility of the automated diabetic retinopathy classification system by integrating complementary structural and vascular biomarkers (Sandhu, "Discussion", 16, Page 3159; Choi, Introduction, p. 2-3; Choi, Discussion, ¶8, p. 8–9).
Also regarding claim 15, the modified Sandhu does not teach that the clinical data includes at least one of HbA1C and dyslipidemia.
Eleuteri teaches a predictive model for diabetic retinopathy that incorporates retinal screening information and clinical risk factors as model inputs, including HbA1C and lipid-related parameters. Specifically, Eleuteri teaches that “covariates with sufficient predictive value were: duration of known disease, HbA1c, age, systolic BP and total cholesterol” (Eleuteri, Abstract), demonstrating that HbA1C and total cholesterol are used as inputs to a predictive model for diabetic retinopathy. Eleuteri further discloses that “values of clinical risk factors prior and nearest to the screen episode date were used” (Eleuteri, p. 2175, 'Methods'), demonstrating that such clinical variables are obtained and used as inputs to the predictive model. Total cholesterol, as a lipid-related clinical parameter, represents a lipid measurement corresponding to dyslipidemia. Total cholesterol is a primary lipid measurement used in the clinical assessment of dyslipidemia, and one of ordinary skill in the art would understand that inclusion of total cholesterol as a model input corresponds to the use of dyslipidemia-related clinical data.
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 Sandhu in view of Eleuteri to include HbA1C and dyslipidemia as part of the clinical data received and used by the machine learning classifier to generate a diabetic retinopathy diagnosis. Eleuteri demonstrates that HbA1C and lipid-related clinical parameters such as total cholesterol are used as inputs to a predictive model for diabetic retinopathy, and thus one of ordinary skill in the art would have incorporated such established clinical variables into the diagnostic framework of Sandhu. HbA1C is a well-established clinical biomarker of glycemic control and a recognized risk factor for diabetic retinopathy, and thus the inclusion of HbA1C alone satisfies the claimed limitation of “at least one of HbA1C and dyslipidemia.” Furthermore, Eleuteri teaches that “retinopathy grade informed baseline risk” (Eleuteri, Abstract), demonstrating that retinal condition derived from imaging is used in combination with clinical variables within the predictive model, thereby reinforcing the compatibility of Eleuteri’s model with Sandhu’s imaging-based classifier framework. Incorporating such clinical inputs alongside imaging-derived features in a predictive or classification model represents the predictable use of known elements according to their established functions. One of ordinary skill in the art would have been motivated to combine these teachings to improve diagnostic accuracy and provide a more comprehensive evaluation of the subject’s condition, and would have had a reasonable expectation of success in implementing this combination.
Regarding claim 17, the modified Sandhu teaches processing the OCT image data to segment the OCT image of the subject retina includes processing the OCT image data to segment the OCT image of the subject retina into a plurality of retinal layers. (Sandhu, "The CAD System" ¶1, Page 3156: "A novel noninvasive framework for early diagnosis of DR using OCT images then was developed"; "Methods", ¶3, Page 3156: ... First, 12 distinct retinal layers are segmented using previously described methods", describing segmentation of OCT image data into multiple retinal layers for diagnostic purposes).
Regarding claim 19, Sandhu teaches that a non-transitory computer-readable storage medium having computer program instructions stored thereon that, when executed by a processor, cause the processor to perform the following instructions(Sandhu, Abstract: "We determine the feasibility and accuracy of a computer-assisted diagnostic (CAD) system to diagnose and grade nonproliferative diabetic retinopathy", inherently describing a computer-based system performing diagnostic analysis, which necessitates the storage of imaging data, extracted features, and trained machine learning models. Such a system would require non-transitory computer-readable storage to persistently retain diagnostic data, enable model training, and support real-time classification, ensuring continuous and reliable operation): receiving at least one OCT feature extracted from OCT image data of a subject retina (Sandhu, Abstract: "system to diagnose and grade nonproliferative diabetic retinopathy (NPDR) from optical coherence tomography (OCT) images... These images then were analyzed by a novel CAD system that... quantifies the reflectivity, curvature, and thickness of each layer", describing the extraction of features from segmented retina layers for analysis from OTC image data); and receiving demographic data and clinical data associated with the subject retina (Sandhu, "The CAD System", ¶2, Page 3156: "First, a set of 12 normal OCTs (from six males and six females) was used as the template for automated segmentation. Second, OCTs from 200 normal patients, aged 18 to 75, were segmented by four different retinal specialists and used as a gold standard, or ‘‘ground truth,’’ for normal OCT images", and "Methods", ¶1, Page 3156: "All patients were diagnosed clinically as having no DR or DR based on dilated fundus exam", describing the receipt of demographic data, including patient age, and clinical data, as the segmentation by specialists provides an expert-validated standard for diagnosing retinal conditions).
Also regarding claim 19, Sandhu does not directly teach receiving at least one OCTA feature extracted from OCTA image data of the subject retina. As shown above, Sandhu already receives features related to OCT. Additionally, Sandhu discusses the importance of analyzing microvascular changes in diabetic retinopathy using OCT angiography (OCTA) and notes that factors such as blood vessel density and the foveal avascular zone (FAZ) are correlated with DR severity (Sandhu, 'Discussion', ¶6, Page 3159), but does not teach receiving OCTA features extracted from OCTA image data of the subject retina. Choi teaches the extraction and assessment of such OCTA features by visualizing and analyzing vascular abnormalities including blood vessel density, blood vessel caliber, and foveal avascular zone enlargement (Choi, Abstract; 'Nonproliferative diabetic retinopathy', ¶2, p. 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 modified Sandhu in view of Choi to receive at least one OCTA feature extracted from OCTA image data. The combination would be possible because Sandhu’s feature-based machine learning framework is designed to process and integrate multiple types of extracted features, and a person of ordinary skill would recognize that relevant additional imaging features such as those from Choi’s OCTA analysis could be readily incorporated as further classifier inputs. Both references discuss the importance of microvascular analysis and feature extraction for improved diabetic retinopathy classification, and the benefit would be enhanced diagnostic accuracy and earlier detection by utilizing additional vascular biomarkers (Sandhu, "The CAD System", 15, Page 3157; Sandhu, "Discussion", 16, Page 3159; Choi, Abstract; Choi, Discussion, ¶8, p. 8–9).
Also regarding claim 19, the modified Sandhu does not teach that said extracting includes extracting from the segmented vasculature at least one of bifurcation points and crossover points.
Bhuiyan, however, teaches detecting and classifying retinal vascular bifurcation, branch, and crossover points from segmented vessel images. In particular, Bhuiyan describes that “identifying the vascular bifurcation, branch and crossover points in the retinal image is helpful for predicting many cardiovascular diseases” and proposes “a novel method to detect and classify the vascular bifurcation, branch and crossover points (landmarks) based on the vessel geometrical features. We utilize the vessel’s centerline and width information to detect and classify these landmarks, which can be used for image matching in medical diagnosis…” (Bhuiyan, Abstract); further, the method “segment[s] the blood vessels” and then performs “vessel segmentation, skeletonization and vessel geometrical features in the junction area” to detect these landmarks (Bhuiyan, p. 2-3; Sec. 2). This disclosure directly relates to processing a segmented retinal vasculature image to extract vascular bifurcation and crossover points as quantitative features.
It would have been prima facie obvious before the effective filing date of the claimed invention to have further modified the modified the Sandhu in view of Bhuiyan so that said extracting includes extracting from the segmented vasculature at least one of bifurcation points and crossover points as recited. Sandhu and Choi already teach using OCT and OCTA imaging to obtain segmented retinal vasculature for diabetic retinopathy analysis, while Bhuiyan teaches that bifurcation and crossover points are important retinal vascular landmarks whose locations can be automatically detected from segmented vessel images using vessel centerlines and geometrical features. Because both sets of references operate on segmented images of the retinal vasculature for medical analysis, one of ordinary skill in the art would have found it obvious to apply Bhuiyan’s bifurcation and crossover detection techniques to the segmented OCTA vasculature in Sandhu/Choi and to use the resulting bifurcation and crossover points as additional extracted vascular features for classification, thereby improving characterization of retinal vascular architecture.
Also regarding claim 19, the modified Sandhu teaches classifying, using a machine learning random forest classifier (Sandhu, “Results”, Table 1, Page 3158: “The proposed deep fusion classification network (DFCN) was compared to other commonly used classifiers, including Random Forest, Support Vector Machine (SVM), and k-Nearest Neighbors (k-NN)”, describing Random Forest as a standard classifier for diabetic retinopathy classification; it would have been prima facie obvious before the effective filing date of the claimed invention to substitute the explicitly-taught Random Forest classifier for the deep learning classifier in Sandhu’s CAD system as a known, conventional alternative performing the same supervised classification function), the subject retina as normal or indicative of diabetic retinopathy based at least in part on the at least one OCT feature, the demographic data, and the clinical data (Sandhu, "The CAD System", 15, Page 3157: "In the final step, after segmenting the 12 retinal layers and extracting the three key features, the CAD system classified normal and DR subjects... we used a form of a neural network called a deep learning network that had the ability to learn these features and fuse them together. To learn characteristics of normal and DR subjects, CDFs were calculated for each feature and fed into the proposed network", in conjunction with the previously established dataset "The CAD System", 12, Page 3156: "First, a set of 12 normal OCTs (from six males and six females) was used as the template for automated segmentation. Second, OCTs from 200 normal patients, aged 18 to 75, were segmented by four different retinal specialists and used as a gold standard, or 'ground truth,' for normal OCT images", and "Methods", 11, Page 3156: "All patients were diagnosed clinically as having no DR or DR based on dilated fundus exam", describing the use of a machine learning classifier to generate a diagnosis based on extracted OCT features, demographic data (age and gender), and clinical data (diabetes status, prior diagnosis, and treatment history)), but does not teach at least one OCTA feature being used as part of the classifying. Sandhu teaches classifying the retina using a machine learning classifier based on extracted OCT features, demographic, and clinical data, but does not teach using at least one OCTA feature as part of the classification process (Sandhu, 'The CAD System', 15, Page 3157). Choi, however, teaches the extraction and assessment of OCTA features that reflect microvascular abnormalities in diabetic retinopathy (Choi, Abstract; Choi, Nonproliferative diabetic retinopathy, ¶2, p. 6), and discusses the future potential for these features to be used in automated, algorithmic, or quantitative classification, stating that "OCTA can be rapidly acquired and ultimately can be automatically graded without reader intervention" and emphasizing the need for image processing algorithms for automatic quantitative assessment (Choi, Discussion, ¶8, p. 8–9). Sandhu’s machine learning framework is expressly feature-based and is designed to process and integrate multiple types of extracted data, “a neural network that had the ability to learn these features and fuse them together” (Sandhu, 'The CAD System', ¶5, Page 3157). While Sandhu specifically discloses fusing OCT-derived, demographic, and clinical features, Sandhu also explicitly motivates combining OCT information with OCTA data for improved diagnostic accuracy (Sandhu, 'Discussion', 16, Page 3159). A person of ordinary skill in the art would recognize that such a classifier could readily accept additional relevant imaging features, such as those derived from OCTA images as taught by Choi, by simply incorporating these as new inputs into the classification model. Both references highlight the importance of integrating structural (OCT) and vascular (OCTA) biomarkers for improved diabetic retinopathy diagnosis and grading (Sandhu, "Discussion", 16, Page 3159; Choi, Introduction, p. 2-3). 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 the Sandhu in view of Choi to use at least one OCTA feature as part of the machine learning-based classification of diabetic retinopathy. The combination would be possible because both references discuss using imaging features (structural and vascular) to improve diagnostic accuracy, and Sandhu’s machine learning framework is readily adaptable to accept additional inputs such as OCTA-derived features. The benefit of the combination would be improved accuracy, robustness, and clinical utility of the automated diabetic retinopathy classification system by integrating complementary structural and vascular biomarkers (Sandhu, "Discussion", 16, Page 3159; Choi, Introduction, p. 2-3; Choi, Discussion, ¶8, p. 8–9).
Also regarding claim 19, the modified Sandhu does not teach that the clinical data includes at least one of HbA1C and dyslipidemia.
Eleuteri teaches a predictive model for diabetic retinopathy that incorporates retinal screening information and clinical risk factors as model inputs, including HbA1C and lipid-related parameters. Specifically, Eleuteri teaches that “covariates with sufficient predictive value were: duration of known disease, HbA1c, age, systolic BP and total cholesterol” (Eleuteri, Abstract), demonstrating that HbA1C and total cholesterol are used as inputs to a predictive model for diabetic retinopathy. Eleuteri further discloses that “values of clinical risk factors prior and nearest to the screen episode date were used” (Eleuteri, p. 2175, 'Methods'), demonstrating that such clinical variables are obtained and used as inputs to the predictive model. Total cholesterol, as a lipid-related clinical parameter, represents a lipid measurement corresponding to dyslipidemia. Total cholesterol is a primary lipid measurement used in the clinical assessment of dyslipidemia, and one of ordinary skill in the art would understand that inclusion of total cholesterol as a model input corresponds to the use of dyslipidemia-related clinical data.
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 Sandhu in view of Eleuteri to include HbA1C and dyslipidemia as part of the clinical data received and used by the machine learning classifier to generate a diabetic retinopathy diagnosis. Eleuteri demonstrates that HbA1C and lipid-related clinical parameters such as total cholesterol are used as inputs to a predictive model for diabetic retinopathy, and thus one of ordinary skill in the art would have incorporated such established clinical variables into the diagnostic framework of Sandhu. HbA1C is a well-established clinical biomarker of glycemic control and a recognized risk factor for diabetic retinopathy, and thus the inclusion of HbA1C alone satisfies the claimed limitation of “at least one of HbA1C and dyslipidemia.” Furthermore, Eleuteri teaches that “retinopathy grade informed baseline risk” (Eleuteri, Abstract), demonstrating that retinal condition derived from imaging is used in combination with clinical variables within the predictive model, thereby reinforcing the compatibility of Eleuteri’s model with Sandhu’s imaging-based classifier framework. Incorporating such clinical inputs alongside imaging-derived features in a predictive or classification model represents the predictable use of known elements according to their established functions. One of ordinary skill in the art would have been motivated to combine these teachings to improve diagnostic accuracy and provide a more comprehensive evaluation of the subject’s condition, and would have had a reasonable expectation of success in implementing this combination.
Regarding claim 20, the modified Sandhu teaches that the at least one OCT feature extracted from OCT image data of the subject retina is extracted from OCT image data of the subject retina segmented into a plurality of retinal layers (Sandhu, "The CAD System" ¶1, Page 3156: "A novel noninvasive framework for early diagnosis of DR using OCT images then was developed"; "Methods", ¶3, Page 3156: ... First, 12 distinct retinal layers are segmented using previously described methods", describing segmentation of OCT image data into multiple retinal layers for diagnostic purposes), but does not teach that the at least one OCTA feature extracted from the OCTA image data of the subject retina is extracted from the OCTA image data of a segmented vasculature of the subject retina. Rather, Sandhu discusses the use of OCT angiography (OCTA) to assess microvascular changes in diabetic retinopathy, particularly through the analysis of capillary density in different plexuses (Sandhu, "Discussion", 16, Page 3159), but does not explicitly disclose segmenting vasculature from OCTA image data or extracting features specifically from segmented vascular layers. Choi, however, teaches the acquisition (Choi, Abstract; Introduction, ¶2, p. 2–3; Nonproliferative diabetic retinopathy, ¶2, p. 6), segmentation (Choi, Methods, ¶5–6, p. 4), and feature extraction (Choi, Abstract; Nonproliferative diabetic retinopathy, ¶2, p. 6; Discussion, ¶8, p. 8–9) from OCTA images for analyzing diabetic retinopathy. Choi teaches segmentation of OCTA images by generating three-dimensional OCTA volumes from hundreds of spatially resolved B-scan locations and by forming en face images projected through individual vascular plexuses or slabs of specific retinal layers (Choi, Methods, ¶5–6, p. 4: "OCTA images were generated by calculating a decorrelation signal on a pixel-by-pixel basis... performed at all 500 B-scan locations in order to obtain a three-dimensional OCTA volume"; "the OCTA volume was either mean projected through individual vascular plexuses... En face OCTA CC slabs were generated by selecting a single en face plane, ~4.4 µm in thickness, immediately below Bruch’s membrane"). Choi further teaches feature extraction from segmented OCTA images by identifying and assessing microvascular abnormalities such as "clustered capillaries, dilated capillary segments, tortuous capillary segments, regions of capillary dropout, reduced capillary density, abnormal capillary loops, and foveal avascular zone enlargement" (Choi, Abstract; Nonproliferative diabetic retinopathy, ¶2, p. 6). Because Sandhu’s automated diagnostic system extracts and analyzes features from segmented OCT data, a person of ordinary skill in the art would recognize that additional imaging features—such as those resulting from Choi’s segmented vasculature in OCTA—could be incorporated into the same feature extraction and classification pipeline with routine experimentation. Both Sandhu and Choi describe feature extraction from segmented images for diagnostic purposes, and the benefit would be enhanced diagnostic accuracy through detailed vascular assessment (Sandhu, "Discussion", 16, Page 3159; Choi, Methods, ¶5–6, p. 4; Choi, Nonproliferative diabetic retinopathy, ¶2, p. 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 further modified the modified the Sandhu in view of Choi to have the at least one feature extracted from OCTA image data of the subject retina be of a segmented vasculature of the subject retina. This would have the benefit of improving the accuracy of diabetic retinopathy diagnosis by allowing for direct vascular segmentation from OCTA images, providing detailed information on capillary density and retinal blood flow.
Claim 18 is rejected under 35 U.S.C. 103 as being unpatentable over Sandhu et al. (Sandhu, Harpal Singh et al. “Automated Diagnosis and Grading of Diabetic Retinopathy Using Optical Coherence Tomography.” Investigative ophthalmology & visual science 59.7 (2018): 3155–3160. Web.), hereto referred as Sandhu, and further in view of Choi et al. (Choi, WooJhon et al. ”Ultrahigh Speed OCT Angiography of Retinal and Choriocapillaris Alterations in Diabetic Patients with and without Retinopathy Using Swept Source Optical Coherence Tomography”. Retina 37(1):p 11-21, January 2017), hereto referred as Choi, and further in view of Bhuiya et al. (Bhuiyan et al. “Detection and Classification of Bifurcation and Branch Points on Retinal Vascular Network.” 2012 International Conference on Digital Image Computing: Techniques and Applications. IEEE, 2012. 1–8. Web.), hereto referred as Bhuiyan, and further in view of Eleuteri et al. (Eleuteri, Antonio et al. “Individualised Variable-Interval Risk-Based Screening for Sight-Threatening Diabetic Retinopathy: The Liverpool Risk Calculation Engine.” Diabetologia 60.11 (2017)), hereto referred as Eleuteri, and further in view of Huang et al. (US 20160284085 A1), hereto referred as Huang.
The modified Sandhu teaches claim 15 as described above.
Regarding claim 18, the modified Sandhu does not teach that processing the OCTA image data to segment the OCTA image of the subject retina includes processing the OCTA image data to segment a vasculature of the OCTA image of the subject retina. It discusses the use of OCT angiography (OCTA) to assess microvascular changes in diabetic retinopathy, particularly through the analysis of capillary density in different plexuses (Sandhu, "Discussion", ¶6, Page 3159). However, Sandhu does not explicitly disclose segmenting vasculature from OCTA image data. Huang, who investigates choroidal neovasculatization (CNV) in the outer retina using OCT angiography, describes OCTA processing for generating en face angiograms to quantify retinal vasculature, including segmentation techniques to extract blood vessel data (Huang, ¶[0028]). This disclosure directly relates to segmenting vasculature from OCTA image data, filling the gap in Sandhu’s disclosure. It would have been obvious for a person of ordinary skill in the art to incorporate Huang’s OCTA segmentation techniques into Sandhu’s automated diagnostic system. Both references focus on using advanced imaging modalities to analyze diabetic retinopathy, and segmenting vasculature from OCTA would provide additional critical diagnostic data. Given that Sandhu already discusses using OCTA for microvascular analysis, applying Huang’s segmentation method would be a logical and beneficial enhancement. 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 the Sandhu in view of Huang to process the OCTA image data to segment vasculature of the subject retina. This would have the benefit of improving the accuracy of diabetic retinopathy diagnosis by allowing for direct vascular segmentation from OCTA images, providing detailed information on capillary density and retinal blood flow, as noted in Huang, "Once generated, the en face angiogram image may be used to quantify various features of the retinal vasculature" (Huang, ¶[0028]).
Claim 11 is rejected under 35 U.S.C. 103 as being unpatentable over Sandhu et al. (Sandhu, Harpal Singh et al. “Automated Diagnosis and Grading of Diabetic Retinopathy Using Optical Coherence Tomography.” Investigative ophthalmology & visual science 59.7 (2018): 3155–3160. Web.), and further in view of Choi et al. (Choi, WooJhon et al. ”Ultrahigh Speed OCT Angiography of Retinal and Choriocapillaris Alterations in Diabetic Patients with and without Retinopathy Using Swept Source Optical Coherence Tomography”. Retina 37(1):p 11-21, January 2017), hereto referred as Choi, and further in view of Huang et al. (US 20160284085 A1), hereto referred as Huang, and further in view of Bhuiya et al. (Bhuiyan et al. “Detection and Classification of Bifurcation and Branch Points on Retinal Vascular Network.” 2012 International Conference on Digital Image Computing: Techniques and Applications. IEEE, 2012. 1–8. Web.), hereto referred as Bhuiyan, and further in view of Eleuteri et al. (Eleuteri, Antonio et al. “Individualised Variable-Interval Risk-Based Screening for Sight-Threatening Diabetic Retinopathy: The Liverpool Risk Calculation Engine.” Diabetologia 60.11 (2017)), hereto referred as Eleuteri, and further in view of Samec et al. (US 20170007111 A1), hereto referred as Samec.
The modified Sandhu teaches claim 1 as described above.
Regarding claim 11, the modified Sandhu does not fully teach that the clinical data further includes at least one of visual acuity and hypertension. Rather, it discloses a system that processes retinal images and integrates clinical data, such as diabetes duration, in the classification of diabetic retinopathy (Sandhu, "Results", ¶4, Page 3157). Additionally, Sandhu references studies indicating that visual acuity correlates inversely with the size of the foveal avascular zone in diabetics (Sandhu, "Discussion", ¶6, Page 3159). However, Sandhu does not explicitly process or incorporate other clinical data such as hypertension.
Samec, who investigates a system for clinically relevant patient diagnostics/monitoring/therapy, provides a system for ophthalmic diagnosis that includes clinical data as part of its analysis. Specifically, Samec describes performing visual acuity tests as part of an ophthalmic assessment (Samec, ¶[1746]), as well as correlating retinal vessel characteristics with hypertensive retinopathy and other vascular diseases (Samec, ¶[1929]). This demonstrates that hypertensive retinopathy, visual acuity, and related vascular factors can be included in an ophthalmic assessment. It would have been obvious for a person of ordinary skill in the art to incorporate Samec’s method of utilizing clinical data (including visual acuity and vascular health indicators) into Sandhu’s automated DR classification system. Since Sandhu already considers clinical parameters such as diabetes status, extending its classification system to incorporate other well-known ophthalmic indicators of DR severity would be a natural extension. The integration of Samec’s methodology would enhance Sandhu’s system by providing a more comprehensive diagnostic assessment.
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 the combined Sandhu, Choi, Huang, and Bhuiyan in view of Samec to have the clinical data explicitly include at least one of visual acuity and hypertension. This would have the benefit of improving diagnostic accuracy by integrating additional clinical factors such as visual acuity and hypertensive retinopathy indicators, as noted in Samec: "New vessel formation around the optic disc and/or microaneurysms may be indicative of diabetic retinopathy" (Samec, ¶[1929]).
Response to Arguments
Claim Rejections – 35 U.S.C. § 101
Applicant's arguments filed 3/17/2026, pages 8-13, regarding the rejections of claims 1–4, 6–15, and 17–21 under 35 U.S.C. § 101 have been fully considered but are not persuasive for at least the following reasons.
Applicant's Argument: Applicant argues that the claims are not directed to an abstract idea because they recite a technological process for diagnosing diabetic retinopathy using OCT and OCTA data.
Examiner's Response: The argument is not persuasive. The claims recite collecting image data, segmenting the image data, extracting features including bifurcation points and crossover points, and applying a machine learning classifier to generate a diagnosis. These limitations constitute mathematical concepts and data analysis techniques, including feature extraction and classification, which fall within the abstract idea grouping of mathematical concepts and mental processes. The recitation of specific imaging modalities (OCT and OCTA) does not transform these data-gathering and data-processing steps into something other than an abstract idea; the modalities function only as conventional data sources. Similarly, the amendments adding HbA1C and dyslipidemia as clinical data inputs and specifying a random forest classifier do not alter this analysis, as selecting particular categories of input data and naming a conventional classifier remain part of the abstract idea of collecting and analyzing data to generate a result.
Applicant's Argument: Applicant argues that the claims are not comparable to the concepts identified in Alice, Bilski, or Benson.
Examiner's Response: This argument is not persuasive because the determination of whether a claim recites an abstract idea does not depend on whether the claimed concept is identical to those cases, but rather whether the claim recites subject matter falling within a recognized category of abstract ideas. Here, the claims recite data processing and mathematical analysis of medical data, which is a well-established abstract idea category. The Federal Circuit has consistently found that applying mathematical or computational analysis to a new domain or a new type of data does not move the claim outside the abstract idea category. See Electric Power Group, LLC v. Alstom S.A., 830 F.3d 1350, 1354 (Fed. Cir. 2016) (collecting, analyzing, and displaying information, even when applied to a specific technical field, remains an abstract idea).
Applicant's Argument: Applicant argues that the claims integrate the alleged abstract idea into a practical application by reciting specific types of medical imaging data (OCT and OCTA), specific extracted features (bifurcation points and crossover points), and the use of a machine learning classifier.
Examiner's Response: This argument is not persuasive. The additional elements of receiving OCT and OCTA data, segmenting retinal layers and vasculature, extracting features, and applying a classifier merely amount to data gathering and data processing steps that use the abstract idea as a tool. The claims do not recite any improvement to OCT or OCTA imaging technology, any specific improvement to image segmentation techniques, or any improvement to the functioning of the machine learning model itself. The specification does not describe a new random forest architecture, a novel training procedure, or any modification to the classifier that solves a specific technical problem in computer science or ML. Instead, the claims merely apply known techniques to analyze data and generate a diagnosis. Applicant's characterization of the claimed process as "quintessentially technological" is not persuasive because the technology (the imaging hardware, the computing system, the classifier) serves only as the environment in which the abstract idea is practiced. The mere application of the abstract idea in a medical or imaging context does not integrate the exception into a practical application. The use of specific types of data or specific features does not integrate the abstract idea into a practical application where the claim as a whole still amounts to collecting information, analyzing it using mathematical techniques, and reporting a result. The claimed invention does not effect a transformation of an article, does not control a particular machine in a meaningful way, and does not provide a specific technological improvement beyond the abstract data analysis itself. See MPEP Section 2106.05(a)-(c).
Applicant's Argument: Applicant argues that the claims recite an inventive concept because they involve a non-conventional combination of elements, including the use of OCT/OCTA data, clinical data such as HbA1C and dyslipidemia, and a machine learning classifier.
Examiner's Response: This argument is not persuasive. The use of machine learning classifiers, including random forest classifiers, for analyzing medical data is well-understood, routine, and conventional. Similarly, combining imaging data with clinical or demographic data for diagnostic purposes is a common practice in the medical and data science fields. As discussed in the new 35 U.S.C. Section 103 rejection set forth above, the use of HbA1C and dyslipidemia as clinical inputs in ML-based diabetic retinopathy analysis is supported by the prior art as being known and used in the field, confirming that this combination is not unconventional. The additional elements, individually and in combination, do not amount to significantly more than the abstract idea because they merely implement the abstract idea using generic data processing techniques and conventional machine learning tools. The claims do not recite any unconventional algorithm, specialized hardware, or specific technical implementation that would provide an inventive concept. Applicant's argument that conflating eligibility and patentability is improper does not apply here: the Step 2B inquiry specifically and permissibly looks to whether the additional elements are well-understood, routine, and conventional, and that determination is supported by the prior art of record as further expanded below.
Accordingly, the claims do not integrate the judicial exception into a practical application and do not recite significantly more than the abstract idea. The rejection under 35 U.S.C. Section 101 is therefore maintained.
Claim Rejections – 35 U.S.C. § 103
Applicant's arguments filed 3/17/2026, pages 13-15, regarding the previous 103 Rejections of claims 1-4, 7, 9-11, 13-15, and 17-21 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.
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
/MATTHEW KREMER/Primary Examiner, Art Unit 3791
Prosecution Insights