Retinal Disease Classification via Merged OCT and OCTA Vectors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for classifying diabetic retinopathy (DR) using optical coherence tomography (OCT) and OCT angiography (OCTA) face challenges such as low sensitivity and specificity, particularly in detecting diabetic macular edema, and rely on accurate retinal layer segmentation, which can become unreliable with advanced pathology, leading to misclassification.
Innovation Solution
A deep learning-based approach utilizing a convolutional neural network (CNN) processes a single 3D volumetric image of the retina to generate a single-dimensional vector, which is then used to classify DR levels without relying on additional images or complex segmentation techniques, enabling accurate classification of DR severity using OCT and OCTA data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If fundus photographs are used for DR classification, then the system is simple and cost-effective, but the sensitivity and specificity for detecting diabetic macular edema are low (60-73% sensitivity, 67-79% specificity)
Solution Approach 1:
The patent combines structural OCT and OCTA data into a unified deep learning framework. The 3D volumetric OCT data and OCTA angiographic data are merged and processed together through a single deep learning model, allowing simultaneous utilization of both structural and vascular information for DR classification, thereby improving detection accuracy without requiring separate analysis pipelines
Solution Approach 2:
The deep learning model is designed to perform multiple functions: it classifies DR severity levels, detects diabetic macular edema, and differentiates between various DR stages all within a single unified system. This multi-functional approach eliminates the need for separate specialized systems for each diagnostic task
2Reliability
If multiple imaging modalities (fundus photography + OCT) are used to improve DME detection, then the diagnostic accuracy improves, but the logistic challenges and cost increase
Solution Approach 1:
The patent merges structural OCT and OCTA data processing into a single deep learning framework. By combining these modalities at the data processing level rather than requiring separate analysis systems, the approach improves DME detection accuracy while reducing the logistic complexity of managing multiple independent imaging systems
Solution Approach 2:
The patent uses OCTA to capture vascular information that complements structural OCT data. The OCTA volume data is processed to generate en face views and vascular maps that serve as additional diagnostic information, effectively creating a comprehensive diagnostic copy that includes both structural and vascular characteristics
3Measurement precision
If retinal layer segmentation is performed to improve classification accuracy, then the precision of biomarker measurement improves, but the system becomes unreliable with advanced pathology due to segmentation errors
Solution Approach 1:
The patent extracts and utilizes OCTA vascular biomarkers that are less susceptible to segmentation errors compared to structural OCT measurements. By focusing on vascular features from OCTA data such as vessel density and caliber, the system maintains measurement precision even when retinal layer segmentation becomes unreliable in advanced pathology cases
Solution Approach 2:
The patent changes the measured parameters from primarily structural OCT-based metrics to include OCTA-based vascular parameters. This parameter shift allows the system to maintain accuracy in advanced DR cases where structural changes make segmentation difficult, as vascular features remain measurable through OCTA angiography
4Productivity
If a single 3D volumetric OCT image is used for classification, then the system simplicity and processing efficiency improve, but the classification accuracy may be insufficient compared to multi-modal approaches
Solution Approach 1:
The patent merges structural and vascular information within a single 3D volumetric OCT framework by integrating OCTA data. This allows the system to maintain processing efficiency of a single modality while achieving classification accuracy comparable to multi-modal approaches through the combined information content of structural and vascular features
Solution Approach 2:
The patent adds the vascular dimension to the traditional structural OCT analysis by incorporating OCTA data. This transforms the analysis from purely structural (one dimension) to include both structural and vascular characteristics (multiple dimensions), thereby improving classification accuracy while maintaining the efficiency of a single imaging session
Data Source
AI summary
Methods and systems for identifying levels of an ophthalmic disease are described. An example method includes generating, by a convolutional neural network (CNN) and using a 3D image of a retina, a vector. The method further includes generating, by a first model and using the vector, a first likelihood that the retina exhibits a first level of an disease and generating, by a second model and using the vector, a second likelihood that the retina exhibits a second level of the disease. The method further includes determining whether the retina exhibits an absence of the ophthalmic disease, the first level of the disease, or the second level of the disease based on the first likelihood and the second likelihood. Further, an indication of whether the retina exhibits the absence of the ophthalmic disease, the first level of the ophthalmic disease, or the second level of the ophthalmic disease is output.