Retinal OCT Segmentation for Faster Visual Acuity Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional machine-learning models struggle to accurately and quickly predict future visual acuity in subjects with eye-related diseases, leading to delayed and ineffective treatment selection.
Innovation Solution
A system utilizing segment-detecting and metric-processing machine-learning models to analyze OCT images of the retina, detecting specific retinal structures and fluids, and generating predictions based on segment-specific metrics using gradient-boosting and deep-learning techniques.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional machine-learning models are used to predict future visual acuity, then the prediction can be made, but the accuracy is insufficient and the time required is excessive
Solution Approach 1:
The patent segments the retinal image into multiple distinct regions (macula, fovea, optic disc, blood vessels, fluid collections) and processes each segment separately using specialized models. This segmentation allows the system to extract relevant features from specific retinal areas without being distracted by irrelevant information, thereby improving prediction accuracy while maintaining efficient processing time.
Solution Approach 2:
The patent transforms the prediction approach from a single holistic image analysis to a multi-dimensional analysis that processes different retinal segments simultaneously at various resolution levels. By analyzing segments at multiple scales and combining results through ensemble models, the system achieves higher accuracy without linearly increasing computation time.
2Reliability
If treatment selection is delayed until efficacy is confirmed, then treatment effectiveness can be verified, but the window of opportunity for early intervention is lost
Solution Approach 1:
The system performs preliminary prediction of future visual acuity using machine-learning models before treatment begins. By predicting the likely outcome of different treatment options based on current retinal imaging, the system enables clinicians to select the most effective treatment in advance, avoiding the need to wait for efficacy confirmation while maintaining reliable treatment selection.
3Measurement precision
If comprehensive retinal analysis is performed to improve prediction accuracy, then more detailed information is obtained, but the processing complexity increases
Solution Approach 1:
The patent divides the complex retinal image analysis into separate modular segments, each handled by specialized models optimized for specific structures (e.g., fluid collection detection, retinal thickness measurement, vascular analysis). This modular segmentation reduces overall processing complexity by allowing independent optimization and parallel processing of each segment.
Solution Approach 2:
The system applies different processing quality levels to different retinal segments based on their clinical relevance. Critical areas such as fluid collections and macular edema receive enhanced analysis, while less critical areas are processed more lightly. This local quality adjustment maintains high prediction accuracy for relevant features while reducing overall processing complexity.
Data Source
AI summary
Methods and systems disclosed herein relate generally to systems and methods for predicting a future visual acuity of a subject by using machine-learning models. An image of at least part of a retina of a subject can be processed by one or more first machine-learning models to detect a set of retina-related segments. Segment-specific metrics that characterize a retina-related segment of the set of retina-related segments can be generated. The segment-specific metrics can be processed by using a second machine-learning model to generate a result corresponding to a prediction corresponding to a future visual acuity of the subject.