Machine-Learning Future Visual Acuity Prediction from Segmented Retinal Images

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional machine-learning models are inadequate in accurately and quickly predicting future visual acuity of subjects with eye-related diseases, leading to delayed and ineffective treatment selection.

Innovation Solution

A computer-implemented method using segment-processing and metric-processing machine-learning models to detect retina-related segments and generate segment-specific metrics, which are then processed to predict future visual acuity, incorporating deep-convolutional neural networks and gradient-boosting machines for enhanced accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If conventional machine-learning models are used to predict future visual acuity, then the prediction process can be performed, but the accuracy and timeliness of the prediction are insufficient

Engineering Contradiction:
Improveprediction accuracyVSAvoidprediction timeliness
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent segments the retinal image into multiple distinct regions (macula, fovea, optic disc, blood vessels, and background) using a segmentation network. This segmentation allows the system to extract and analyze specific morphological features from each region independently, improving prediction accuracy by focusing on clinically relevant areas while maintaining computational efficiency through parallel processing of segmented regions.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent transforms 2D retinal images into 3D feature representations by extracting morphological features (area, perimeter, circularity, compactness) from segmented regions and combining them with temporal dimensions (baseline and follow-up time points). This dimensional transformation enables the system to capture dynamic changes in retinal structure over time, improving both prediction accuracy and timeliness by analyzing structural evolution rather than static images.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

2Reliability

If treatment selection is delayed until efficacy is confirmed, then treatment effectiveness can be verified, but the window for early intervention is lost

Engineering Contradiction:
Improvetreatment efficacy verificationVSAvoidintervention timing
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The patent performs preliminary prediction of future visual acuity using machine-learning models before treatment initiation. By analyzing baseline retinal images and extracting morphological features from segmented regions, the system predicts likely treatment outcomes in advance. This preliminary action enables clinicians to select treatments based on predicted efficacy rather than waiting for actual treatment results, thereby maintaining the window for early intervention while still verifying effectiveness through follow-up imaging.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent incorporates feedback loops where follow-up retinal images are compared against baseline images and treatment responses are monitored. The system uses this feedback to verify predicted treatment efficacy and adjust future predictions. This feedback mechanism ensures that treatment selection is both timely (based on preliminary predictions) and reliable (verified through ongoing monitoring of treatment response and visual acuity changes).

Inventive Principle:
Principle #23Feedback

Data Source

PatentEP4049287B1Machine-learning techniques for prediction of future visual acuity
Publication Date: 2025.08.20 F HOFFMANN LA ROCHE & CO AG
  • EP4049287B1 patent drawingFigure 1
  • EP4049287B1 patent drawingFigure 2
  • EP4049287B1 patent drawingFigure 3

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.