Multi-Task OCT CNN for Geographic Atrophy Growth Forecasting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current imaging modalities, such as two-dimensional Fundus Auto-Fluorescent (FAF) images, are limited in providing refined structural information for geographic atrophy (GA) lesion area, and there is a need for improved assessments of GA onset and progression to understand and predict geographic-atrophy lesion growth.
Innovation Solution
A three-dimensional convolutional neural network (CNN) is used to process OCT data, enabling simultaneous prediction of current GA lesion size and subsequent growth, directly from baseline 3D OCT images without additional processing steps, utilizing multi-task models to enhance accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If two-dimensional FAF images are used to quantify geographic-atrophy lesion area, then the measurement process is simple and fast, but the structural information and measurement precision are limited
Solution Approach 1:
The patent transitions from two-dimensional FAF imaging to three-dimensional OCT imaging to quantify geographic-atrophy lesion volume. By adding the depth dimension, the system captures comprehensive structural information including RPE, photoreceptors, and choriocapillaris layers, enabling more precise measurement of lesion characteristics while maintaining automated processing through machine learning models.
2Loss of information
If traditional FAF imaging is used, then the assessment of GA progression is straightforward, but the ability to provide refined structural information and predict growth rates is insufficient
Solution Approach 1:
The patent segments the three-dimensional OCT data into distinct anatomical layers (RPE, photoreceptors, choriocapillaris) and identifies specific lesion characteristics within each layer. This segmentation enables comprehensive structural information extraction and facilitates detailed characterization of geographic-atrophy lesions, including volume, shape, and layer-specific involvement.
Solution Approach 2:
The patent replaces manual assessment methods with automated machine learning models that process three-dimensional OCT data. The convolutional neural network automatically extracts structural features, quantifies lesion characteristics, and predicts growth rates, eliminating the need for complex manual measurements and providing consistent, reproducible results.
3Productivity
If manual assessment methods are used for GA lesion area, then the process is interpretable, but the productivity and consistency of measurements are reduced
Solution Approach 1:
The patent implements self-service through automated machine learning models that independently process three-dimensional OCT data, extract lesion characteristics, and generate measurements without human intervention. The convolutional neural network automatically identifies geographic-atrophy lesions, quantifies their volume and shape, and predicts growth rates, ensuring consistent and reproducible results while significantly increasing assessment throughput.
Solution Approach 2:
The patent incorporates feedback mechanisms where the machine learning models are trained on labeled data and continuously improved through validation against ground truth measurements. The system provides feedback on measurement confidence levels and allows for iterative refinement, ensuring both high productivity and measurement consistency across different cases.
Data Source
AI summary
Embodiments disclosed herein generally relate to predicting geographic-atrophy lesion growth and/or geographic atrophy lesion size in an eye. The predictions can be generated by processing a data object using a neural network. The data object may include a three-dimensional data object representing a depiction of at least part of the eye or a multi-channel data object representing one or more decorresponding pictions of at least part of the eye. The neural network can include a convolutional multi-task neural network that is trained to learn features that are predictive of both lesion-growth and lesion-size outputs.