OCT Retina Image Analysis for Confidence-Based DIRF Mapping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current systems struggle to accurately and reliably identify and quantify diffuse intraretinal fluid (DIRF) in Optical Coherence Tomography (OCT) images, which is crucial for diagnosing and treating macular diseases, as existing segmentation techniques lack precision and consistency.
Innovation Solution
A machine learning-based system using a trained model to determine the location and volume of DIRF in OCT images by employing probability-based labeling and aggregation of expert annotations, enabling the identification of high and low confidence regions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing segmentation techniques are used to identify DIRF, then the analysis process can be performed, but the precision and reliability of identification are insufficient
Solution Approach 1:
The patent replaces manual segmentation methods with an automated machine learning-based system that uses neural networks to identify DIRF regions. The system processes OCT images through multiple neural network models that have been trained to detect fluid density patterns, replacing the mechanical/manual process of image segmentation with an intelligent automated system that achieves superior precision and reliability.
Solution Approach 2:
The patent transforms the identification process by changing from binary classification (fluid present/absent) to probability-based labeling that captures fluid density variations. The system uses multiple probability thresholds to categorize regions into different confidence levels, enabling nuanced differentiation of DIRF regions based on fluid density and spatial distribution characteristics.
2Measurement precision
If probability-based labeling with multiple thresholds is applied, then the confidence levels and differentiation of DIRF regions improve, but the complexity of the analysis system increases
Solution Approach 1:
The patent segments the analysis process into distinct stages: first applying a neural network to generate probability maps, then applying multiple probability thresholds to categorize regions into different confidence levels. This segmentation allows the complex task of DIRF differentiation to be broken down into manageable steps, each handled by a specialized component of the system.
Solution Approach 2:
The patent introduces probability thresholding as an intermediary mechanism between the neural network output and the final DIRF classification. This intermediary layer translates continuous probability values into discrete confidence categories, serving as a bridge that simplifies the interpretation of neural network outputs while maintaining the nuanced differentiation capability.
Data Source
Figure 1
Figure 2
Figure 3A
AI summary
Methods and systems of analyzing images of a retina captured by an Optical Coherence Tomography (OCT) scanner are disclosed. The methods and systems use a processor configured to implement a series of instructions that include creating a training set of images of a retina of a patient captured by an OCT scanner including labeled regions of diffuse intraretinal fluid (DIRF) such as by assessing maximum and minimum regions of DIRF and regions of pathology of the retina to generate a model, and analyzing the DIRF region or regions of pathology by the model to derive an assessment of the retina of the patient. The assessment of the retina enables, for instance, the treatment of the patient's retina to be evaluated and determined.