Medical Image Label Prediction With Segmentation-Based Explanations
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing artificial intelligence systems for digital image classification, particularly in medical imaging, lack explainability, making it difficult for users to understand the reasoning behind predictions and comply with regulatory requirements for transparency.
Innovation Solution
A method involving a segmentation neural network and a classification neural network, optimized through a learning phase using cost functions to maximize label quality, probability of background classification, and sparsity, providing explainable predictions by associating pixel classifications with image labels.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a multilayer neural network is used for automatic image classification, then predictive accuracy is improved, but explainability deteriorates (the system becomes a black box)
Solution Approach 1:
The patent introduces an intermediary explanation module that acts as a mediator between the neural network's internal representations and the user. This module generates human-interpretable explanations by analyzing which image regions most influenced the prediction, using techniques like gradient-based sensitivity analysis or attention mechanisms to highlight relevant areas without altering the core prediction engine.
Solution Approach 2:
The patent segments the image into multiple regions and processes each region through the neural network separately, then aggregates the results. This segmentation approach allows the system to identify which specific regions contributed to the final prediction, providing explainability while maintaining the power of deep learning for accurate classification.
2Measurement precision
If the neural network processes all pixels in the image, then predictive accuracy is improved, but processing time increases
Solution Approach 1:
The patent extracts and processes only the most relevant portions of the image rather than all pixels. By identifying and focusing on salient regions that contain diagnostic information, the system reduces the amount of data to be processed while maintaining predictive accuracy, thereby decreasing processing time.
Solution Approach 2:
The patent applies partial processing by focusing computational resources on the most informative regions of the image rather than uniformly processing all pixels. This selective approach uses sufficient (but not excessive) processing on critical areas to achieve accurate predictions with reduced overall computation time.
Data Source
AI summary
Method for the prediction of labels associated with a digital image, comprising a prediction phase consisting of:supplying the image to a segmentation neural network configured to predict a classification of the pixels of the image into a first set of classes; andsupplying at least part of this classification to a classification neural network configured to predict a set of labels for said image, based on the classification P of the pixels;said segmentation and classification neural networks being determined by a learning phase comprising, for each image of a training set, the first and second steps; and determining a location of the background of said image, based on the classification of the pixels, and optimizing the weights of the neural networks according to a set of cost functions configured, by iteration, to maximize the quality of the set of labels as a function of labels previously established and associated with the image, and to maximize the probability of not predicting any label for the background.