SHAP-Based Multi-Lead ECG Explanations with Lead Contributions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing deep learning models for cardiovascular disease classification using multi-lead electrocardiograms provide results without explaining the reasoning behind their decisions, leading to a lack of trust and understanding among medical professionals.
Innovation Solution
A processor-implemented method and system that generate quantifiable explanations by classifying multi-lead electrocardiogram signals, computing contribution values for each data point, filtering and normalizing these values, determining prominent leads, and analyzing their contribution to disease classes, using SHapley Additive exPlanations (SHAP) with post-processing steps for enhanced explainability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If deep learning models are used for disease classification, then classification accuracy is improved, but explainability deteriorates
Solution Approach 1:
The patent introduces an explainer model as an intermediary component that bridges the deep learning classifier and the user. The explainer model receives inputs from the classifier and generates human-interpretable explanations, making the black-box predictions transparent without affecting the classifier's accuracy. This mediator layer translates complex model decisions into understandable terms for medical professionals.
Solution Approach 2:
The system segments the explanation generation process into distinct components: contribution value calculation for individual data points, filtering of significant contributions, and aggregation into lead-level explanations. This segmentation allows the complex explanation task to be broken down into manageable steps, each handled by specialized sub-components that work together to provide comprehensive explanations.
2Ease of operation
If qualitative explanations are provided, then ease of understanding is improved, but quantifiability deteriorates
Solution Approach 1:
The patent transforms qualitative explanations into quantitative form by calculating specific contribution values for each data point using the trained classifier. These contribution values are numerical parameters that can be filtered, normalized, and aggregated. By changing the representation from purely qualitative to quantitative, the system enables both easy understanding through numerical metrics and rigorous measurement of explanation quality.
Solution Approach 2:
The system incorporates feedback mechanisms where the explained data points and their contribution values are used to verify and refine the explanations. The feedback loop allows the system to compare generated explanations against ground truth and continuously improve the quantifiability and accuracy of explanations, creating a self-validating system.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Deep Learning (DL) performs well in cardiovascular disease (CVD) classification using 12-lead Electrocardiogram (ECG). However, explainable artificial intelligence in CVD classification still remains largely qualitative. Embodiments of the present disclosure provide systems and methods that implement a Region of Interest (ROI) based quantifiable explanation for multi-lead ECG. CVD specific post-processing steps are added, to increase the explanation performance. Furthermore, the system enables selection of an optimal DL model, within the performance space defined by classification, explanation, and time-complexity.