CEFEs: A CNN Explainable Framework for ECG Signals.

Area of Science:

• Artificial Intelligence in Healthcare
• Machine Learning Interpretability
• Biomedical Signal Processing

Background:

• Deep learning (DL) models are increasingly used in healthcare decision support systems for tasks like Electrocardiogram (ECG) signal classification.
• The "black-box" nature of DL models hinders trust and understanding among domain experts regarding their decision-making processes.
• Existing research on explaining DL models, particularly Convolutional Neural Networks (CNNs), primarily focuses on non-medical domains and 2D data, leaving a gap for medical time-series data.

Purpose of the Study:

• To address the need for interpretable DL models in healthcare, this study proposes a novel framework for explaining 1D-CNN models used for ECG signal analysis.
• The framework aims to provide functional understanding of how these models classify patient data by offering insights into their internal mechanisms.
• Enhance trust and confidence in AI-driven healthcare decision support systems through transparent model explanations.

Main Methods:

• A modular framework, CNN Explanations Framework for ECG Signals (CEFEs), was developed for interpretable explanations of 1D-CNN models.
• CEFEs modules offer functional understanding through data descriptive statistics, feature visualization, feature detection, and feature mapping.
• The framework evaluates model capacity by correlating learned features with raw signals and classification performance.

Main Results:

• CEFEs provides interpretable explanations of 1D-CNN models for ECG data, going beyond traditional performance metrics.
• Feature visualization and quantifiable metrics offered by CEFEs reveal the quality of the DL model's learned features.
• The framework demonstrates the correlation between a model's classification capacity and its learned features, validated across different dataset volumes.

Conclusions:

• The proposed CNN Explanations Framework for ECG Signals (CEFEs) offers a valuable tool for interpreting deep learning models in the context of medical time-series data.
• CEFEs enhances the transparency and trustworthiness of AI in healthcare by providing functional insights into model behavior.
• Interpretable AI, as demonstrated by CEFEs, is crucial for the adoption and reliable application of deep learning in clinical decision support systems.