Interpretable ECG classification via a query-based latent space traversal (qLST)

TL;DR

This paper introduces qLST, a novel interpretability method for ECG classification that uses latent space traversal of a variational autoencoder to generate explanations for black box models, enhancing clinical interpretability.

Contribution

The paper proposes qLST, a query-based latent space traversal technique that provides explanations for ECG classifiers by generating representative ECGs, addressing limitations of existing saliency methods.

Findings

qLST successfully explains various ECG classifiers.

Generated ECGs reflect important features for classification.

Method improves interpretability of deep learning models in ECG analysis.

Abstract

Electrocardiography (ECG) is an effective and non-invasive diagnostic tool that measures the electrical activity of the heart. Interpretation of ECG signals to detect various abnormalities is a challenging task that requires expertise. Recently, the use of deep neural networks for ECG classification to aid medical practitioners has become popular, but their black box nature hampers clinical implementation. Several saliency-based interpretability techniques have been proposed, but they only indicate the location of important features and not the actual features. We present a novel interpretability technique called qLST, a query-based latent space traversal technique that is able to provide explanations for any ECG classification model. With qLST, we train a neural network that learns to traverse in the latent space of a variational autoencoder trained on a large university hospital dataset with over 800,000 ECGs annotated for 28 diseases. We demonstrate through experiments that we can explain different black box classifiers by generating ECGs through these traversals.