An Interpretable Hand-Crafted Feature-Based Model for Atrial Fibrillation Detection

Abstract

Atrial Fibrillation (AF) is the most common type of cardiac arrhythmia. Early diagnosis of AF helps to improve therapy and prognosis. Machine Learning (ML) has been successfully applied to improve the effectiveness of Computer-Aided Diagnosis (CADx) systems for AF detection. Presenting an explanation for the decision made by an ML model is considerable from the cardiologists' point of view, which decreases the complexity of the ML model and can provide tangible information in their diagnosis. In this paper, a range of explanation techniques is applied to hand-crafted features based ML models for heart rhythm classification. We validate the impact of the techniques by applying feature selection and classification to the 2017 CinC/PhysioNet challenge dataset. The results show the effectiveness and efficiency of SHapley Additive exPlanations (SHAP) technique along with Random Forest (RF) for the classification of the Electrocardiogram (ECG) signals for AF detection with a mean F-score of 0.746 compared to 0.706 for a technique based on the same features based on a cascaded SVM approach. The study also highlights how this interpretable hand-crafted feature-based model can provide cardiologists with a more compact set of features and tangible information in their diagnosis.

Keywords: atrial fibrillation; classification; computer-aided diagnosis; feature importance; feature selection; interpretability.

Figure 1

Feature importance based on LR.

Figure 2

Feature importance based on PT.

Figure 3

Feature importance based on RF.

Figure 4

Feature importance based on SHAP.

Figure 5

Comparison of feature importance values obtained from different techniques.

Figure 6

Results of heart rhythm classification in terms of ${\mathcal{F}}_{\mu }$, by RF classifier applied to only the most important features. SHAP generates the best classification results based on only 28 features.

Figure 7

Results of heart rhythm classification in terms of ${\mathcal{F}}_M$, by RF classifier applied to only the most important features. SHAP generates the best classification results based on only 28 features.

Figure 8

ROC curves for different classes obtained from feature importance by SHAP and RF classification and 10-fold cross-validation.

Figure 9

LIME feature importance for different samples, selected randomly from test set, corresponding to class atrial fibrillation (A), class normal sinus rhythm (B), class other rhythms (C), and class noisy recordings (D), respectively.

Figure 10

Force plot for different samples, selected randomly from test set, corresponding to class atrial fibrillation (A), class normal sinus rhythm (B), class other rhythms (C), and class noisy recordings (D), respectively.