Tfdginixml: a novel explainable machine learning framework for early detection of cardiac abnormalities based on nonlinear time-frequency distribution gini index features

Cardiovascular diseases (CVDs) are the leading cause of global death, with approximately 80% of such CVD mortalities occurring in low and middle-income regions. Early detection of cardiac abnormalities is essential for timely intervention and minimizing mortalities. Automated CVD detection methods are vital, particularly in areas with limited healthcare resources. However, most existing AI-based techniques lack three critical aspects: model interpretability, longer-duration analysis, and effective use of nonlinear time-frequency approaches, which are necessary for ECG signals due to their nonlinear, nonstationary, and multi-component nature. This study proposes explainable intelligent classifiers incorporated with a novel sequence of time-frequency energy Gini Index (GI) features from the QRS complexes of ECG signals to address these challenges and enable early-stage CVD detection. These features are extracted using the Choi-Williams Time-Frequency method, reporting the first instance application of GI measures to nonlinear time-frequency distribution (TFD) for ECG analysis. Features are computed from one-minute windows, covering 30 minutes of ECG recordings. These interpretable features provide clear insights into normal and abnormal ECG patterns. The proposed method was trained and validated using the MIT-BIH Arrhythmia and Fantasia-Normal databases. Eight machine learning classifiers, including SVM, Random Forest, XGBoost, Gaussian Naïve Bayes, KNN, LinearBoost, CatBoost, and Logistic Regression were tested. The best model achieved 100% sensitivity, 94.4% accuracy, 95.24% F1-score, 90% precision, and 92.59% AUC on the test dataset. High sensitivity ensures reliability for medical screening by reducing False Negatives, making the approach suitable for integration into any type of smart device for accurate online and offline monitoring of CVD abnormalities.

Text 121567.pdf - Published Version Available under License Creative Commons Attribution. Download (3MB) Official URL or Download Paper: https://ieeexplore.ieee.org/document/10981715/

Actions (login required)

View Item