Federated learning-inspired smart ECG classification: an explainable artificial intelligence approach

Cardiovascular Disease (CVD) treatment often requires the use of digital cardiac data, which can be leveraged to improve patient outcomes in critical cases such as heart attacks. It can be challenging to compile a sizable volume of medical information from one place. To address these challenges, a novel end-to-end framework has been proposed that utilizes federated learning to ensure data availability. In this article, a new algorithm that employs deep learning is introduced for classifying Electrocardiogram (ECG) arrhythmias in a federated environment. The proposed approach integrates a Support Vector Machine (SVM) classifier with a Bi-directional Long Short-Term Memory (Bi-LSTM)-based Auto-Encoder (AE) network. The AE network is employed for learning the characteristics of ECG arrhythmias signals, while the SVM identifies the signals based on the learned characteristics. The AE network comprises an encoding module that uses an LSTM network to extract high-level features from ECG arrhythmias signals and a decoding module uses the second LSTM network to reconstruct the signals from the extracted high-level features. The algorithm can accurately identify Normal (N), Right Bundle Branch (RBB), Left Bundle Branch (LBB), Premature Ventricular Contractions (PVC) heartbeats, and Atrial Premature Complexes (APC). By employing ten-fold cross-validation, the proposed method surpasses state-of-the-art techniques, achieving an accuracy of up to 95.57% for noisy data and 99.12% for clean data. Furthermore, the suggested approach contains an Explainable Artificial Intelligence (EAI) module to improve the interpretability of categorization findings, allowing medical professionals to make more precise and timely judgments. The framework has been verified and validated for ECG signal categorization, but it may also be used in a variety of other healthcare applications.