Latent Space Classification for Cardiovascular Disease Detection: A Deep Convolutional Autoencoder-Based Approach for Telemedicine Applications

Cardiovascular disease is a leading global cause of mortality, often due to abnormal heart function. Early detection and timely treatment are essential to prevent fatalities. Electrocardiograms (ECGs) are critical non-invasive tools for diagnosing such conditions. With the rise of telemedicine, remote monitoring of cardiovascular patients has become more accessible. However, telemedicine sensor’s limited bandwidth and battery life demand efficient data transmission. This study introduces a Latent Space Classification System (LSCS) that compresses ECG signals into lower dimensions while preserving diagnostic accuracy, enhancing performance, and energy efficiency in telemedicine. The study analyses energy consumption through FLOPs, inference time, and transmission size to overcome sensor limitations across various feature extraction techniques. The proposed LSCS encompasses a deep convolutional autoencoder trained on the MIT-BIH arrhythmia database that compresses ECG signals. The compressed features are classified using seven ML models: K-Nearest Neighbors (KNN), Random Forest (RF), Extreme Gradient Boosting (XGBoost), Histogram Gradient Boosting Trees (HGBT), and three Support Vector Machine variants (SVML, SVMP, SVMR). LSCS-KNN, LSCS-XGBoost, and LSCS-HGBT achieved top performance, with over 98.8% accuracy, 98.7% precision, and 98.7% F1 scores. These results confirm LSCS’s ability to classify cardiovascular diseases accurately from compressed ECG data. Additionally, deep learning-based ECG sensors significantly reduced inference time and energy use (0.000107 J) compared to traditional methods. LSCS offers an efficient ECG transmission approach for telemedicine, effectively addressing bandwidth and energy limitations in remote cardiovascular monitoring.