Time adaptive ECG driven cardiovascular disease detector

Electrocardiograms (ECGs) are widely used to detect cardiovascular disease (CVD). Deep learning is a topic of interest in healthcare, which timely detection ECG anomalies can play vital role patient monitoring. However, automatic CVD via ECGs complex problem, where state-of-the-art performance achieved typically by the rule-based diagnosis models, inefficient deal with large amount heterogeneous data and requires significant analysis medical expertise achieve adequate precision diagnosis. In this paper, we propose two-stage multiclass algorithm. The first stage performs segmentation based on Convolutional Bidirectional Long Short-Term Memory neural networks attention mechanism. A second time adaptive Neural applied beats extracted from for several intervals. converted 2D images using Short-Time Fourier Transform automatically discriminate normal cardiac adverse events such as arrhythmia congestive heart failure predict sudden death. Model accuracy was compared across different scales. Data train test models were MIT/BIH-PhysioNet databases. By 4 min ECG, an 100% events, 97.9% deaths. This offers unprecedented results supporting domain-experts work, computing signal characteristics automated complete system showed be promising algorithms similar purposes.