Segmentation-based Extraction of Key Components from ECG Images: A Framework for Precise Classification and Digitization

The physical and paper Electrocardiography (ECG) contain valuable insights into the history and diversity of cardiovascular diseases (CVDs). The development of algorithms that can digitize and classify these images could significantly improve our understanding and treatment of CVDs, particularly in underrepresented and underserved populations. As part of the George B. Moody PhysioNet Challenge 2024, we propose a deep learning approach to digitize and classify ECG images. Our methodology employs a deep learning segmentation model to extract key components, which are then used to train a classification model for the detection of CVDs and to digitize the signal. Our team, BAPORLab, achieved a signal-to-noise ratio of 5.493 placing 2nd in the digitization task. In the classification task, we achieved a macro F-measure of 0.730, ranked 3rd.