From Paper to Digital: ECG Processing with U-Net Digitization and ResNet Classification

Introduction: Electrocardiograms (ECGs) exist in both paper and digital formats.While digital ECGs offer ease of analysis and categorization, paper ECGs are still widely used.The PhysioNet Challenge 2024 aims to digitize and classify ECGs captured from images or paper printouts.Method: YOLOv8 Tiny was utilized for image correction.The YOLO identified the tilt angle between the upper and lower lead names.And the angle was used to correct the image.For the digitization task, a ResUNet model with an integrated CBAM module was employed.A novel method was proposed to process input and output images, followed by a column-by-column scanning of the binary output image to derive one-dimensional signals.For the classification task, ResNet50 was utilized.ECGs were classified based on the predicted values of 11 output classes.Result: Our team, USST Med, received a challenge score of an SNR of 2.202 for the digitization task (ranked 5th out of 16 teams) and a macro F-measure score of 0.393 (ranked 8th out of 16 teams) for the classification task on the hidden test set.Conclusion: The proposed approaches had good digitalization and classification effects on paper ECGs.