کنفرانس مهندسی برق ایران

صفحه اصلی / سی و سومین کنفرانس بین المللی مهندسی برق

Digitizing Analog ECGs: A Deep Learning Pipeline for Converting Historical Records into High-Quality Digital Signals

نویسندگان :

Sahar Askari¹ Somayeh Afrasiabi²

1- دانشگاه شیراز 2- دانشگاه شیراز

کلمات کلیدی :

historical ECG records،ECG digitization،deep learning in healthcare،UNet segmentation

چکیده :

Paper-based electrocardiogram (ECG) records pose challenges such as data degradation, limited retrospective analysis, and lack of standardization. This study presents a robust digitization pipeline to convert analog ECGs into high-quality digital formats. Using the PTB Diagnostic ECG Dataset (PTB-XL) and synthetic images with real-world distortions, the process includes segmentation with ResUNet18, alignment correction with The Oriented Features from Accelerated Segment Test (FAST) and Rotated Binary Robust Independent Elementary Feature (BRIEF) descriptor (ORB) and Random Sample Consensus (RANSAC), and lead detection using you only look once (YOLO) model. The binary images are then transformed into calibrated digital signals and processed for noise reduction and baseline correction. Evaluation shows the method achieves a median-based signal-to-noise ratio (SNRmed) of 11.803 dB, demonstrating its effectiveness for preserving signal quality. This approach enables integration of historical ECG data into modern workflows, supporting large-scale research and machine learning applications.