ECGLight: Compute-Light Framework For Paper ECG Digitization and Myocardial Infarction Screening

As cardiovascular diseases remain a leading cause of global mortality, timely diagnosis via electrocardiography (ECG) is critical. However, millions of ECGs are still stored as paper printouts in remote or resource-limited clinics, rendering them inaccessible to modern AI-based clinical decision support systems. This study addresses the need for a lightweight, on-device pipeline capable of digitizing these paper records and performing automated diagnostic screening without requiring high-end computational resources or high-speed internet.

The researchers created an end-to-end pipeline that processes a smartphone photo or scan of a paper ECG. The system uses a YOLOv11-based segmentation model to isolate waveforms, detect lead labels, and identify reference pulses for physical calibration. By leveraging a patch-based inference strategy, the model maintains high fidelity even when processing noisy or degraded images. Once digitized, the 12-lead signals are processed for baseline correction and R-peak detection, followed by classification models trained to detect Myocardial Infarction (MI) and Occlusion Myocardial Infarction (OMI). The entire system is optimized for CPU-only execution, ensuring it can run on standard laptops or edge devices.

The framework demonstrates high performance across both synthetic and real-world datasets. On the PTB-XL dataset, the system achieved 95.51% accuracy (F1 = 0.9519) for MI detection. When validated on the hospital-acquired ECG-Matrix dataset, it maintained an accuracy of 88.89% (F1 = 0.8862) for OMI detection. Crucially, the entire digitization-to-diagnosis workflow completes in under 30 seconds on a standard CPU, proving that legacy paper records can be effectively democratized for AI-assisted clinical decision support in resource-constrained environments.

This work provides a scalable solution for integrating legacy paper-based medical data into modern digital health workflows. By enabling on-device, offline analysis, the framework helps bridge the diagnostic gap in remote clinics, potentially accelerating the detection of life-threatening conditions like acute coronary occlusion and facilitating timely reperfusion therapy where digital infrastructure is currently lacking.