Automated Optical Reading of Scanned ECGs

Abstract

Electrocardiogram (ECG) is a valuable tool for medical diagnosis used worldwide. Its use has contributed significantly to the prevention of cardiovascular diseases including infarctions. Although physicians need to see the printed curves for a diagnosis, nowadays there exist automated tools based on machine learning that can help diagnosis of arrhythmias and other pathologies, these tools operate on digitalized ECG data that are merely one-dimensional discrete signals (a kind of information that is much similar to digitized audio). Thus, it is interesting to have both the graphical information and the digitized data. This is possible with modern, digital equipment. Nevertheless, there still exist many analog electrocardiogram machines that plot results on paper with a printed gris measured in millimeters. This paper presents a novel image analysis method that is capable of reading a printed ECG and converting it into a sampled digital signal.