Examining Human ECG Interpretability Through Eye Movement to Improve Automated Outcomes

The electrocardiogram (ECG) has been the de-facto tool for graphically representing the electrical activity of the human heart. The 12-lead ECG displays this activity from 12 ”perspectives” called ”leads”, in relation to the position of the 12 sensors attached to the body. The ECG is a practical and affordable diagnostic medical test, used in medical settings around the world. Despite the fact that more than three hundred million electrocardiograms are performed each year, the task of correctly interpreting them is considered a tedious and complex one. Failing to correctly interpret an ECG may result in injury or death. Therefore, several attempts at automating the interpretation of ECGs were made and are being developed with the advent of machine learning. Yet, these automation efforts are still aspiring to reach a sufficiently reliable level compared to human interpretation. Since the task requires both cognitive and visual efforts, eye tracking is a suitable methodology to reveal the dynamics of the interpretation procedure. This thesis aims at revealing indicators in the visual behavior of the interpreter that specify nuances useful in forthcoming ECG interpretation accuracy. This work uses eye-tracking as an enabling methodology to analyze how different health practitioners proceed into interpreting an ECG with relation to the accuracy of interpretation. Results of this work included quantifying the eye movement behavior of medical practitioners using eye tracking data, mainly fixations and gaze data, and correlating it with the roles of medical practitioners reflected through their accuracy diagnosing different heart arrhythmias.