Improving Accuracy of Double Reading in Chest X-Ray Images by Using Eye-gaze Metrics

Abstract

Accurate diagnosis of issues using medical images is crucial. One common approach for decreasing the risk of human diagnosis error is the double reading which is based on the assessment of each medical image by more than one radiologist. A novel method which increases the accuracy of the double reading approach while decreasing the required number of expert radiologist is proposed. The core idea behind the method is the better selection of the second expert by using subjective difficulty level of the medical image for the first doctor. The method estimates the subjective difficulty level by using eye-gaze metrics and the method assigns difficult images to the more experienced radiologists. The expertise level of the doctors are determined by the number of their previous decisions which are matched with the other experts' decisions. Effectiveness of the proposed system is shown by using real chest x-ray images and eye-gaze recordings.