A Data-Driven Approach to Improve the Quality of Radiologist's Chest X-Ray Scanning
Perceptual errors in radiography account for most errors encountered while scanning Chest X-rays, leading to most misdiagnosed. In this scenario, radiologists train themselves in such a way that they develop their search-pattern strategy. These strategies, however, are untrustworthy due to a lack of quantitative descriptions and are not transferable. Therefore, we propose the use of a spatiotemporal feature encoding method. This method divides eye-fixation data into multiple temporal bins and is used by machine learning classifiers to distinguish between two radiologists, one expert and one novice. Classifiers using the proposed methodology outperform the current state-of-the-art in differentiating between the two radiologists. Following that, we propose an educational framework for improving scan quality. Radiologists receive appropriate feedback on quality metrics like their heterogeneity in scanning X-rays and how long it takes to complete a full scan. The number of interruptions made per scan, how much of the area of interest they covered, how accurate they were in identifying an area of interest within a scan, and finally, a complete summary of their performance throughout the training period. To achieve this goal, we used eye-gaze data collected by eye-tracking technology. This educational framework not only helps radiologists improve the quality of their Chest X-ray scans, but it also helps them develop more accurate search patterns based on the assessed metrics. The findings demonstrate the effectiveness of feedback and intervention in improving radiologists' skills and accuracy levels after four sessions and the usefulness of continuous improvement in the healthcare system.