Detection of improper fixation in MTI photoscreening images

Purpose: To determine the effect of fixation shift on photoscreening crescents, the ability of human interpreters to detect fixation shift, and the potential improvement by image processing. Methods: MTI photoscreening (Medical Technologies & Innovations, Inc, Lancaster, PA) images, measured at 11 positions of gaze, were obtained from 10 subjects (9 with refractive error warranting spectacle correction). Photographs were taken with subjects fixating at 20, 15, 10, and 5 cm to the left and the right of the camera fixation target (1 m distant); 5 cm above and below the camera fixation target; and on-axis. Photographs were inspected by 11 experienced raters, who indicated if the subject appeared to be looking directly at the camera. The photographs were digitized, enlarged, contrast enhanced, and measured by 3 raters. For each photograph, distance from the corneal light reflex to the nasal limbus was measured and a measure of asymmetry computed. Results: Raters could reliably detect off-axis fixation greater than 10 cm away from the intended fixation target. Raters correctly identified on-axis subject viewing 73% of the time. Crescents became larger when the fixation shifted off-axis for both the myopic and hyperopic subjects. Image analysis correctly classified 10 of 10 on-axis measurements and 34 of 39 off-axis measurements. Conclusion: Direct inspection of photoscreening images by trained raters can result in the failure to detect small but relevant errors of fixation. These fixation shifts can cause crescents to become larger than expected, resulting in false-positive classification. Image analysis offers a potential improvement in the detection of off-axis fixation in MTI photoscreening images.