Abstract
Introduction: Radiotherapy using the CyberKnife (CK) system for treating eye tumors requires tracking skull movements and immobilizing the eye. This study aims to develop a custom algorithm for a noninvasive eye position monitoring system in the CK framework.
Material and Methods: Retrospective data were analyzed from 13 patients treated with CK over five fractions. Eye immobilization was achieved using two guides above the patient’s head, with a black dot as the fixation eye. Camera recording was used to analyze pupil movements. A Python-based algorithm was developed to calculate eye position angles relative to the desired direction. A Quality Index (QI) was defined as the Common Area (CA) ratio of the 80% isodose and Planning Target Area (PTA) to the isodose area shifted by eye movement. For simplicity, only horizontal eye movement (angle φ) was considered. Dose distribution and CT data were processed in ImageJ, and a script calculated CA values for all integer φ angles. MediaPipe was used to map facial landmarks, estimate iris contours, and track pupil center positions on the XY plane.
Results: The algorithm analyzed camera recording during fractions of radiation, issuing visual and auditory warnings when gaze angles exceeded 20°. Instantaneous and average QI values were also monitored, with violations prompting controlled therapy interruptions. Analysis of 7 patients showed effective eye tracking. Two patients struggled to maintain their gaze, leading to frequent warnings. For six recordings, issues arose due to missing facial landmarks, 90° image rotations, or partially closed eyes.
Conclusion: The custom algorithm provides real-time eye movement analysis and can enhance treatment quality monitoring in CK-based eye tumor therapy.