Deep Learning-Based Eye Monitoring and Tracking System For Ocular Proton Therapy in a Regular Gantry Room

Document Type

Conference Proceeding

Publication Date

12-2025

Publication Title

International Journal of Particle Therapy

Abstract

Background and Aims: Proton therapy is an effective treatment for uveal melanoma. A novel camera-based eye-tracking system was developed to achieve noninvasive eye positioning and motion monitoring during proton therapy. Methods: An infrared camera, paired with an infrared light source, was placed anteriorly to illuminate the pupil (Figure 1). Patients were instructed to focus on a gaze-point on a customized gaze-fixation device. A deep learning-based, Mediapipe’s FaceMesh module was used for face recognition and eye localization. The eye frame was fed into the model and trained using TensorFlow to predict the pupil coordinates. The system incorporated a real-time alert mechanism to monitor pupil displacement, enabling responses to user-defined tolerance. The eye-tracking system was clinically implemented for uveal melanoma proton therapy in a regular gantry room. Daily cone-beam CT (CBCT) and kV images guided patient alignment. The system offers real-time eye position data, which was confirmed and calibrated against pre-treatment CBCT. Pupil displacement was quantified as the difference between the real-time and calibrated positions. Results: The eye-tracking system achieved real-time localization with sub-millimetric precision and a response time of 0.5 seconds. Figure 2(a) shows the pupil position for a representative ocular patient; points 1 to 6 correspond to the beam-on and beam-off time points for three treatment fields. The dashed line represents the 1/2/ 3 mm tolerance levels, with observed displacements remaining within 1.5 mm (figure 2(b)), consistent with kV measurement. Figure 2(c, d) illustrates the system’s real-time alert mechanism. The x-y coordinates below the eye frame show the pupil position. Green dots indicate the displacement is within tolerance (2(c)), whereas red dots in Figure 2(d) highlight deviation beyond tolerance, observed in this case when the patient relaxed after the beam delivery completed. Conclusions: The in-house camera-based eye-tracking system effectively monitors and tracks eye position, providing eye movement alerts during ocular proton therapy. This system ensures the quality and efficiency of treatment by maintaining precise eye positioning throughout the treatment.

Volume

17

Issue

Suppl

First Page

100908

Last Page

100908

Comments

63rd Annual PTCOG (Particle Therapy Cooperative Group) Conference, June 2-7, 2025, Buenos Aires, Argentina

DOI

10.1016/j.ijpt.2025.100908

Link to Full Text

Share

COinS