Convolutional Neural Networks Cascade for Automatic Pupil and Iris Detection in Ocular Proton Therapy

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Deep learning-based eye tracking is increasingly used in various fields.
• Ocular proton therapy (OPT) requires precise patient alignment.

Purpose of the Study:

• To develop and evaluate an automated deep learning eye tracking system for ocular proton therapy.
• To assess the system's accuracy in detecting iris and pupil during OPT treatments.

Main Methods:

• A two-stage convolutional neural network (CNN) approach was implemented for eye tracking.
• The system was trained and tested on 707 video frames from clinical OPT sessions.
• Performance was evaluated against manual annotations using Dice, Szymkiewicz-Simpson, Intersection over Union, and Hausdorff distance metrics.

Main Results:

• The automated system achieved high accuracy in iris and pupil detection, with median Dice coefficients of 0.94 and 0.97, respectively.
• Quantitative metrics demonstrated that the system's predictions closely matched manual ground truths.
• The framework showed comparable performance to clinical operator delineations.

Conclusions:

• The proposed deep learning eye tracking framework offers an automatic and accurate method for evaluating pupil and iris misalignments in OPT.
• This tool can support clinical activities by providing quantitative assessments without altering established workflows.