Visual light perceptions caused by medical linear accelerator: Findings of machine-learning algorithms in a...

Area of Science:

• Medical Physics
• Radiation Oncology
• Ophthalmology

Background:

• Visual light perceptions (VLPs) during radiation therapy (RT) are poorly understood.
• Potential causes include radiation effects, phosphenes, or psychological factors.

Purpose of the Study:

• To investigate the incidence of VLPs during RT.
• To identify factors influencing VLP occurrence, such as radiation energy, dose, patient demographics, and RT location.
• To determine if VLPs are radiation-induced.

Main Methods:

• Analysis of 293 treatment fields from 78 patients undergoing head-and-neck or brain RT.
• Comparison of Random Forests (RF) and Logistic Regression (LR) machine learning models using Area Under the Curve (AUC).
• Patient questionnaires and interviews post-RT.

Main Results:

• VLPs were detected in 40.16% of exposure portals.
• RF model achieved a higher AUC (0.888) than LR (0.773).
• Retina fraction dose was the most significant continuous variable, and age was the most significant categorical variable influencing VLPs.

Conclusions:

• VLPs during RT are a real phenomenon induced by radiation.
• Radiation dose to the retina and patient age are key factors affecting VLP occurrence.
• Findings suggest VLPs are not self-suggested hallucinations or solely due to phosphenes.