Stereotactic Radiation Therapy or Protons for Uveal Melanoma Patients? An Artificial Intelligence (AI)-Based Clinical Treatment Decision-Making Tool Predicting Doses To Radiation Therapy Constraints

Abstract

PURPOSE: For ocular melanoma, selecting between stereotactic radiation therapy (SRT) and protons requires a lengthy plan comparison process. The purpose of this brief report is to describe an artificial intelligence (AI) decision-making tool to predict dosimetric and clinical outcomes based on easy-to-access tumor characteristics. METHODS AND MATERIALS: The AI tool was based on a retrospective database of 66 patients with uveal melanoma treated in a single center with robotic SRT. A supervised machine learning model was developed to correlate the risk of toxicity for each radiation modality and clinical features. Clinical toxicity risks were built in various profiles: Profile I for maculopathy, optic-neuropathy, and visual acuity deterioration; Profile II for neovascular glaucoma; Profile III for radiation-induced retinopathy; and Profile IV for dry-eye syndrome. RESULTS: Machine learning-based toxicity prediction accuracy for selecting the correct treatment modality was 81%, 77%, 91%, and 93% for Profiles I, II, III, and IV, respectively. CONCLUSION: The study shows that a machine learning method based on easy-to-access clinical characteristics can predict which toxicity would be greater with SRT or protons. This AI tool could support patients in making informed treatment decisions in an ophthalmology clinic, without the lengthy wait for computed tomographic simulation results and extensive plan comparisons.