Elham Raiesi Nafchi - Department of Radiation Sciences, Faculty of Allied Medicine, Iran University of Medical Sciences, Tehran, Iran
Pedram Fadavi - Department of Radiation Oncology, School of Medicine, Iran University of Medical Science, Tehran, Iran
Sepideh Amiri - Department of Computer Science, University of Copenhagen, Copenhagen, Denmark
Susan Cheraghi - Department of Radiation Sciences, Faculty of Allied Medicine, Iran University of Medical Sciences, Tehran, Iran .Radiation Biology Research Center, Iran University of Medical Sciences, Tehran, Iran
Maryam Garousi - Department of Radiation Oncology, School of Medicine, Iran University of Medical Science, Tehran, Iran
Mansoureh Nabavi - Radiation Oncology Research Center (RORC), Cancer Institute, Tehran University of Medical Sciences, Tehran, Iran

Introduction: We aimed to build a machine learning-based model to predict radiation-induced optic neuropathy (RION) in patients who had treated brain and head and neck cancers with radiation or chemo radiation therapy.Methods: The visual evoked potential values were obtained in the case (۵۲ patients) and control (۵۲ patients) groups. The radiomics features were extracted from the segmented area including right and left nerve optic and optic chiasm using ۳D-Slicer software. We implemented ۵-fold cross-validation to evaluate ۵ supervised ML models Bernoulli Naive Bayes, Decision Tree, Gradient Boosting Decision Trees, K-Nearest Neighbor, and Random Forest on ۴ input datasets to predict radiation induced visual complications. The F۱ score, accuracy, sensitivity, specificity, and area under the ROC curve were the evaluation criteria.Results: RION affected ۳۱% of the patients. ۹۰۵ radiomic characteristics were extracted from each segmented area. Gradient Boosting Decision Trees was the most powerful algorithm to predict RION and had the highest AUC among the five classifiers with AUC ≥ ۹۸%. Chiasm dataset can predict RION better than right or left nerve optic or combination of features from all radiomics datasets.Conclusion: We found that combination of radiomic, dosimetric, and clinical factors can predict RION after radiation treatment with high accuracy. To acquire more reliable results, it is suggested VEP is conducted before and after radiation therapy, with multiple follow-up courses, more additional optometric tests, and more patients.

Radiation-induced optic neuropathy, Visual evoked potential, computed tomography, Machine learning, Radiation therapy, Radiomics