Deep Learning Model to Differentiate Intracranial Germ Cell Tumors Subtypes and Predict Survival

Abstract

Background Pretherapeutic differentiating subtypes of primary intracranial germ cell tumors (iGCTs), including germinomas (GEs) and non-germinomatous germ cell tumors (NGGCTs) is essential for clinics because of their distinct treatment strategies and prognosis profiles. This study aimed to develop a deep learning model, iGNet, to assist in differentiation of iGCT subtypes by employing pretherapeutic MR images.Methods The iGNet model was developed using a retrospective dataset of 280 pathology-confirmed iGCT patients, including 83 GE and 117 NGGCTs in train dataset, and 31 GEs and 49 NGGCTs in the retrospective internal test dataset. The model's diagnostic performance was then tested through area under the receiver operator characteristics curve (AUC) in a prospective internal dataset (n = 22) and two external datasets (n = 22 and 20). Next, we compared the diagnostic performance in six neuroradiologists with or without the assistance of iGNet. Lastly, the predictive ability of the iGNet outputs for progression-free and overall survival was assessed in comparation with pathological diagnosis.Results iGNet achieved high diagnostic performance with AUCs between 0.869 and 0.950 across the four test datasets. With the assistance of iGNet, neuroradiologists' diagnostic AUCs (average of the four test datasets) were increased by 9.22–17.90% across six neuroradiologists. The iGNet output can predicting the progression-free and overall survival, comparable to that based on pathological diagnosis (P = .889).Conclusions The iGNet, leveraging pretherapeutic MR imaging, accurately differentiates iGCT subtypes, thereby facilitating clinical stratified treatment and prognostic evaluation.