Deep learning based on preoperative MR to differentiate germinomas from nongerminomatous cell tumors.

Abstract

19 Background: Intracranial germ cell tumors (iGCTs) are highly malignant brain tumors that predominantly occur in children. Two types of iGCTs exist: germinomas (GEs) and non-germinomatous germ cell tumors (NGGCTs), each having different treatments and prognoses. However, predictive biomarkers for differentiation are limited. Methods: This multicenter Development-validation set study included patients with iGCTs with pathology-confirmed for the developed dataset (n = 280), which was split into training (n = 200) and testing sets (n= 80), and three independent validation sets (n = 64), respectively. Patients were included between January 2010 and December 2021. We developed a DL model based on 3D nnU-Net and preoperative T2-weighted images to differentiate GE from NGGCT, and we compared the model performance with the diagnoses made by five physicians of different majors and experiences with and without referring to the DL model’s output. Data analysis was performed from February 2021 to May 2022.We calculated the diagnostic performances of the model and physicians using the area under the curve (AUC), accuracy, sensitivity, and specificity. Results: The DL pipeline showed an average AUC of 0.837 in the testing set and three independent validation sets, superior to that using conventional information alone (combine with MR features and clinical information, AUC=0.768, p=0.009). The physicians’ diagnoses showed an average AUC of 0.695 without reference to the DL output and 0.842 concerning the DL output in the testing and prospective internal validation sets, with a 21.15% improvement (p=0.014). Conclusions: We developed and validated a DL model using a large dataset to accurately differentiate GE from NGGCT using preoperative T2-weighted images, which is relatively easily incorporated into the workflow of clinical diagnosis at minimal additional cost, and may enhance precision therapy.