Two-level classification for differential diagnosis and molecular subtype classification of pediatric medulloblastoma from other posterior fossa tumors

Brain tumors are the most common solid malignancies in the pediatric population, with medulloblastoma being the most prevalent malignant brain tumor in children. Despite advances in multimodal treatment, which typically involves surgery, chemotherapy, and radiation, survivors often face significant long-term side effects. Recent molecular studies have classified medulloblastoma into four major subgroups: WNT, SHH, Group 3 and Group 4, each with distinct clinical and biological characteristics. Accurately identifying medulloblastoma from brain tumors and differentiating the specific molecular subtype are crucial for developing targeted and effective treatment strategies. This study aims to propose a twolevel classification schema to enhance the accuracy of differential diagnosis and molecular classification of medulloblastoma using machine learning algorithms applied to radiomic features extracted from MRI scans. Our dataset comprised samples from patients with posterior fossa tumors, including a significant subset of medulloblastoma cases. We employed feature extraction and machine learning classification to develop a robust pipeline for both differential diagnosis and molecular subtype classification. The differential diagnosis model demonstrated strong performance in accurately distinguishing medulloblastoma from other tumor types, reaching a balanced accuracy over 85% in both training and testing dataset. Additionally, the molecular subtype classification showed high efficacy, reaching a balanced accuracy over 80% for all subtype cohorts. These findings underscore the potential of machine learning to improve diagnostic precision and enable personalized treatment approaches for medulloblastoma.