A New Risk Stratification System Based on MRI Features and Machine Learning to Predict Differential Prognosis for Pediatric Medulloblastoma Patients

Ziyang Liu,Heng Zhang,Chiyi Liao,Zhiming Liu,Xu An,Jian Cheng,Chunde Li,Jian Gong,Haijun Niu,Zhenzhou Wu,Jing,Zixiao Li,Tao Liu,Yongji Tian

DOI: https://doi.org/10.2139/ssrn.4030558

2022-01-01

SSRN Electronic Journal

Abstract:Background: Medulloblastoma (MB) is highly malignant and impairs the health and longevity of children. The current risk stratification was based on gene characteristics and pathological performance. Magnetic resonance imaging (MRI) was a good tool to depict the tumors’ shape features. However, it was not brought into the risk stratification system.Methods: We retrospectively sampled 139 MB patients who were treated in a single-center Beijing Tiantan hospital. A two-step model was proposed for molecular grouping and prognosis prediction. The MRI features were extracted for molecular groups by a novel multi-modality convolutional neural network called Bi-ResNet-MB. Then, it was to establish a prognosis prediction model to predict the probability of survival state (i.e., Machine learning-based Medulloblastoma Risk Score, M 2 R Score) using the MR features extracted by Bi-ResNet-MB. Finally, we established a novel risk stratification system to stratify the patients based on the M 2 R Score for MB patients. Findings: Bi-ResNet-MB achieved excellent performance in molecular classification, with AUCs of 0·985 for the WNT-activated, 0·958 for the SHH-activated group, 0·904 for group 3, and 0·944 for group 4. For the prognostic prediction, our machine learning model achieved excellent performance at three time points (1 year, 3 years, and 5 years) with AUC values of 0·927, 0·913, and 0·906. Based on M 2 R Score, the new risk stratification strategy clustered individual subjects with significant differences in their survival time (P < 0·001).Interpretation: Machine learning-based risk stratification system had potential in assessing medulloblastoma patients’ prognosis. Individuals at higher risk of recurrence or mortality may benefit from optimized management.Funding Information: This work was supported by Natural Science Foundation of Beijing Municipality (Z210010), Capital’s Funds for Health Improvement and Research (2020-2-2047), Beijing Natural Science Foundation (Grant No. 7172041 and Z200016), and National Natural Science Foundation of China (Grant No. 81871434 and 61971017). Declaration of Interests: The authors have declared that no competing interest exists.Ethics Approval Statement: This study is a retrospective cohort study approved by the Ethics Committee of Beijing Tiantan Hospital (Approval Number: KY2018-020-01).