RadxTools for assessing tumor treatment response on imaging

点击翻译按钮获取中文摘要

ABSTRACT: Medulloblastoma (MB) is a malignant, fast-growing pediatric brain tumor with heterogenous outcomes and a 5-year survival rate of 70-80%. Current treatment strategies for MB patients include surgical resection, chemotherapy, and craniospinal irradiation (CSI), with dose-intensification in high-risk patients (defined as residual tumor >1.5 cm2, evidence of leptomeningeal metastases, or large-cell/anaplastic histology) to improve clinical outcomes, while de-escalation of therapy to reduce long-term sequelae in standard-risk MB patients. Unfortunately, this treatment protocol has only proven useful as a rough guide for predicting prognosis with the existing clinical stratification; particularly, the 5-year survival rate for the high-risk patients are currently at about 60%. Additionally, the existing clinical risk stratification fails to identify about 20–30% of standard-risk patients who might be overtreated and eventually suffer from long-term morbidities that significantly affect their quality of life. Consequently, there is a critical need for reliable tools to risk-stratify MB patients based on their survival, with the goal of identifying high-risk MB cases who are most likely to receive added benefit from adjuvant and concomitant therapy, while de-escalating therapy in low/standard-risk cases. Through an ongoing NCI U01 award (1U01CA248226-01) from the Informatics Technology for Cancer Research (ITCR), our group has been leading the development of peri-tumoral (Eur. Rad 20172, AJNR 20183) and intra-tumoral spatial heterogeneity radiomics, that go beyond texture, shape-based approaches, for characterization of adult tumors. As an extension to our U01 efforts, in this supplemental project, we propose to develop two informatics modules for (1) radiomic analysis for tumor characterization on clinical MRI scans (Gd-T1w, T2w, FLAIR), and (2) a risk-stratification module, for survival risk-stratification of pediatric MB patients. In our preliminary work, we demonstrated that our biophysical deformation descriptor that characterizes subtle changes in vasodilation from brain parenchyma on Gd-T1w MRI scans, had higher concordance-index (C- index) in predicting overall survival in MB patients compared to employing the molecular subgroup-based stratification [n=89, p<0.05 vs. p=0.6, C-index=0.831 vs. 0.80]. The 2 modules developed in our supplement project will be leveraged to improve on our initial model (using deformations alone), to (a) include features relating to (1) 3D topology, (2) localized entropy, and (3) peri-tumoral features from the vicinity of the tumor and (b) perform MRI-based risk-stratification of MB patients based on their survival characteristics, independent of molecular stratification. Our collaborative efforts with Children's Hospital Cinncinati, Nationawide Childrens Columbus, and Children's Brain Tumor Network, will led to creation of a rich, one-of-the-largest MB cohorts for the pediatric cancer community and will serve as the foundation of ongoing and future studies focused on resolving MB aetiology. Following successful completion, we will make the multi-institutional studies and the associated mRRisc features publicly available by leveraging ongoing efforts through ITCR.