MRI Radiomics Signature of Pediatric Medulloblastoma Improves Risk Stratification Beyond Clinical and Conventional MR Imaging Features

Abstract

Background: Prognostic evaluation is important for personalized treatment in children with medulloblastoma (MB). Limited data are available for risk stratification using a radiomics-based model. Purpose: To evaluate the incremental value of an MRI radiomics signature in stratifying the risk of pediatric MB in terms of overall survival (OS). Study type: Retrospective. Subjects: A total of 111 children (mean age 5.82 years) with pathologically confirmed MB divided into training and validation cohorts (77 and 34 children, respectively). Field Strength/Sequence: A 3 T, contrast-enhanced T1-weighted imaging with inversion recovery. Assessment: The study endpoint was OS defined as the time between the preoperative MRI study and death or last follow-up. The radiomics signature model and a clinical-MRI model were developed for personalized OS prediction. An integrative model, which combined the radiomics signature and clinical-MRI features, was also built using multivariable Cox regression model. The performance of the three models was evaluated with the C-index. The performance of integrative model was assessed by calibration curve and decision curve analysis (DCA). Statistical Tests: Independent T-test, Mann-Whitney U test, Fisher's exact tests or chi-square test, logistic regression analysis, Kaplan-Meier survival analysis, C-index, intraclass correlation coefficients (ICC). P < 0.05 was considered statistically significant. Results: The media OS was 2.83 years (3.87 AE 1.85 years). Two clinical and one conventional MR imaging features (remnant, adjuvant treatment, and peritumoral edema) were selected for clinical-MRI model building. The integrative model evaluated OS (C-index 0.823) better than either the radiomics signature (C-index 0.702) or the clinical-MRI model (C-index 0.771). And it also showed good performance in the validation cohort (C-indices: 0.786, 0.756, 0.721), which was validated by the good calibration (P > 0.05) and more benefit. Data Conclusions: This study demonstrated that the integrative model, which combined radiomics signature, clinical, and conventional MRI features, showed best performance in OS evaluation for children with MB. The radiomics signature may confer incremental value over clinical-MRI features. Evidence Level: 3. Technical Efficacy: Stage 2.

Opportunities and Advances in Radiomics and Radiogenomics for Pediatric Medulloblastoma Tumors

“…Pre-processing is then followed by segmenting the region of interest (i.e., the tumor and its compartments of interest) from the imaging modalities [41][42][43][44][45]. Following tumor segmentation, radiomic analysis is performed, which involves extracting the different feature classes (e.g., texture, shape, size, structural deformations) from the tumor compartments [28][29][30][31][32][33][34][35][36][37][38][39][40]. Typically, the feature extraction stage is followed by some operations for pruning and reduction of the feature sets to remove redundant, highly correlated features.…”

“…Typically, the feature extraction stage is followed by some operations for pruning and reduction of the feature sets to remove redundant, highly correlated features. This can be conducted using different statistical approaches, such as logistic regression [29,31,32], minimum redundancy, maximum relevance [33], Pearson's correlation coefficient [30,39], and principal component analysis [28]. The set of selected features is then fed into different machine learning and statistical models that pertain to a specific application.…”

“…Performance metrics are finally employed to assess the model's ability to perform the designated task, as well as its generalizability, to be applied in different clinical applications. There is a wide variety of performance metrics that could be included, from Area Under the Curve (AUC) [33,34,40], to Kaplan-Meier curves [28][29][30][31][32] for survival analysis, hazard ratios, etc. ignated task, as well as its generalizability, to be applied in different clinical applications.…”

“…ignated task, as well as its generalizability, to be applied in different clinical applications. There is a wide variety of performance metrics that could be included, from Area Under the Curve (AUC) [33,34,40], to Kaplan-Meier curves [28][29][30][31][32] for survival analysis, hazard ratios, etc.…”

“…Secondly, we provide a detailed literature review of the radiomic and radiogenomic approaches that have thus far been explored for pediatric MB. We have grouped these studies by their endpoint, resulting in three main categories: (1) applications in survival prognostication [28][29][30][31][32], (2) applications in molecular subgroup classification [32][33][34][35][36][37][38][39][40] (i.e., radiogenomics), and (3) MB tumor segmentation [41,42]. Finally, we address the current challenges and future directions pertaining to applications of radiomic and radiogenomic approaches in pediatric MB tumors.…”