# A Novel Radiomic Model for Risk Stratification of Cerebral Herniation in Radiation-Induced Cystic Brain Necrosis

## TL;DR

This study develops a radiomic model using MRI scans to predict the risk of cerebral herniation in patients with radiation-induced brain necrosis.

## Contribution

A novel radiomic model combining MRI features and clinical variables for risk stratification of cerebral herniation in RCN patients.

## Key findings

- The radiomic model showed strong predictive performance with C-indices of 0.841 and 0.867 in training and testing cohorts.
- The model successfully stratified patients into high- and low-risk groups for cerebral herniation.
- Calibration and decision curve analyses confirmed the model's clinical utility and accuracy.

## Abstract

Radiation-induced cystic brain necrosis (RCN) can progress rapidly to life-threatening cerebral herniation. In this study, we identified a radiomic signature derived from baseline magnetic resonance images (MRIs) to stratify the risk of cerebral herniation in nasopharyngeal carcinoma survivors with RCN. By incorporating the radiomic signature and ratios of perilesional enhancement, a radiomic model was developed and showed favorable performance in the training and testing cohorts. Our findings demonstrate that radiomic features extracted from MRI can predict the risk of cerebral herniation in patients with RCN. The radiomic model can serve as an easy-to-use and non-invasive tool for managing patients with RCN. Specifically, patients identified as high-risk should receive more frequent imaging surveillance and clinical monitoring, with surgical intervention considered when necessary.

Background: Radiation-induced cystic brain necrosis (RCN) can progress rapidly to life-threatening cerebral herniation. This study aimed to develop a predictive model integrating radiomic features and clinical variables to assess the risk of cerebral herniation in RCN patients. Methods: A total of 130 patients diagnosed with RCN following radiotherapy for nasopharyngeal carcinoma were retrospectively enrolled and randomly assigned to training (n = 91) and testing (n = 39) cohorts in a 7:3 ratio. Radiomic features were extracted from baseline T2-weighted magnetic resonance imaging (MRI), and a radiomic signature was constructed using least absolute shrinkage and selection operator regression. A multivariate Cox regression model was then developed by incorporating the radiomic signature and clinical variables to predict cerebral herniation. The model’s discriminative ability, calibration, and clinical utility were evaluated. Results: The radiomic signature based on five selected radiomic features demonstrated good predictive performance. The radiomic model, which integrated the radiomic signature and ratios of perilesional enhancement, exhibited favorable performance in both the training cohort (C-index: 0.841) and testing cohort (C-index: 0.867). The model successfully stratified patients into high- and low-risk groups. The calibration curves showed good agreement and the decision curve confirmed the clinical utility of the model. Conclusions: The MRI-based radiomic model, which integrates radiomic features and clinical variables, demonstrates robust performance in predicting cerebral herniation in RCN patients, offering a practical and user-friendly tool to support clinical decision-making.