Predicting postoperative recurrence and survival in glioma patients using enhanced MRI-based delta habitat radiomics: an 8-year retrospective pilot study.

Abstract

Objective: This study aimed to develop predictive models for postoperative recurrence and overall survival in patients with brain glioma (BG) by integrating preoperative contrast-enhanced MRI-derived delta habitat radiomics features with clinical characteristics.

Methods: In this retrospective study, preoperative contrast-enhanced MRI data and clinical records of 187 BG patients were analyzed. Patients were stratified into non-recurrence (n = 100) and recurrence (n = 87) cohorts based on postoperative outcomes. The dataset was randomly divided into training and test sets (7:3 ratio). Delta habitat radiomic features were extracted from intratumoral and peritumoral edema regions. A radiomic score (Radscore) was generated via LASSO regression with ten-fold cross-validation in the training cohort. Clinical variables (gender, IDH1 mutation, 1p19q co-deletion, MRI enhancement patterns) and radiomic features were compared between groups using χ² or Student's t-tests. Multivariate logistic regression models incorporating significant predictors were developed. Model performance was evaluated using AUC comparisons (DeLong test), decision curve analysis (clinical utility), and validated via XGBoost machine learning. Nomograms were constructed to visualize recurrence and survival predictions.

Results: The training cohort revealed significant intergroup differences in gender, IDH1 mutation, 1p19q co-deletion, MRI enhancement patterns, and delta habitat radiomic scores (Radscore1/2, p < 0.05). The combined model (clinical + radiomic features) demonstrated superior predictive performance for recurrence [AUC 0.921 (95% CI 0.861-0.961), OR 0.023, sensitivity: 87.18%, specificity: 82.03%] compared to clinical-only [AUC 0.802 (0.745-0.833), OR 0.036] and radiomic-only [AUC 0.843 (0.769-0.900), OR 0.034] models (p < 0.05, DeLong test). Decision curve analysis confirmed greater clinical net benefit for the combined model. These findings were replicated in the test cohort. The survival nomogram incorporated IDH1 mutation status, gender, and Radscore1/2, with Kaplan-Meier analysis verifying their prognostic significance (p < 0.01).

Conclusion: Delta habitat radiomics derived from preoperative contrast-enhanced MRI may enhance the accuracy of postoperative recurrence and survival predictions in BG patients. The validated nomograms provide actionable tools for optimizing postoperative surveillance and personalized clinical decision-making.