MRI Based Intratumoral-Peritumoral Habitat Radiomics for Prediction of Overall Survival in Rhabdomyosarcoma: A Multicenter Study.

RATIONALE AND

OBJECTIVES

This study aimed to develop and validate an MRI-based habitat radiomics model integrating intratumoral and peritumoral heterogeneity for predicting overall survival (OS) in rhabdomyosarcoma (RMS).

MATERIALS AND

METHODS

This retrospective study included 486 patients with histologically confirmed RMS, divided into training (n = 289), validation (n = 124), and test (n = 73) cohorts. All patients underwent standardized MRI scans, and T1 contrast-enhanced sequences were used for feature extraction. The tumor was segmented into intratumoral and peritumoral regions, and voxel-wise clustering using K-means was applied to identify subregions or "habitats" with similar imaging characteristics. A total of 1762 radiomic features were extracted from these regions and subregions, including texture, shape, and fractal features. Feature selection was performed using Spearman correlation, univariate Cox regression, and LASSO-Cox regression. Prognostic models were constructed using multivariate Cox proportional hazards models. The models were evaluated using concordance index (C-index), time-dependent AUC, Kaplan-Meier survival analysis, calibration curves, and decision curve analysis (DCA).

RESULTS

The habitat-based models significantly outperformed conventional radiomics models. The intratumoral habitat model (Intra_Habitat) achieved a C-index of 0.876 in the training cohort, surpassing the whole-tumor model (C-index: 0.823). Similarly, the peritumoral habitat model (Peri_Habitat) demonstrated superior performance (C-index: 0.876). The integrated intratumoral-peritumoral habitat model (IntraPeri_Habitat) showed the best overall prognostic performance across all cohorts, with a C-index of 0.876 in the training cohort and 0.770 in the external test cohort. Time-dependent AUC for 5-year overall survival further confirmed its robust discriminative power. Kaplan-Meier analysis confirmed significant stratification between high- and low-risk groups (log-rank P < .001). Calibration curves showed excellent agreement between predicted and observed survival outcomes and DCA demonstrated superior net clinical benefit.

CONCLUSION

MRI-based habitat analysis offers a refined method to capture tumor heterogeneity and provide insights into survival outcomes in RMS patients. The integrated intratumoral-peritumoral habitat model demonstrates promising prognostic value, potentially aiding personalized treatment strategies in the future.