Mri-based habitat and peritumoral radiomics for predicting the proliferative activity of stromal cells in giant cell tumor of bone.

Purpose: This study aims to explore the feasibility of MRI-based habitat and peritumoral radiomics for predicting the proliferative activity of stromal cells in giant cell tumor of bone (GCTB) Material and methods: A retrospective study was performed on 133 patients (102 in training cohort and 31 in validation cohort) diagnosed with GCTB from four centers. The tumor was meticulously segmented into three distinct habitat subregions using K-means clustering, incorporating a 1-pixel peritumoral expansion to capture the microenvironments surrounding the tumor. After feature extraction and selection, habitat, intratumoral and peritumoral models integrating three different machine learning classifiers were constructed respectively to identify GCTB patients with high and low proliferation. The performance of the models was assessed by receiver operating characteristic (ROC) curves, calibration plots, and decision curve analysis (DCA). SHAP analysis was utilized to enhance model interpretability. Results: Among the eligible patients, 43 (32.3 %) diagnosed with high proliferative activity of stromal cells in GCTB by pathological diagnosis. Among all models tested in the validation cohort, the Logistic Regression (LR) algorithm for habitat model exhibited superior performance in the validation cohort (AUC: 0.956, 95 % CI: 0.887-1.000). The calibration curves and DCA exhibited fit for the habitat model while providing great clinical net benefit. Conclusion: MRI-based habitat radiomics had the potential to predict the proliferative activity of stromal cells in GCTB. This model may help determine optimal treatment strategies and improve patient outcomes.