Predicting early recurrence of craniopharyngioma using multi-omics radiomic modeling: a retrospective cohort study.
AI interpretation is pending for this paper.
Open original publication →What the AI sees
Not AI summarized yet.
Research significance
Pending deeper interpretation.
Source abstract
Craniopharyngiomas (CPs) are benign tumors arising in the sellar and suprasellar regions, but they frequently recur after surgery because of their proximity to critical neurovascular and hypothalamic structures. This study aimed to develop and validate a habitat-based multimodal radiomics model for early postoperative recurrence prediction in CP. This retrospective single-center cohort included 106 patients with pathologically confirmed CP selected from 136 consecutive surgically treated cases. Preoperative magnetic resonance imaging (MRI) scans (T1-weighted imaging, contrast-enhanced T1-weighted imaging, and T2 -weighted imaging) were used for tumor habitat analysis based on unsupervised clustering of voxel-level radiomic features. Radiomic features were extracted from whole-tumor and habitat-defined subregions, followed by feature selection within the training cohort using univariate filtering, correlation pruning, minimum redundancy maximum relevance (mRMR), and least absolute shrinkage and selection operator (LASSO). Clinical, radiomics, habitat-based, and Combined models. Model performance was evaluated using Receiver Operating Characteristic (ROC), calibration, and decision curve analyses. Shapley Additive exPlanations (SHAP) analysis and representative habitat maps were used for model interpretability. Among the evaluated classifiers, the ExtraTrees model showed the most stable discrimination performance across the training and validation cohorts. The HabitatAll model achieved an AUC of 0.891 in both cohorts, while the Combined model showed the best overall predictive performance, with AUCs of 0.944 and 0.891 in the training and validation cohorts, respectively, and the lowest Brier scores. SHAP analysis identified T1C-derived and H3-related habitat features as major contributors to recurrence prediction. Representative habitat maps showed that H3 was frequently distributed in the peripheral portion of the tumor in both recurrent and non-recurrent cases. This study developed a promising habitat-based multimodal radiomics model for early postoperative recurrence prediction in craniopharyngioma. The Combined model demonstrated the most favorable overall discrimination and calibration. H3-related habitat features may capture intratumoral heterogeneity associated with recurrence risk. However, given the retrospective single-center design, these findings should be interpreted as preliminary and require further validation in larger multicenter cohorts.