MRI‐Based DeltaHabitat Radiomic Model Predicts Pathological Complete Response in Oral Cavity Cancer Treated With Neoadjuvant Chemoimmunotherapy

Objective This study aimed to develop and validate an MRI‐based DeltaHabitat radiomics model to predict pathological complete response (pCR) in oral cavity squamous cell carcinoma (OCSCC) patients treated with neoadjuvant chemoimmunotherapy. Methods Patients treated with neoadjuvant chemoimmunotherapy and surgery were retrospectively included from one institution and randomly divided into training and test cohorts using a 7:3 ratio. The region of interest (ROI) for the primary tumor was manually delineated on contrast‐enhanced T1‐weighted MRI, and radiomic features were extracted. The volume of interest was segmented into three subregions using the K‐means clustering algorithm. Following feature selection, five models were constructed to predict pCR in both the training and test cohorts. The efficacy of the models was evaluated using receiver operating characteristic (ROC) curves, calibration curves, and decision curve analysis (DCA). Results One hundred and ninety‐five patients were enrolled. The median age was 57 years, and 127 (65.1%) patients were male. Features were extracted from three separate regions, and a total of 5502 features were yielded. After the feature selection process, 12 features were retained. Among radiomic models, the DeltaHabitat model demonstrated a satisfactory area under the receiver operating characteristic curve (AUC) in both the training and test cohorts (0.923, 95% CI: 0.880–0.967; 0.878, 95% CI: 0.791–0.964, respectively). Conclusion MRI‐based DeltaHabitat radiomics model demonstrated good performance in predicting pCR in OCSCC patients treated with neoadjuvant chemoimmunotherapy. This non‐invasive approach may facilitate early identification of responders and support individualized treatment decision‐making.