Abstract
BACKGROUND: Neoadjuvant chemoimmunotherapy (NACI) shows promise in oral squamous cell carcinoma (OSCC), but reliable noninvasive biomarkers for predicting pathologic complete response (pCR) remain scarce. Radiomics integrating intratumoral and peritumoral heterogeneity across multi-sequence MRI may offer novel insights into treatment response evaluation. METHODS: The data of 212 patients with OSCC after NACI were retrospectively collected and analyzed. Among these patients, 56 (26.4%) achieved pCR after NACI. Intratumoral and peritumoral habitat imaging (HI) was achieved using the K-means clustering algorithm applied to T1-weighted imaging (T1WI), axial T2-weighted imaging with fat suppression (T2WI), and contrast-enhanced T1-weighted imaging with fat suppression (T1C). Moreover, intratumoral and peritumoral HI models were constructed and compared using the receiver operating characteristic curve (ROC). Five-fold cross-validation was performed to mitigate model overfitting. RESULTS: Intratumoral HI models derived from different sequences and the Intratumoral Fusion model exhibited favorable predictive ability, with AUCs of 0.738-0.817 and 0.729-0.789 in the training and testing cohorts, respectively. Moreover, peritumoral HI models displayed marginally higher predictive abilities compared to intratumoral HI and Fusion models, with AUCs of 0.734-0.869 and 0.788-0.802 in the training and testing cohorts, respectively. Meanwhile, the decision model with peritumoral habitat features (PHF S ), intratumoral habitat features (IHF S ), and 3 clinical features displayed the highest performance, with average AUCs of 0.913 and 0.843 in both respective cohorts. Among the most important features screened by SHAP, three IHFs and one PHF could effectively distinguish between the lower and higher groups of programmed cell death ligand 1 (PD-L1) Combined Positive Score (t = 2.027-2.275, P < 0.05), whilst two PHFs were highly correlated with CD45+ white blood cell densities in the stroma (r = 0.958, -0.920, P < 0.05), which were associated with pCR. CONCLUSIONS: Integrated intratumoral and peritumoral HI derived from multi-sequence MRI offers a high predictive capacity for pCR following NACI in OSCC patients.