Radiomics Based on the Tumor–Parenchyma Invasive Interface Predicts Major Pathological Response to Neoadjuvant Immunochemotherapy in Non-small Cell Lung Cancer

Background Accurate prediction of tumor response to neoadjuvant immunochemotherapy (NAIC) enables personalized perioperative therapy for resectable non-small cell lung cancer (NSCLC). Objective The present aimed to evaluate the predictive value of radiomics derived from the tumor-parenchyma invasive zone for response to NAIC in resectable NSCLC, with the goal of developing a more accurate and clinically applicable model. Methods Patients with pathologically proven NSCLC from August 2019 and March 2025 were retrospectively included from two medical centers. In the training set, radiomics features were extracted from the whole tumor region and tumor margin region (6mm) respectively. Following feature selection via intraclass correlation coefficient and least absolute shrinkage and selection operator, the Whole Tumor Model (WTM) and Tumor Margin model (TMM) were developed to non-invasively predict major pathological response (MPR) following NAIC. The performance was evaluated using the area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, specificity, positive predictive value, and negative predictive value in the internal validation and external test sets. The optimal radiomics model and clinical characteristics were combined to build the hybrid model (HM). Results A total of 169 patients (median age, 60 years; 154 men) were divided into training, internal validation and external test sets, with 104 patients (61.5%) achieving MPR. In the test dataset, WTM and TMM achieved AUCs of 0.71 (95% CI: 0.54–0.89) and 0.84 (95% CI: 0.71–0.97), respectively. After incorporating tumor margin radiomics features and clinical predictors(pathology), the HM demonstrated satisfactory performance in the training set (AUC: 0.88, 95% CI: 0.81–0.95) and internal validation set (AUC: 0.86, 95% CI: 0.74–0.98). In the independent external test set, the HM obtained satisfactory performance (AUC = 0.87, 95% CI: 0.76–0.98). Decision curves analysis indicated that the radiomics-clinical combined nomogram provided significant clinical utility. Conclusion A radiomics model based on the tumor margin region outperformed the whole-tumor model in predicting MPR in NSCLC. Our study developed a novel tool to predict the response of NSCLC to NAIC, which demonstrated excellent performance.