Preoperative CT Radiomics for Assessing Pathological Response to Neoadjuvant Chemoimmunotherapy in Locally Advanced...

Background/Objectives: Reliable preoperative assessment of treatment response after neoadjuvant chemoimmunotherapy in locally advanced non-small-cell lung cancer (NSCLC) remains challenging because conventional imaging may not accurately distinguish viable tumour from treatment-related fibrosis or immune-mediated changes. This study investigated whether CT-derived radiomic features combined with machine learning could improve the identification of patients achieving pathological complete response (pCR). Methods: Twenty-nine consecutive patients with stage III NSCLC who underwent surgical resection following neoadjuvant chemoimmunotherapy were retrospectively analysed. Radiomic features were extracted from preoperative CT scans and used to develop supervised machine-learning models based on Random Forest, Support Vector Machine, K-Nearest Neighbors, Multi-Layer Perceptron, and Logistic Regression algorithms. Histopathological findings after surgery served as the reference standard. Feature distributions were compared between patients with and without pCR using the Mann-Whitney U test with Bonferroni correction. Results: Surgical procedures included 19 lobectomies, 2 bilobectomies, 3 pneumonectomies, and 5 complex major resections. Pathological complete response was observed in 9 of 29 patients (31%). The proposed radiomics-based model achieved an area under the ROC curve of 0.92 (95% CI, 0.81-1.00), with 91% accuracy, 90% sensitivity, and 92% specificity (p < 0.05). All patients classified by the model as complete responders were confirmed to have pathological complete response at postoperative histological examination. Conclusions: CT-based radiomics combined with machine learning demonstrated promising performance for the preoperative prediction of pathological response after neoadjuvant chemoimmunotherapy in stage III NSCLC. Although these findings require external validation in larger prospective cohorts, this approach may support preoperative treatment assessment and surgical decision-making.