Enhancing prediction of pathological complete response to neoadjuvant immunotherapy in non-small cell lung cancer by whole-lung radiomics

To evaluate the utility of whole-lung radiomics for predicting pathological complete response (pCR) to neoadjuvant immunotherapy in patients with non-small cell lung cancer (NSCLC), and further explore whether integrating whole-lung radiomics with tumor-specific radiomics could enhance the predictive performance for pCR. This dual - center retrospective study enrolled stage II-IIIB NSCLC patients receiving neoadjuvant immunotherapy. A total of 284 patients from Center 1 (August 2018–June 2025) were split into training and internal test sets via stratified randomization (7:3 ratio, balanced by pCR status). Another 82 patients from Center 2 (April 2023–March 2026) formed the external test set. The primary tumoral (PT) and whole-lung (WL) radiomic features were extracted from baseline CT images. Two independent LightGBM-based radiomic models (PT model and WL model) were established, and their predicted probabilities were fused to develop the combined PTWL radiomic model. Model discriminative performance was assessed using receiver operating characteristic (ROC) curve, and recurrence-free survival (RFS) outcomes were evaluated via Kaplan-Meier survival analysis. The overall cohort included 366 patients (median age [IQR], 63 [58–67] years; 317 males), among whom 151 (41.3%) achieved pCR after neoadjuvant immunotherapy. The WL radiomic model exhibited comparable performance to the PT model (internal AUCs: 0.676 vs. 0.701; external AUC 0.694 vs. 0.732). The PTWL model achieved higher AUCs (0.788, internal test set; 0.787, external test set). Integrating carcinoembryonic antigen (CEA) into the PTWL signature yielded a combined model with further elevated positive predictive value (PPV), which reached 0.769 and 0.636 in the internal and external test sets, respectively. A hierarchical decision-making framework was established, in which the PTWL model completed initial risk stratification and the combined model acted as the secondary confirmation for tiered clinical decision-making. Patients classified as low-risk by this hierarchical framework exhibited significantly prolonged RFS (log-rank P < 0.05). The proposed whole-lung radiomic approach demonstrates potential for predicting pCR in patients with NSCLC receiving neoadjuvant immunotherapy. The PTWL model conducts preliminary risk stratification, and the combined model strengthens pCR confirmation. This hierarchical decision-making framework facilitates individualized neoadjuvant treatment by reinforcing pCR identification.