通过生境影像组学预测 TIL（肿瘤浸润淋巴细胞）并探索新辅助免疫化疗治疗肺癌的应答机制

Neoadjuvant immunochemotherapy (NAIC) induces tumor microenvironment remodeling in non-small cell lung cancer (NSCLC), presenting challenges for treatment response assessment. This study developed and validated a habitat radiomics approach for non-invasive prediction of tumor-infiltrating lymphocyte (TIL) status to evaluate NAIC response in NSCLC.

This retrospective study enrolled 238 NSCLC patients following NAIC for clinical analysis, of which 201 patients met criteria for radiomics analysis. Patients were classified into TIL-positive and TIL-negative groups based on pathological assessment. Post-treatment computed tomography (CT) images were analyzed using K-means clustering to identify tumor habitat sub-regions for radiomic feature extraction. Seven machine learning algorithms were evaluated for TIL status prediction. Model interpretability was assessed through SHapley Additive exPlanations (SHAP) analysis. Single-cell RNA sequencing (scRNA-seq) data were analyzed to compare major pathological response (MPR) and non-MPR tumor microenvironments through cell type annotation, differentiation trajectory analysis, and intercellular communication network analysis.

Pre-treatment neutrophil-to-lymphocyte ratio (NLR) showed association with pathological response in multivariable analysis. The radiomics cohort was randomly divided 7:3 into training (n = 140) and test (n = 61) sets. The Random Forest model achieved an area under the receiver operating characteristic curve (AUC) of 0.823 (95% CI: 0.694-0.932) in the test set, and the habitat radiomics model stratified patients into high and low recurrence risk groups. Single-cell analysis identified immunosuppressive features in non-responding tumors, characterized by expansion of SERPINB9 + regulatory T cells (Tregs) that regulated suppressive intercellular communication networks.

This study establishes a habitat radiomics model for non-invasive assessment of TIL status following neoadjuvant immunochemotherapy in NSCLC. The model shows reliable predictive performance and prognostic stratification capability, offering potential clinical utility for treatment response evaluation and patient selection.