Artificial intelligence-based assessment of tumor bed stroma in non-small cell lung cancer after neoadjuvant therapy: Association with pathologic response and survival. A multicenter study.

Histopathologic assessment of the tumor bed following neoadjuvant therapy in non-small cell lung cancer (NSCLC) is increasingly relevant, but comparative data across treatment modalities remain limited. We evaluated tumor bed features and artificial intelligence (AI)-assisted stromal quantification in advanced NSCLC, comparing patients treated with immunotherapy, tyrosine kinase inhibitors (TKIs), and chemotherapy. This multicenter retrospective study included 71 patients with stage IIIB-IV NSCLC who underwent salvage surgery after neoadjuvant therapy at five Italian centers between 2018 and 2022. Thirty-eight patients received immune-based therapy, 16 received TKIs, and 17 received chemotherapy alone. Tumor bed response was assessed according to International Association for the Study of Lung Cancer recommendations, including residual viable tumor, necrosis, fibrosis, and inflammation. Immune-related features were recorded. AI-assisted morphometric analysis quantified fibrosis on Azan-Mallory staining and inflammatory burden on CD45 immunohistochemistry. The immunotherapy group showed the most favorable regression profile, with higher pathological complete response rates than the TKI and chemotherapy groups, respectively (40.5% vs 21.4% vs 11.8%), and lower residual viable tumor burden (median, 5% vs 45% vs 45%). This group also showed a more immune-reactive tumor bed phenotype and significantly higher AI-quantified inflammatory burden (p = 0.022). In the overall cohort, multivariable analysis identified residual viable tumor percentage (HR, 1.02; p = 0.018) and AI-derived fibrosis (HR, 0.97; p = 0.013) as independent predictors of recurrence. Tumor bed evaluation after neoadjuvant therapy provides prognostically relevant information beyond residual viable tumor alone. AI-assisted fibrosis quantification may complement viable tumor assessment and refine post-surgical risk stratification.