AI-driven body composition atlas reveals its association with NSCLC immunotherapy outcome and molecular background: a multicenter study.

Although previous studies have linked body composition to immunotherapy efficacy, comprehensive multidimensional analyses with biological explanations remain lacking. This study integrated eight independent cohorts comprising 2,132 non-small cell lung cancer (NSCLC) patients, including five immune checkpoint inhibitor prognostic cohorts (n = 1,919), two bulk RNA-seq cohorts (n = 190), and one prospective single-cell RNA-seq cohort (n = 23). Using deep learning algorithms, we automatically extracted 92 body composition parameters from computed tomography images. The AI-based segmentation system demonstrated high consistency with manual measurements (intraclass correlation coefficient >0.87) with significantly improved efficiency. In male patients, higher intermuscular fat volume (IMFV) and 14 other indicators were independent predictors of overall survival; in female patients, T12 subcutaneous fat density and 6 other indicators showed potential associations with survival. Male patients with high IMFV exhibited significant upregulation of interferon-related pathways in CD8 + T cells and NK cells, along with lower exhaustion scores, while female patients with high T12 subcutaneous fat density showed macrophage polarization toward the M1 phenotype. This study underscores the importance of multidimensional body composition in NSCLC patient management, demonstrating that specific parameters are not only closely related to survival outcomes but also exhibit unique gender differences and location variations, providing new insights for optimizing immunotherapy strategies.