AI-driven quantitative assessment of abdominal organs and immunotherapy outcomes across multicenter hepatocellular carcinoma cohorts

Host-related imaging phenotypes beyond conventional muscle and fat measurements remain insufficiently explored in hepatocellular carcinoma (HCC). This study evaluated associations between abdominal organ features and outcomes in 1512 patients with HCC receiving immune checkpoint inhibitor-based therapy across 10 cohorts from six centers. Deep learning automatically quantified nine abdominal organs from pretreatment CT or MR images, yielding 18 volume and height-normalized indices. Continuous organ measurements were assessed using multivariable Cox regression with Benjamini–Hochberg false discovery rate correction; exploratory external validation used WHUH-derived median cutoffs. The primary outcomes were overall survival (OS) and progression-free survival (PFS). In the primary analysis, organ measurements were evaluated as continuous variables using multivariable Cox regression, followed by Benjamini–Hochberg false discovery rate (FDR) correction. Exploratory external validation used median cut-offs derived from the WHUH cohort and fixed in the remaining centers. Liver volume, left adrenal gland volume, and right kidney index were associated with overall survival after FDR correction. No continuous organ metric remained significant for progression-free survival, whereas higher right kidney volume and right kidney index were associated with shorter progression-free survival in exploratory external validation. Automated multi-organ quantification from routine imaging may provide complementary prognostic information without requiring additional examinations or radiation exposure.