Prognostic value of automated computed tomography body composition in hepatocellular carcinoma: a volumetric versus single-slice comparison
Abstract
Background/Aims To compare the prognostic value of fully automated volumetric versus single-slice computed tomography (CT)-derived body composition analysis for predicting overall survival in patients with newly diagnosed hepatocellular carcinoma (HCC). Methods In this retrospective study of 448 patients with newly diagnosed HCC, deep-learning algorithm quantified volumetric (L1–L5) and single-slice (L3) skeletal muscle and adipose tissue metrics using CT images. Spearman correlation assessed the relationship between volumetric and single-slice body composition metrics. Cox proportional hazards models, adjusted for established clinical prognostic factors, were used to evaluate association between body composition parameters and overall survival. Harrell’s C-index was used to compare prognostic performance of Cox models. Results Strong correlations (correlation coefficient ≥0.82) were observed between volumetric and single-slice measurements for muscle attenuation, visceral fat index, and subcutaneous fat index, while muscle index demonstrated only moderate correlation (correlation coefficient= 0.50). Single-slice low muscle index (hazard ratio [HR], 1.63; 95% confidence interval [CI], 1.15–2.31; P=0.006), single-slice low muscle attenuation (HR, 1.38; 95% CI, 1.01–1.88; P=0.045) and volumetric low muscle attenuation (HR, 1.42; 95% CI, 1.03–1.95; P=0.034) were identified as independent predictors of overall survival. Median follow-up was 6.6 years. The single-slice and volumetric models showed comparable discrimination (C-index, 0.816 vs. 0.817; P=0.11). Conclusion Automated single-slice muscle parameters at the L3 level provide significant prognostic stratification in patients with hepatocellular carcinoma, establishing it as a viable, time-efficient tool for routine oncologic practice. In contrast, automated volumetric body composition assessment did not provide superior discrimination compared with the single-slice approach.