Population-Scale Liver Volume Nomograms: Data-Driven Insights from High-Throughput Imaging and Clinical Data Integration

BackgroundLiver volume is a potentially useful biomarker for clinical and research applications. However, comprehensive population-scale understanding of the association between liver volume and disease states is lacking. PurposeTo identify key factors influencing liver volume in a large and diverse cohort using a data- driven method. Materials and MethodsThis retrospective study included 78,983 patients who underwent abdominal CT and MR imaging within a single health system between 2014 and August 2024. Images were processed using an automated pipeline that employed a nnUNet model for liver segmentation and a ResNet-based quality control module. Clinical data, including demographic, physiological, diagnosis codes, and lab results within three months of imaging were integrated. Feature selection utilized LASSO regression, and predictive models were validated across a reserved test set and a temporal validation cohort. ResultsEight variables associated with liver volume were identified: body surface area (BSA), weight, age, liver steatosis, race (Black or African American), smoking status, nicotine dependence, cigarettes (uncomplicated), and fever (unspecified). Nomograms derived from these predictors achieved root mean squared errors (RMSE) of 374 mL to 413 mL (Z-score 0.70-0.78) and correlation coefficients (R2) of 0.41-0.44 across datasets. Adjusted models incorporating diagnostic codes and lab results consistently outperformed demographic-only models. ConclusionOur findings determined the association between patient characteristics and liver volume, and establish nomograms that may be applicable across populations. The identification of these factors advances our understanding of the drivers of liver volume. SummaryOur study leverages high-throughput analysis of imaging and clinical data to identify key predictors of liver volume, providing robust nomograms for accurate prediction and improving understanding of liver volumetry. Key ResultsO_LIFrom analysis of 145,165 abdominal scans, we identified eight key predictors of liver volume, with BSA (668 mL), liver steatosis (+268 mL), and unspecified fever (+150 mL) showing the strongest associations. C_LIO_LIA model incorporating these factors had moderate performance for predicting liver volume (R2=0.41-0.44, RMSE=374.07-405.7) in test sets. C_LIO_LIA simplified model based on five "easily observable" factors had similar performance for predicting liver volume (R2, RMSE) in the same test sets. C_LI