Development of an Algorithm to Estimate Fat-Free Mass to Optimize Contrast Injection for Computed Tomography Imaging of the Liver

OBJECTIVES: Contrast-enhanced computed tomography (CT) is central to liver imaging. Inadequate enhancement can compromise diagnostic accuracy and impact treatment decisions. Fat-free mass (FFM) has emerged as a key predictor of liver enhancement quality, enabling personalized contrast protocols. However, direct FFM measurement is impractical in clinical settings due to equipment costs and time requirements. METHODS: We developed a simple machine learning (ML) algorithm to estimate FFM using routinely available patient characteristics, including weight, height, age, and sex. A data set of 689 patients with hepatic CT scans was used to train, validate, and test the algorithm. FFM was benchmarked against measurements from a bioelectrical impedance meter (Biotekna, Italy). Model performance was evaluated through K-fold validation, yielding metrics such as R-squared ( R2 ), mean absolute percentage error (MAPE), and root mean squared error (RMSE). Clinical validation was also performed by measuring the contrast enhancement after the algorithm was implemented using 265 cases. RESULTS: The ML model demonstrated high accuracy in predicting FFM ( R2 =0.915±0.019; MAPE=0.033±0.003). Clinical validation after model implementation in clinical practice showed optimal liver enhancement, corresponding to acceptable image quality, in 89% of cases (235), with a mean enhancement centered at 53 Hounsfield units. The model outperformed existing FFM estimation formulas, demonstrating superior accuracy and generalizability across diverse populations. CONCLUSIONS: Our ML-based FFM estimation model facilitates personalized contrast protocols, eliminating the need for expensive equipment and reducing procedural complexity. This approach optimizes liver imaging quality, enhances lesion detection, and supports treatment planning.