An ultrasound multiparametric method to quantify liver fat using magnetic resonance as standard reference

Laura De Rosa,Antonio Salvati,Nicola Martini,Dante Chiappino,Simone Cappelli,Marcello Mancini,Libertario Demi,Lorenzo Ghiadoni,Ferruccio Bonino,Maurizia R. Brunetto,Francesco Faita
DOI: https://doi.org/10.1111/liv.16078
IF: 8.754
2024-08-30
Liver International
Abstract:Background & Aims There is an unmet need for a reliable and reproducible non‐invasive measure of fatty liver content (FLC) for monitoring steatotic liver disease in clinical practice. Sonographic FLC assessment is qualitative and operator‐dependent, and the dynamic quantification range of algorithms based on a single ultrasound (US) parameter is unsatisfactory. This study aims to develop and validate a new multiparametric algorithm based on B‐mode images to quantify FLC using Magnetic Resonance (MR) values as standard reference. Methods Patients with elevated liver enzymes and/or bright liver at US (N = 195) underwent FLC evaluation by MR and by US. Five US‐derived quantitative features [attenuation rate(AR), hepatic renal‐ratio(HR), diaphragm visualization(DV), hepatic‐portal‐vein‐ratio(HPV), portal‐vein‐wall(PVW)] were combined by mixed linear/exponential regression in a multiparametric model (Steatoscore2.0). One hundred and thirty‐four subjects were used for training and 61 for independent validations; score‐computation underwent an inter‐operator reproducibility analysis. Results The model is based on a mixed linear/exponential combination of 3 US parameters (AR, HR, DV), modelled by 2 equations according to AR values. The computation of FLC by Steatoscore2.0 (mean ± std, 7.91% ± 8.69) and MR (mean ± std, 8.10% ± 10.31) is highly correlated with a low root mean square error in both training/validation cohorts, respectively (R = 0.92/0.86 and RMSE = 5.15/4.62, p
gastroenterology & hepatology
What problem does this paper attempt to address?