Machine learning based on quantitative ultrasound for assessment of chronic liver disease

Abstract

Chronic liver disease (CLD) is a highly prevalent condition characterized by the coexistence of histopathological changes, including liver steatosis, inflammation and fibrosis. Based on a multi-parametric approach, the goal was to assess the ancillary value of quantitative US (QUS) parameters to point shear-wave elastography (pSWE), based on random forests, on a cohort of subjects with CLD. Ninety-one individuals were recruited in this prospective institutional review board approved study, and 82 patients were included after applying exclusion criteria. Measurements of pSWE and radiofrequency ultrasound images were acquired with a clinical scanner using a convex probe. QUS features were extracted from homodyned-K parametric maps. Total and local attenuation coefficient slopes were also included as spectral QUS features, based on reference phantom methods. Dichotomous classification of grades and stages were performed. Receiver operating characteristics (ROC) curves were estimated with bootstrapping, which yielded area under each ROC curve (AUC). The reference standard was histopathological analysis of liver biopsy specimens for grading steatosis and inflammation, and staging fibrosis. QUS parameters improved the classification of liver steatosis, inflammation, and fibrosis compared to pSWE alone. For instance, to classify liver steatosis grades 0 vs ≥ 1, ≤ 1 vs ≥ 2, ≤ 2 vs 3, respectively, AUCs increased from 0.60, 0.63, and 0.62 to 0.90, 0.81, and 0.78, respectively. Examples of parametric maps are reported.