A multi-parametric model for progression of metabolic dysfunction-associated steatohepatitis (MASH) in humans

Multiparametric analysis of quantitative ultrasound parameters was previously shown to improve assessment of metabolic dysfunction-associated fatty liver diseases (MAFLD). In this study, we aim to develop a multiparametric model for metabolic dysfunction-associated steatohepatitis (MASH), which contains more complex disease progression as an advanced version of MAFLD.We extracted quantitative ultrasound parameters, including H-scan frequency, Burr distribution λ and b, B-mode intensity, and shear wave speed (SWS). The parameters were categorized and displayed in multiparametric space. Support vector machine (SVM) was used to produce hyperplanes to differentiate MASH stages. Gaussian mixture model (GMM) was used to identify the centroids of the MASH stages. The centroids of MASH stages 0, 2, and 4 were then used to find early and late stage MASH progression vectors.To evaluate the multiparametric model, we performed an in vivo human study. 39 patients were enrolled and unterwent clinical tests, such as biopsy, blood biochemistry, metabolomics test (OWLiver), and ultrasound B-mode and shear wave elastography (SWE). A clinician confirmed MASH stages based on the clinical test results (M0: no disease; M1: steatosis; M2: steatohepatitis; M3: steatohepatitis with fibrosis; M4: steatohepatitis with cirrhosis).Complex disease progression was not well characterized by individual parameters, but the multiparametric model captured the trajectory of MASH progression. SVM classification resulted in 87.0% and 76.8% accuracy for training and testing, respectively. SVM and GMM produced a consistent trajectory in the multiparametric space. In conclusion, our multiparametric model was able to track nonlinear MASH progression trajectory accurately.

I. Introduction

Quantitative ultrasound (QUS) approaches improve diagnostic performance over traditional B-mode imaging by enhancing the qualitative nature and morphological features from the B-mode images with quantitative information about local tissue. This has lead to the development of a variety of QUS parameters for metabolic dysfunction-associated fatty liver diseases (MAFLD) [1]. These MAFLD-associated QUS parameters have been shown to be accurate alternatives to MRI-estimated proton density fat fraction (MRI-PDFF) [2]. However, to apply QUS approaches effectively in the clinic, clinicians need to incorporate information from multiple QUS measures linked to different underlying disease properties. Multiparametric analysis is particularly attractive because it can simplify diagnostics by integrating information from multiple parameters. A multiparametric approach has previously been verified with MAFLD, yielding a single output parameter incorporating multiple input parameters, which showed higher correlation MRI-PDFF compared to all individual parameters and thus enabled improved hepatic steatosis identification [3]-[5]. Multiparametric models have also been applied to simple liver disease models, such as classification of normal/fibrosis/fat/inflammation [6], [7] and classification of normal/steatosis/fibrosis/tumor [8].