I. Introduction
Nonalcoholic fatty liver disease (NAFLD) is now the most common form of chronic liver disease in the world. The prevalence of this disease has been estimated to be over 25% in the general worldwide population [1]. This disease encompasses a spectrum of changes in the liver related to fat deposition. The changes range from non-alcoholic fatty liver (NAFL) with simple steatosis (> 5% liver fat content) with minimal or no inflammation, to a progressive form of the disease called nonalcoholic steatohepatitis (NASH). NASH is characterized by steatosis, inflammation and hepatocellular injury, with eventual progression to various stages of fibrosis [2]. As the progressive form of the disease, NASH is associated with increased morbidity and mortality; therefore, determination of this disease is important during diagnosis. Liver biopsy is the gold standard for determining the fibrosis stage and diagnosing NASH from NAFLD activity score (NAS) to differentiate it from simple steatosis. However, the cost of an invasive procedure such as liver biopsy combined with the possibility of complications such as bleeding, infections, and rarely death, rule out its routine use in clinical practice [3]. In addition, intra-observer variance along with sampling variance add significant error to the manual interpretation of liver biopsy histopathology [4], [5]. Therefore, it is of great value to develop computational methods for NAS score prediction and fibrosis staging from non-invasive imaging modality data. Besides, when the biopsy data is available, it is meaningful to build computational models to predict scores from the biopsy data, which can assist pathologists in the diagnosis process by saving time and reducing the unreliability of less-experienced doctors.