Artificial intelligence and deep learning: New tools for histopathological diagnosis of nonalcoholic fatty liver disease/nonalcoholic steatohepatitis

Abstract

Nonalcoholic fatty liver disease (NAFLD)/nonalcoholic steatohepatitis (NASH) is associated with metabolic syndrome and is rapidly increasing globally with the increased prevalence of obesity. Although noninvasive diagnosis of NAFLD/NASH has progressed, pathological evaluation of liver biopsy specimens remains the gold standard for diagnosing NAFLD/NASH. However, the pathological diagnosis of NAFLD/NASH relies on the subjective judgment of the pathologist, resulting in non-negligible interobserver variations. Artificial intelligence (AI) is an emerging tool in pathology to assist diagnoses with high objectivity and accuracy. An increasing number of studies have reported the usefulness of AI in the pathological diagnosis of NAFLD/NASH, and our group has already used it in animal experiments. In this minireview, we first outline the histopathological characteristics of NAFLD/NASH and the basics of AI. Subsequently, we introduce previous research on AI-based pathological diagnosis of NAFLD/NASH.

Keywords: Artificial intelligence; Machine learning; Nonalcoholic fatty liver disease; Nonalcoholic steatohepatitis; Pathological diagnosis.

Graphical abstract

Fig. 1

Steatosis in nonalcoholic fatty liver disease (NAFLD)/nonalcoholic steatohepatitis (NASH). Macrovesicular steatosis is observed mainly in zone 3 (Masson’s trichrome staining).

Fig. 2

Lobular inflammation in NASH. Inflammatory foci consisting of various inflammatory cells (lymphocytes, polymorphs, and Kupffer cells) are scattered in the hepatic lobule (arrows) (hematoxylin and eosin (H&E) staining).

Fig. 3

Hepatocellular ballooning in NASH. Ballooned hepatocytes are observed as swollen hepatocytes with rarefied cytoplasm reflecting hepatocyte injury (arrows) (H&E staining).

Fig. 4

Fibrosis in NASH. Perisinusoidal/pericellular (chicken wire) fibrosis is observed in the hepatic lobule (Masson’s trichrome staining).

Fig. 5

Detection of steatosis and fibrosis in NASH by artificial intelligence (AI). (A) On the H&E staining slide, hepatic steatosis is observed mainly in zone 3. (B) Hepatic steatosis is accurately detected by AI and displayed in yellow. Non-steatotic area is displayed in green. (C) On the Masson’s trichrome staining slide, fibrotic area is stained blue. (D) Fibrotic area is accurately detected by AI and displayed in red. Non-fibrotic area is displayed in green. The insets show appearances in higher magnification.