Artificial intelligence in prediction of steatosis and fibrosis of nonalcoholic fatty liver disease

Accurate prediction of hepatic steatosis and fibrosis is an important part of managing nonalcoholic fatty liver disease (NAFLD) for a diagnosis and prognosis. Conventional approaches such as liver biopsy are invasive with low patient acceptance and high sampling error. Non-invasive assessments for hepatic steatosis and fibrosis are more welcomed and achieve much higher applicability. Common noninvasive assessments include laboratory-based scores and various imaging modalities such as ultrasonography, transient elastography, and magnetic resonance imaging. Their accuracies range from modest to satisfactory. Artificial intelligence (AI) contributes significantly to the accurate prediction of hepatic steatosis and fibrosis in NAFLD. The most mature area is AI-assisted histologic scores, which are close to clinical application. They may improve the quality of clinical trials and facilitate comparison across studies for drug development. Laboratory-based AI models for NAFLD are developed with common laboratory parameters that are well-validated and may be applied in the clinical setting and at the population level because of their high applicability. Radiomics-based AI models involve a large number of quantitative features from medical images using data characterization algorithms. Data from histopathology, clinical parameters, and imaging data may also be combined to develop more sophisticated AI models. Further evaluation of their ability to assess and monitor disease severity and predict future outcomes is the next step to establishing the role of these AI tools in patient care.