Observational and Prospective Study of Hepatic Steatosis and Related Risk Factors Using Ultrasound and Artificial Intelligence

In recent years, ultrasound has taken on a predominant role in the evaluation of liver steatosis, as it is a non-invasive, non-irradiating method that is easily reproducible and inexpensive. Of particular effectiveness is the use of the hepatorenal index, evaluated as the intensity ratio (echogenicity) between the hepatic parenchyma and the renal cortical parenchyma. The main limitations of detecting the hepato-renal index during abdominal ultrasound, however, are operator dependence and the use of a relatively long time span to complete the sequence of operations and calculations required to determine the index itself. The use of Artificial Intelligence (AI) techniques for image analysis in the medical field is yielding excellent results. AI-based algorithms are increasingly a powerful tool that allows the physician to improve their performance in terms of speed and accuracy of clinical evaluations. Today, there is already evidence of the effectiveness of using AI on ultrasound images for clinical evaluations. The use of AI as an aid in diagnosing liver diseases through ultrasound is still under-researched. The hypothesis to be tested is the utility that AI can have in the evaluation, its general and specific uses in reducing calculation times of the hepatorenal index.

In this study, 134 patients were enrolled with no clinical suspicion of liver steatosis. All patients underwent abdominal ultrasonography (US) and magnetic resonance imaging fat fraction (MRI-PDFF), assumed as reference technique to evaluate the grade of steatosis. The hepatorenal index (US) was manually calculated (HRIM) by 4 skilled operators. An automatic hepatorenal index calculation (HRIA) was also obtained by an algorithm. The accuracy of HRIA to discriminate different grades of fatty liver was evaluated by Receiver operating characteristic (ROC) analysis using MRI-PDFF cut-offs.