Introduction

Non-alcoholic hepatic steatosis (NAFLD) affects approximately 20-30% of the adult population in developed countries and causes hepatocellular carcinoma. The non-invasive method used to diagnose NAFLD is liver ultrasound. In recent years, machine learning has been recognised as a low-cost diagnostic method. Therefore, an intuitive and easy-to-use neural network-based web application was created to support medical decisions during the diagnostic phase using simpler and less expensive tools. It is based, in fact, on the input of simple biochemical and anthropometric variables into the web app that are readily available in healthcare databases.

Technical features

It is based on a Neural Network type artificial intelligence algorithm trained with anthropometric and biochemical data of both healthy and proven NAFLD subjects. The developed algorithm is able to provide the NAFLD condition as an output, based on the input variables, corresponding readily available biochemical and anthropometric values are used. In practice, the user (Clinician) will enter the anthropometric and biochemical data of the subject into the app, the Neural Network will output the prediction made. The Neural Network algorithm employs a model consisting of: AVI + glucose + GGT plus age + gender, where AVI is the Abdominal Volume Index according to a formula that is able to detect NAFLD with an accuracy rate of 75.3% and a Roc curve value (AUC) of 0.81.
TRL: 3

Possible Applications

• Epidemiological studies and screening;
• Follow-up visits (general practitioner);
• Large-scale application in epidemiological studies;
• Supporting the clinician in the diagnosis of NAFLD.

Advantages

• Reduction of waiting lists and healthcare expenditure;
• Ease of application (parameters available in databases) and use (no additional equipment required);
• Due to the high discriminating power of the algorithm, only patients positive for “Presence of NAFLD” are sent for in-depth ultrasound examination;
• Algorithm easily used even by general practitioners.