A study carried out by the University of Trieste in collaboration with Yale University, published by Nature Npj Digital Medicine, has entered the top ten of the most cited articles in 2024 in the prestigious scientific journal, coming eighth in the ranking drawn up by the publisher.

The article entitled ‘Optimization of hepatological clinical guidelines interpretation by large language models: a retrieval augmented generation-based framework’ saw the contribution of an interdisciplinary UniTS research team, consisting of Simone Kresevic, PhD student in Biomedical and Clinical Engineering, Miloš Ajčević and Agostino Accardo from the Department of Engineering and Architecture, and Lory Saveria Crocè, gastroenterologist from the Department of Medical, Surgical and Health Sciences.

The study benefited from the collaboration between the group of researchers from the University of Trieste and the Yale School of Medicine, in particular the Human+Artificial Intelligence in Medicine centre in New Heaven (Connecticut, USA), with contributions from Dennis L. Shung, director of the research laboratory, and Mauro Giuffrè, co-authors of the paper.

The researchers explored the use of generative AI systems capable of understanding and generating human language by processing large amounts of data, Large Language Models (LLM), to optimise clinical decision support in the field of medicine.

In this study, a digital infrastructure (framework) based on LLM was therefore developed which, through the correct formatting of clinical guidelines, could improve their consultation and application in clinical practice.

The research focuses, in particular, on the integration of these Artificial Intelligence models to improve the interpretation of medical guidelines relating to the management of chronic infections caused by the hepatitis C virus (HCV).

This system, using Retrieval Augmented Generation (RAG) techniques, a method of extracting relevant information from clinical guidelines, reprocessing it using LLM and, through the generative process, providing answers to guideline-related questions in a clear and accurately structured manner.

The research activity now continues to expand the functioning of the framework on different liver diseases. This strand of research could offer a system capable of supporting physicians with decisions based on the best available scientific evidence.

‘With this approach,’ explains Dr Simone Kresevic, first author of the article together with Mauro Giuffrè, ‘we are laying the foundation for using artificial intelligence in everyday clinical practice. Evidence-based medicine, a fundamental pillar of modern medicine, aims to integrate the best available scientific evidence with clinical experience and patients' needs. However, the complexity and volume of clinical guidelines often represent a significant barrier in their application.'

'Through this framework,' Kresevic concludes, ‘we can offer a tool to support the clinician and thus support evidence-based and personalised medicine, bridging the gap between high-quality research and practical healthcare, especially in complex areas such as hepatology.’