Liver diseases: classification is changing with artificial intelligence
Investing in more accurate diagnosis means providing targeted treatment, avoiding unnecessary therapies and, in the long term, identifying which medicines are beneficial and which are harmful
It is often believed that diseases can be defined clearly and precisely, but in reality the situation is frequently very different. Doctors and researchers strive to ‘draw a line in the sand’ (or rather, ‘with a scalpel’…) around the limits and boundaries of diseases, but in clinical practice what actually happens is that a patient may display characteristics of both one disease and another. In such cases, the medical field often uses the English term ‘overlap’ to indicate the concurrent presence of features attributed to distinct conditions in a priori classifications, or ‘variant’ to indicate that the patient presents with atypical or mixed clinical features compared to the available classifications. Overlap/variant syndromes exist in many medical specialities, including liver diseases.
How the research came about
Over the past two years, I have coordinated an international consensus among the world’s leading experts to redefine one of these complex conditions: that in which two rare, autoimmune liver diseases – primary biliary cholangitis and autoimmune hepatitis – co-exist in the same patient. The findings, published in July 2026 in the journal *Hepatology*, represented a significant step forward. However, a fundamental limitation remains: they are based on the judgement of hundreds of specialists, not on new scientific data.
It is precisely this limitation that gave rise to LiverMapAI, the research project I began as Principal Investigator on 1 July 2026 at the University of Milan-Bicocca. This project was selected and funded under the competitive ‘Starting Grant – Italian Science Fund (FIS2)’ call for proposals issued by the Ministry of Universities and Research, which provides funding for early-career researchers with a budget ranging from 1.2 to 1.5 million euros. The idea is simple to state but ambitious to achieve: to use artificial intelligence to completely rethink the classification of autoimmune liver diseases and liver damage caused by drugs and herbal products (DHILI, Drug- and Herbal-Induced Liver Injury), basing the classification no longer solely on expert opinion but on the data itself.
Why include DHILI in the study as well? Because it is a public health issue, representing one of the most common causes of disruption to the development of new medicines. Furthermore, DHILI can present with clinical features very similar to those of autoimmune conditions. To date, there are no tools capable of predicting in advance which patients are at risk of DHILI or how to distinguish immune-like forms of DHILI from genuine autoimmune liver diseases. The direct consequence of a misdiagnosis is that some patients receive immunosuppressive drugs they do not need, whilst others who do need them receive them too late, by which time the liver damage has become irreversible.
The objectives of the LiverMapAI project
The LiverMapAI project operates on two fronts. The first is to use the latest techniques to gather detailed information on patients with typical clinical presentations and on those with more subtle or atypical presentations. The second is to entrust artificial intelligence, in a transparent and verifiable manner, with the task of analysing, in conjunction, liver biopsy images, the patient’s medical records, their genetic profile and the most advanced molecular analyses of the tissue, in order to identify patterns that currently remain invisible. Biopsy images, in particular, are currently examined under a microscope in a largely qualitative manner: they likely contain far more information than the human eye is capable of discerning. This is where an algorithm can make a difference.
