Artificial intelligence can help detect life-threatening liver disease

Twice as many die each year from liver disease than breast cancer, but knowledge of liver disease is not as widespread.

Now team of researchers from the Faculty of Engineering at SDU, in collaboration with medical experts at OUH, has developed a method to detect liver fibrosis in time with the help of artificial intelligence. Or rather exclude almost all persons who don’t have signs of the disease.

- We have succeeded in developing a method where we can exclude people with almost 99 percent certainty based on several data and the use of artificial intelligence. The group at risk can then be examined with either a scan or a biopsy, says associate professor Victoria Blanes-Vidal from the Applied AI and Data Science Unit (AID), at the Maersk Mc-Kinney Moller Institute at SDU.

Costs 2 million lives each year

Worldwide, liver disease costs 2 million lives each year, 1 million due to cirrhosis and 1 million due to viral hepatitis and liver cancer. This places cirrhosis and liver cancer at the 11th and 16th most common causes of death, respectively.

The first sign of most liver diseases is liver fibrosis, where healthy cells are replaced by scars. Often, it worsens in silence, with a fatal outcome when the scar tissue on the liver is not detected in time and the damage to the vital organ is irreparable.

Easy method

Victoria Blanes-Vidal is part of the center for collaboration. The Maersk Mc-Kinney Moller Institute and OUH was established last year under the name CAI-X, which aims to develop technological solutions for the health sector.

This first scientific result has just been published in the journal Scientific Reports, part of the Nature Portfolio.

- Our ambition was to develop a method where, with the help of a standard blood test at your general practitioner, you can quickly and easily determine if you are at risk of having the disease. And we actually succeeded in our experiment, says the Spanish associate professor and specialist in biostatistics and data analysis. She has worked at SDU since 2008.

Although the work on liver fibrosis started before CAI-X saw the light of day last year, the project is an excellent example of the collaboration between OUH and TEK.

- It is a platform for a collaboration where we can jointly develop technology that can improve health efforts in the future, says Victoria Blanes-Vidal.

Liver fibrosis is an excellent example of a disease in which new technology can thus provide a crucial tool. In the future, the AI model from SDU can help many people who otherwise risk getting a fatal message when they discover the disease too late.

So far, it has required rather invasive procedure where you take a biopsy from the liver to analyze or use a second method; an ultrasound scan of the liver, which is expensive and requires you to go to the hospital. Finally, you can also use a blood sample, but it requires unique analysis, and it is not standard, she says.

Based on 3352 people’s health information, blood tests, and data on age, gender, weight, and alcohol intake, the researchers have built an AI model that can exclude people at risk of having the disease.

The research group has included as many as 233 different parameters in the study – but the ambition was to use little data to facilitate the method and make it as accessible as possible.

The test subjects are selected from three groups. One had been in treatment for alcohol abuse, one group was already patients in treatment, and finally, a randomly selected group who were summoned via E-boks.

- Most people associate liver disease with alcohol abuse, but there is also non-alcoholic liver disease. It is far from everyone who gets liver fibrosis who has too much alcohol consumption. Others are overweight, have high blood pressure, or are diabetics, which also poses a risk, says Victoria Blanes-Vidal.

300,000 Danes at risk

She also emphasizes that as many as five percent of the population gets the disease – equivalent to 300,000 people in Denmark, who are not necessarily overweight or have overconsumption of alcohol.

- That group will probably never go to the doctor because they have no signs of the disease. Not until it's too late. With this model, we can detect them from a routine check-up and blood test at the general practitioner, says the researcher.

The results of the research and the model with artificial intelligence, which combine six different AI analyses, show that the model can identify and exclude people who do not have the disease. And it can do so with close to 99 percent certainty.

- It is a fantastic result that we are incredibly proud of. We mustn't send people home, who then two years later are back and turn out to have the disease, says Victoria Blanes-Vidal.

For the remaining and much smaller group, who are in risk of having the disease, it will be possible to perform either an ultrasound scan or a biopsy and thus find carriers of the disease.

Next steps await

Now the next step in developing the analysis model awaits, where it will be in so-called external validation, says the researcher.

- We must have it tested and tested on other groups in Denmark and internationally to see if we can achieve equally good results there, says the Spanish researcher.