Liver cancer detection by machine learning


Liver cancer is diagnosed in 8,200 new individuals each year, the majority being men. Early detection of this cancer greatly increases the chances of recovery. Focus on a new screening method...

The main liver cancer is hepatocellular carcinoma (70%) and almost always occurs in conjunction with cirrhosis, which can be caused by overconsumption of alcohol, or chronic hepatitis B or C. The other types of primary cancers are mainly cholangiocarcinomas and angiosarcomas. If the incidence of some cancers displayed a decreasing trend in recent years, this is unfortunately not the case for liver cancer. Evocative symptoms occur quite late (abdominal pain, fever, loss of appetite and weight, jaundice, etc.), underlining the importance of screening for people at risk.

The first-line examination to detect this cancer is abdominal ultrasound, but this method has certain limitations, hence the need for more effective techniques. An artificial intelligence-based blood test, called DELFI, has been developed by an American team at John Hopkins University. This method showed high specificity ≥80% and sensitivity ≥88% in a study on 724 patients. This test measures modifications in circulating tumor DNA fragmentation in blood, which reflect chromatin and genomic changes in liver cancer. Based on these measurements, the machine learning approach was used to distinguish healthy patients and those with cancer.

According to the scientists behind this new screening method, it could detect twice as many liver cancers currently diagnosed by conventional blood tests. In addition, it is an inexpensive and non-invasive test, which could be a major advance in the prevention of liver cancer.