Screening 50 cancer types by AI blood test?

Already, AI using machine learning algorithms is expected to be applied in the medical field. Results such as detecting lung cancer from CT scan data with higher precision than human doctors or determining whether cancer immunotherapy is effective have been reported. Researchers in the UK and the US have published a study showing that blood tests using a new AI can easily screen for more than 50 types of cancer.

Most cancer diagnosis by conventional blood tests detects cancer cell DNA, which starts to flow in blood when immune cancer cells are destroyed, by DDNA sequencing, which determines the DNA sequence. However, the research team developed a new blood test method that focused on the DDNA methylation pattern of cancer cells, not the DNA sequence. Because the methylation pattern that occurs in cancer cell DNA is different from that of normal cells, it is said that if cancer cells are destroyed by immunity and flow into the blood, the DNA methylation pattern in the blood can be analyzed to determine the presence or absence.

The researchers knew there existed a blood DNA-based approach, but the question is how to tune and perfect the technology for finding cancer in the DNA floating in the blood. For this reason, the research team developed a new blood test method using machine learning algorithms.

First, the research team took blood samples from more than 2,800 cancer patients and learned the DNA methylation patterns contained in the samples into an algorithm. Then, the algorithm was trained using the DNA methylation patterns contained in blood samples of 3,052 people, including 1,531 cancer-diagnosed patients and 1,521 non-cancer patients. The research team said that the algorithm learned not only the presence of cancer, but also the location of the cancer in the sample patient.

The team then tested the training system using blood samples from 1,264 people with cancer in half of the total. The samples included more than 50 types of cancer including breast and esophageal cancer, gastric cancer, ovarian cancer, lung cancer, multiple myeloma, and pancreatic cancer.

As a result, the new system correctly detected cancer in 44% of human blood samples diagnosed as cancer. In fact, the rate of diagnosing non-cancer people as cancer by mistake was very low at 0.7%. Also, the degree of cancer detection improves as the cancer progresses. Stage 1 cancer was detected in 18% of the total, whereas in stage 4 cancer was detected at 93% rate.

In some cancers, the detection precision was particularly high. The detection accuracy of pancreatic cancer was found to be 63% in stage 1 and 100% in stage 4. In addition, the system was able to detect the presence or absence of cancer, as well as to identify the type of cancer. If cancer is present, in 96% of cases judged by the system, the system guessed the cancer site and hit it with 93% accuracy.

The research team developed a blood test that could detect more than 20 cancers in 2019, but increased precision with a system using machine learning algorithms. According to the research team, the system is already in the stage of reviewing clinical trials. The results suggest that cancer detection through blood tests using AI is promising, but there is also a problem that the early cancer detection rate is low. Therefore, it is necessary to further understand the test performance by looking at the clinical trial results before this test becomes generalized. Related information can be found here .