Researchers Create AI-Powered Blood Test to Detect Cancer
Cancer is one of the most common causes of death, but the huge variation in types of cancer can make early detection a challenge. A team of researchers from the US has developed a blood test that might be able to help identify early-stage tumors using the power of AI. The test can detect over 50 different cancers and narrow down tumor location to specific areas of the body.
At its most basic, cancer is uncontrolled cell growth. There are many different types of cells in the human body, and almost all of them can become cancerous. Unfortunately, many cancers don’t present with noticeable symptoms until the disease has progressed to the later stages. Once cancer begins spreading to other body systems, it becomes much harder to treat.
A routine test that could flag potential cancers early on could save untold lives and reduce medical costs. The team, consisting of researchers from the Mayo Clinic, Cleveland Clinic, and other institutions, focused on analyzing signals from so-called “cell-free DNA” (cfDNA). That’s genetic material shed by cells that circulates freely in the blood. All cells leak a little DNA, and that includes the cells in tumors.
While cancer does involve genetic alterations, those changes are minor and often limited to specific sections of DNA. A better way to spot the evidence of cancer in cfDNA is by assessing methylation patterns. In methylation, a methyl group (CH3) replaces a hydrogen atom on cytosine or adenine — those are two of the four base pair molecules that make up your genetic code. Methylation can affect the way your genes are expressed without actually changing the underlying sequence. Importantly, methylation can be a cause or result of cancer.
The team started with blood samples from 3,000 patients, half of which had one of the 50+ cancers included in the study. This was “labeled” data, allowing the AI to learn the methylation patterns in cfDNA for various cancers and for no cancer. To test the algorithm, researchers put to work categorizing 1,200 new samples, and the results are encouraging.
The AI is able to detect 93 percent of stage IV cancers, 81 percent of stage III, 43 percent of stage II, and 18 percent of stage I tumors. The rate of false positives was just 0.7 percent. When it spotted cancer, the AI was able to identify the location of the tumor with 93 percent accuracy.
The team is hopeful this technique could scale to larger populations. Detection of early cancers is still low, but those tumors are rarely found via other means. Identifying 18 percent of cancers early could still be very helpful, and additional training could make the AI better at detecting cancer in the early stages.