Kelly McSweeney

Feb 21st 2019

Machine Learning Approach to Cancer Detection


Researchers at the University of Toronto recently developed a new method for cancer detection that could catch the disease before symptoms appear. When cancer is detected early, patients may have better survival rates, according to the National Cancer Institute, but we don’t have effective screening tests for early detection of many types of cancers. The Toronto team led by Dr. Daniel De Carvalho combined several techniques to develop a blood test that reveals biomarkers for cancer.

A Blood Test for Cancer

The researchers combined liquid biopsy, epigenetic alterations and machine learning to detect cancer by identifying molecular biomarkers in the blood that indicate cancer even at the very early stages, according to Nature. This isn’t the first time that scientists have attempted to develop a blood test for cancer. Liquid biopsies, as these blood tests are called, detect small amounts of circulating-free DNA (cfDNA). In other words, they can find microscopic evidence of tumors circulating in a patient’s blood. It’s a growing segment of the medical field that is expected to hit more than $22 billion by 2022, according to MarketWatch.

Previously, liquid biopsies haven’t been specific enough or sensitive enough to help doctors identify a treatment plan. MarketWatch pointed out that low specificity and sensitivity are restraining factors of the liquid biopsy market, and BioNews reiterated this ongoing challenge, saying that previous liquid biopsy techniques had a low sensitivity for cancers at the earliest stages.

Improving Upcoming Medical Technology

The big challenge with liquid biopsies is that they are looking for a very small amount of cfDNA in blood, which is also filled with other protein that has nothing to do with cancer. There is simply too much data to draw useful conclusions.

“‘We are very excited at this stage,'” Dr. De Carvalho said in a press release published on Science Daily. “‘A major problem in cancer is how to detect it early. It has been a ‘needle in the haystack’ problem of how to find that one-in-a-billion cancer-specific mutation in the blood, especially at earlier stages, where the amount of tumor DNA in the blood is minimal.'”

Carvalho and his team approached the problem as if they were data scientists. They profiled changes in gene expression called “epigenetic alterations,” which helped them identify thousands of modifications unique to each cancer type, according to the press release. Then they applied machine learning to sift through all the data to create classifiers to highlight the presence of tumor DNA within blood samples and, importantly, to determine the type of cancer.

So far, they have profiled and successfully matched more than 700 tumor and blood samples. Looking ahead, the team will analyze data from large population health research studies already under way in several countries, where blood samples were collected months to years before cancer diagnosis.

Are We There Yet?

While this work makes great progress toward an effective liquid biopsy, early cancer detection hasn’t been solved yet.

In Nature, Carvalho and his colleagues explained, “‘This work sets the stage to establish biomarkers for the minimally invasive detection, interception and classification of early-stage cancers based on plasma cell-free DNA methylation patterns.'”

Even if this technique is perfected in the future, liquid biopsies still wouldn’t be a cure-all medical technology. The American Cancer Society pointed out that just because a small amount of cancer is identified doesn’t mean doctors know what to do next. Treatment depends on the type of cancer and how aggressive it is. In fact, some cancers are so slow-growing that they would never need treatment.