Machine learning accurate at colorectal cancer diagnosis, may also predict survival

Researchers in China find that a new model accurately screens for the disease and can predict survival.

Jan. 2 (UPI) -- Can artificial intelligence assist in colon cancer screening?

A study published Thursday in the journal Science Translational Medicine suggests a new machine learning platform can help identify people with colorectal cancers, and perhaps even predict disease severity and chance for survival.

According to researchers, the non-invasive testing approach supplements recent advances in technologies that analyze circulating tumor DNA, or ctDNA, and could assist oncologists in spotting colorectal cancers in at-risk patients at earlier stages.

An accurate, noninvasive diagnostic test, researchers wrote in the study, is "highly desirable."

"The majority of colorectal cancer cases can be successfully treated if detected early," researchers wrote. "Colonoscopy is widely recognized as an effective tool for screening, but its cost and invasive nature limit its use. Moreover, colonoscopy requires bowel cleansing, is often painful, and may at times be biased by inter-observer variability, especially for early lesions, lessening screening efficacy."

Like many other cancers, colorectal cancers are most treatable if they are detected early --- before they are able to metastasize to other regions of the body, like the lungs. Although colonoscopies are the "gold standard" for diagnosis, both the procedure itself and required prep are uncomfortable and can lead to complications.

As a result, many people at risk for this type of cancer are less willing to undergo screening.

To find an alternative, Huiyan Luo, a researcher at Sun Yat-sen University Cancer Center in Guangzhou, China, and colleagues leveraged machine learning techniques to develop a less invasive diagnostic method.

According to Luo, their technology works by screening for methylation markers, which are DNA modifications that are frequently found in tumors.

To test the approach, they created a diagnostic model based on nine methylation markers associated with colorectal cancer, which they identified by studying plasma samples from 801 people with colorectal cancer, as well as 1,021 healthy controls. This model accurately distinguished patients from healthy individuals with a sensitivity and specificity of 87.5 percent and 89.9 percent, respectively, and outperformed a clinically available blood test called CEA.

In addition, a modified prognostic model combined with established clinical characteristics, such as tumor location, helped predict the patients' risk of death over an average follow-up period of 26.6 months. One methylation marker was particularly useful, the authors noted, as screening for it alone spotted cases of colorectal cancer and pre-cancerous lesions in a prospective study of 1,493 at-risk individuals.

However, they noted that studies with longer follow-up periods are needed to further assess their model's reliability for clinicians and patients.

"Collectively, our findings demonstrated the usefulness of cfDNA methylation markers for diagnosis, prognostication, and surveillance of colorectal cancer, with the potential to be used for early detection of asymptomatic patients," the authors wrote. "The results of this study offer support for setting up large-scale randomized clinical trials to validate its clinical applicability."

Latest Headlines