NY Cancer Center Develops Machine-Learning Sensor to Sniff Out Cancer

Researchers at Memorial Sloan Kettering Cancer Center (MSK) have created a tool that can sniff for and identify cancer in blood samples using machine learning.

Like the human nose, the tool uses sensors to detect molecules and how they interact, which generates a unique molecular signature of a given disease. The tool’s sensors are made of fluorescent carbon nanotubes almost 100,000 times smaller than the width of a human hair. The light the nanotubes produce can be used to generate a pattern. These patterns are then given to a machine-learning model trained to differentiate between cancer and normal patterns.

When testing their tool using blood samples from ovarian cancer patients, MSK researchers found that it detected cancer more accurately than standard biomarker tests.

Tests that use blood markers to identify cancer can support early detection and help improve outcomes for patients, the researchers noted in the study. But various types of cancer, including ovarian, cannot be identified at an early enough stage to reduce mortality using currently available methods. This is where the sensor can help.

“Ovarian cancer spreads along the surfaces of the abdomen and pelvis [rather than through the blood], which makes finding it with a blood test especially challenging,” said study author and MSK surgeon Kara Long Roche, MD, in the press release. “This technology could potentially find more subtle, complex changes in the blood, which may be the key to early detection — and early detection will save lives.”

Traditional cancer screening tests also face similar early detection challenges because they rely on cancer biomarker identification, unlike the newly developed tool.

“A major limitation in the development of cancer screening tests has been the lack of sufficient biomarkers,” stated Lakshmi Ramanathan, chief of the clinical chemistry service at MSK, in the press release. “The ability to develop a screening test without the need for a biomarker is an exciting possibility for this type of technology.”

The researchers plan to validate the tool soon so that it can be used in clinical settings. They also hope to develop the tool further to detect other types of disease.

Other efforts to bolster preventive health and improve cancer care are also underway.

Earlier this month, researchers at Children’s Hospital of Philadelphia (CHOP) announced that they had developed a machine-learning platform designed to assist clinicians with identifying cancer mutations and interpreting their significance.

The platform, known as CancerVar, takes clinical information from existing databases related to 13 million somatic variants from 1,911 cancer census genes and combines it with a deep-learning algorithm. From there, clinicians can generate automated descriptive interpretations for variants.

According to the researchers, CancerVar highlights how computational tools can be used to reduce clinician labor, improve consistency of variant classification, and help address challenges related to cancer diagnosis and prognosis.