Molecular testing can determine the course of treatment for advanced non-small cell lung cancer patients. If a patient has an EGFR mutation, they have a higher chance of responding to targeted therapy. If a patient’s cancer cells highly express a protein called PDL1, they can be treated with an immunotherapy.

Biomarker tests use archived tissue or samples taken from a biopsy. Biopsies are invasive procedures and testing may fail if the sample is insufficient. Waiting on results can take weeks—time that many advanced non-small lung cancer patients don’t have.

“The real-world barrier is the time lag,” said Dr. Matthew Schabath, an epidemiologist at Moffitt Cancer Center. “The five-year survival rate for advanced stage lung cancer is dismal if untreated, and sometimes these tests come back indeterminate because there wasn’t enough quality or quantity of the tissue.”

Researchers have been working to find alternatives to biomarker testing that are minimally-invasive, convenient and quick. This includes using a patient’s blood, urine or even breath.

What about using a patient’s scans?

Moffitt researchers presented data at the Society for Immunotherapy of Cancer Annual Meeting that shows artificial intelligence can use images to perform biomarker tests. A retrospective study of 837 non-small cell lung cancer patients found AI performed just as well as a traditional lab biomarker test in identifying EGFR mutations and PDL1 expression.

There are many advantages of using an AI analysis of a medical image. It does not require any lab testing; the images reflect the entire tumor instead of just the portion of the tumor biopsied; and biomarkers can be calculated immediately after a scan. 

In the future, imaging biomarker tests could be potentially integrated into radiology. As soon as a patient has a scan, the radiologist could identify the tumor or region of interest and use AI to make an informed decision.

“This has tremendous potential in speeding up the delivery of therapy,” Schabath said. 

The next steps are to validate the program on larger data sets and eventually open a trial to test AI in a real-world clinical setting.