A group of Stanford researchers has developed a computer program that can evaluate lung cancer more accurately and efficiently than pathologists.

Currently, pathologists still use slides that they review themselves to assess lung cancer, making the process subjective. According to Kun-Hsing Yu M.S. ’14 Ph.D.’16, lead author of the study, when two different pathologists look at the same slide, they only agree 60 percent of the time. This means the same patient would get different diagnoses four times out of 10.

“[Classifying tumors] is done by pathologists using primarily their eyeballs and that obviously seems incredibly crude,” said Michael Snyder, professor and chair of genetics and one of two senior authors of the paper. “It seems like the more proper way to do this would be a very systematic and sophisticated way … where you can do this automatically.”

Using a database of images from patients with lung cancer, the researchers sought a more mechanized approach to classifying tumors. They developed a program that is able to identify thousands of features of cells, going far beyond what the naked eye can see. The new information provided by the software could help clinicians make more informed decisions about treatment plans.

The researchers published their study in Nature Communications on Aug. 16.

Historically, cancer has been classified based on the grade and stage of a tumor, meaning the cell shape and size. Higher-grade tumors receive more aggressive treatment. According to Yu, this treatment system is not always optimal, especially with lung cancer. The computer program allows for a more detailed evaluation.

“We can use our … [program] to distinguish the longer-term survivors from the shorter term survivors and this may inform clinicians in terms of treatment options,” Yu said.

Although the program has been tested on two different databases and gotten accurate results, the researchers are still validating the software with the hope that it will be used in a clinical setting to aid pathologists.

According to Snyder, using computers to analyze images is just the first step in developing more accurate prognoses. Combining the researchers’ program with other tests will allow for even more accuracy.

According to Snyder, combining imaging analysis with molecular tests that analyze the DNA of patients will help clinicians look for genetic mutations and further inform treatment plans.

“[Image analysis and molecular tests] are different kinds of data but together they could wind up being very powerful for classifying cancers and making a prognosis based on what you automatically pull out,” Snyder said.

According to Yu, image analysis software could potentially expand to different forms of cancers and researchers are currently trying to find more data sets to see if the software they have developed for lung cancer can be extended.

The lung cancer software’s potential for expansion and use alongside other tests shows that the medical world is becoming increasingly digital, Snyder said.

“In my mind this is gonna be the future,” Snyder said. “I don’t see physicians, at least at a first pass, looking at slides. I see machines doing readouts and physicians just double checking … I think in the end it’s going to wind up being a much more accurate test than the way pathologists currently do things.”

Contact Shilpa Sajja at 19ssajja ‘at’ castilleja.org.