To the human eye scrutinizing MR images, brain cancer cells can look a lot like cells killed or damaged by radiation treatment. Now comes a computer program capable of beating neuroradiologists at distinguishing between the two.

In a feasibility study published Sept. 15 in the American Journal of Neuroradiology, Pallavi Tiwari, PhD, of Case Western Reserve and colleagues describe their work developing the program, which combines machine-learning algorithms with algorithms for radiomics.

The latter field is concerned with extracting massive datasets of quantitative features from medical images.

Using MRI brain scans acquired at their institution from 43 patients, Tiwari and team trained a computer to pick out radiomic features that separate cancer cells from cells with radiation necrosis.

Next the researchers developed algorithms to find the most discriminating radiomic features—in this case, textures that can’t be seen by “simply eyeballing the images,” according to a press release sent by the university.

It turned out the program was nearly twice as accurate as two neuroradiologists at making the cancer vs. necrosis distinction.

Specifically, one neurorad correctly diagnosed seven of 15 patients, the second correctly diagnosed eight—and the computer program nailed 12 of the 15.

“What the algorithms see that the radiologists don’t are the subtle differences in quantitative measurements of tumor heterogeneity and breakdown in microarchitecture on MRI, which are higher for tumor recurrence,” Tiwari says in the release.

Case Western says the researchers don’t expect their program to displace neuroradiologists but, rather, to work as a decision-support tool.

They plan to test their approach in a multi-site study using many more images.