Researchers at Emory and Northwestern universities have developed artificial intelligence software that can predict the survival of patients diagnosed with glioma, a deadly form of brain tumor, by analyzing digital images of tissue biopsies.

Being able to predict the course of a patient’s glioma at diagnosis is vital given that it carries a bleak prognosis. Gliomas are classified as either low-grade or high-grade based on their appearance under a microscope. The problem is that microscopic examination is extremely subjective, with different pathologists often providing different interpretations.

However, researchers used deep learning to train software to learn visual patterns associated with patient survival using images of brain tumor tissue samples. What they discovered was that when the software was trained using digital images and genomic data their predictions of how long patients survive beyond diagnosis were more accurate than those of human pathologists.

“Our approach surpasses the prognostic accuracy of human experts using the current clinical standard for classifying brain tumors and presents an innovative approach for objective, accurate, and integrated prediction of patient outcomes,” conclude the authors in an article published this week in Proceedings of the National Academy of Sciences.

Also See: Machine learning automatically identifies brain tumors

“The way the (convolutional neural) network works is we show it the images and the patients’ survival—sort of as an input and output pair—and by doing that repeatedly the network learns the relationships between the images and the clinical outcomes of the patients,” says Lee Cooper, the study’s lead author and a professor of biomedical informatics at Emory University School of Medicine.

In addition, researchers demonstrated using visualization techniques that the software learns to recognize many of the same structures and patterns in the tissues that pathologists use when conducting their examinations under a microscope and grading the brain tumors. Public data produced by the National Cancer Institute’s Cancer Genome Atlas project was leveraged to develop and evaluate the algorithm.

“Prognosis depends on both genomics and histology and so it’s important to have a unified framework that can take all the data that comes out of the pathology department and make these predictions,” adds Cooper, who is also a member of the Winship Cancer Institute. “What we’re hoping to do is to identify patients who can benefit from more aggressive therapy and also to avoid applying therapy where there will be no benefit but where there are significant side effects.”

Going forward, he says researchers are planning future studies to assess whether the software can be used to improve outcomes for newly diagnosed glioma patients, with an eye towards clinical trials.

“The eventual goal is to use this software to provide doctors with more accurate and consistent information. We want to identify patients where treatment can extend life,” concludes Cooper. “What the pathologists do with a microscope is amazing. That an algorithm can learn a complex skill like this was an unexpected result. This is more evidence that AI will have a profound impact in medicine, and we may experience this sooner than expected.”

Nonetheless, he acknowledges that being able to explain how algorithms work remains a critical barrier to their adoption for clinical use. “The issue is they’re very complex and so they can’t really be analyzed mathematically typically, so you have to use techniques like visualization to understand what they are doing.”

Register or login for access to this item and much more

All Health Data Management content is archived after seven days.

Community members receive:
  • All recent and archived articles
  • Conference offers and updates
  • A full menu of enewsletter options
  • Web seminars, white papers, ebooks

Don't have an account? Register for Free Unlimited Access