SEP 27, 2016 2:22 PM PDT

Machine Outperforms Doctors at Diagnosing Brain Cancer

“Come in, Mr. Smith. The computer is ready for your diagnosis now.” This could conceivably be in store for brain cancer patients in the future, as a new computer program shows nearly twice the accuracy of doctors at diagnosing the tumors.
Computer training and machine learning has come a long ways – machines have beat humans in Jeopardy!, chess, even Go. In August 2016, machine learning even helped to rekindle seemingly deadened nerves, giving paraplegics limb sensations they hadn’t felt in years. Now, Case Western Reserve researchers say their computer program can tell whether abnormal MRI scans are indicative of dead tissues (radiation necrosis) or the recurrence of brain cancer. And, impressively, the program can do this better than trained neuroradiologists.

MRI scans of patients with radiation necrosis (above) and cancer recurrence (below) are shown in the left column. | Image: Tiwari Lab
"One of the biggest challenges with the evaluation of brain tumor treatment is distinguishing between the confounding effects of radiation and cancer recurrence," said Pallavi Tiwari, assistant professor of biomedical engineering at Case Western Reserve and leader of the study. "On an MRI, they look very similar."
To better distinguish between radiation necrosis and cancer recurrence, the team trained a computer to read radiomic features from an image scan. These features include textures and densities that can’t be gleaned with the naked eye.
"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," said Tiwari.

The algorithm also takes into account the edge space of a mass. This can reveal important information about the identity of the mass. "If the edges all point to the same direction, the architecture is preserved," said Anant Madabhushi, professor at Case Western who is co-author on the study. "If they point in different directions, the architecture is disrupted -- the entropy, or disorder, and heterogeneity are higher."
In a head-to-head test, the team gave 15 MRI scans to 2 physicians and compared the results to that from the computer algorithm. The algorithm was able to correctly diagnose more cases (12 total) than either of the physicians (7 and 8 cases each).
While it may still be unreasonable to solely rely on a computer algorithm to read brain cancer scans, the team hopes this tool will augment the accuracy of diagnoses for patients. Importantly, the right diagnosis can eliminate unnecessary biopsies and other procedures that are both extremely stressful and costly. 

Additional source: Case Western Reserve University
About the Author
  • I am a human geneticist, passionate about telling stories to make science more engaging and approachable. Find more of my writing at the Hopkins BioMedical Odyssey blog and at
You May Also Like
SEP 11, 2019
Health & Medicine
SEP 11, 2019
Diagnosis and Treatment of Neurosarcoidosis
Sarcoidosis is a multisystem inflammatory disease characterized by the formation of non-caseating granulomas in the affected organs. The majority of p...
SEP 17, 2019
Clinical & Molecular DX
SEP 17, 2019
An All-Inclusive Genetic Testing Made Easy for the Prospective Parents
An increasing number of parents are opting their babies in for at-home genetic testing. These at-home ...
NOV 12, 2019
NOV 12, 2019
Allergy Shots May Work for Kids with Pollen Food Allergy Syndrome
It’s not common for young children to develop pollen food allergy syndrome (PFAS), but for those that do, there’s not too much parents can do o...
JAN 14, 2020
Clinical & Molecular DX
JAN 14, 2020
Can I eat this donut? A quick test for celiac disease.
Genetic testing revealed that our ancestors have been eating wheat, rye, spelt and barley for over 8,000 years. Today, gluten, a protein found within these...
JAN 13, 2020
Chemistry & Physics
JAN 13, 2020
Magnetic Field-guided Tethered-probe Can Navigate Complex Vascular Networks
Deep and complex vasculatures such as carotid arteries represent a challenge for diagnosis and treatment because they are buried underneath layers of other...
MAR 23, 2020
Genetics & Genomics
MAR 23, 2020
Diagnosing Cancer by Looking for Microbial DNA in the Blood
Liquid biopsies aim to diagnose a disease with only a bit of biological fluid, usually blood....
Loading Comments...