Scientists have trained a supercomputer to automatically detect brain tumours from scans with 90% accuracy.
George Biros, a mechanical engineering professor at the University of Texas at Austin, has been fine-tuning the system for nearly 10 years.
That system was trained to spot the most common and aggressive kind of tumour that starts in the brain – glioma.
The technique won’t replace the work of radiologists and brain surgeons, but could make for a faster diagnosis, which is an important factor in survival.
The group from Austin tested their algorithm at the Multimodal Brain Tumour Segmentation Challenge, an annual competition that sees teams from around the world present their tumour detection systems.
First they were given 300 sets of brain scans to calibrate their supercomputer, then they had two days with 140 patients’ data to identify tumours and differentiate between the different types of tissue.
They were able to do it in just four hours and correctly identified tumours 90% of the time, which is comparable to a human radiologist.
Professor Biros said their system was “similar to taking a picture of one’s family and doing facial recognition to identify each member, but here you do tissue recognition, and all this has to be done automatically”.