Brain scans and algorithms combine to diagnose concussion


While the symptoms are frighteningly obvious, some types of brain injury are impossible to accurately identify. A new approach developed in Canada might be about to change all that, Angus Bezzina reports.


Magnetoencephalographic imaging, which maps interaction between regions, along with machine learning could one day objectively diagnose concussion.
Vakorin et al

Improved imaging techniques combined with machine learning could soon provide doctors with a precise way of diagnosing types of concussion that currently don’t show up on conventional brain scans.

Mild traumatic brain injury – which often affects colliding football players – does not appear on standard magnetic resonance imaging (MRI) or computed tomography (CT) scans, even though sufferers show symptoms such as headache, fatigue and memory loss.

So diagnosis is based on behavioural symptoms, patient interrogation and the doctor’s best judgement – a potentially risky strategy, given in some cases untreated mild traumatic brain injury ends up in lifelong impairment.

A pilot study by Canadian scientists, led by Vasily Vakorin from Simon Fraser University in British Columbia, found improved magnetoencephalographic (MEG) imaging can map how different areas of the brain interact by directly measuring its activity at fast time scales.

The study, published in the journal PLOS: Computational Biology, reports that mild traumatic brain injury is linked with less connectivity in brain areas that operate in the delta and gamma frequency range (that is, above 30 Hertz).

At the same time, “slow-wave” brain activity in the eight to 12 Hertz alpha range is markedly increased. These could be used to objectively diagnose concussion following a knock to the head.

Will Woods, a specialist in MEG research at Swinburne University in Melbourne, Australia, said the machine learning approach meant that a pool of previously obtained data could be used to better detect new cases of concussion.

“The idea is that once you have trained a machine classifier, somebody new comes along, you use the MEG to measure this composite variable which relates to the connectivity of their brain and then you get the machine to tell you which side of the boundary they are on,” he explained.

To find the hallmarks of concussion, Vakorin and his team took MEG scans from 41 men aged between 20 and 44.

Half had been diagnosed with mild brain trauma within the previous three months, with the remainder healthy controls.

The scans showed that patients with concussion had distinctive changes in communication between different areas of their brains.

Increased levels of slow-wave activity were shown to be characteristic of the injury. Recent advances in MEG imaging sensitivity enabled the researchers to better focus on the source of this change, confirming the hypothesis that mild traumatic brain injury stems from damage to the brain’s white matter microstructures.

Vakorin’s team took readings of resting state activity in 90 cortical and sub-cortical brain regions from each of the mild traumatic brain injury patients and controls. By combining analyses of network connectivity with machine-learning algorithms, they were able to identify the concussion patients with 88% accuracy.

While only a preliminary study, Vakorin’s paper reports that the results “provide potentially clinically translatable methods that will permit the detection of mTBI in single individuals where conventional radiological imaging approaches are inconclusive”.

  1. http://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1004914
Latest Stories
MoreMore Articles