Many people who suffer mild concussions from sports injuries, whiplash, or blows to the head may not realize the long-term health effects of leaving the injury untreated. Despite receiving emergency room treatment, the majority of concussions go undiagnosed, leaving patients vulnerable to serious complications such as brain hemorrhage and cognitive impairment.
A recent collaborative study between the USC Viterbi School of Engineering and the USC Leonard Davis School of Gerontology used advanced machine learning to accurately predict a patient’s concussion status.
A concussion is a type of traumatic brain injury that can result in temporary changes in brain function. Current methods for diagnosing concussions often rely on basic cognitive tests, such as the Glasgow Coma Scale, according to Hacker. However, diagnosing is difficult because many patients with mild concussions do not lose consciousness and may not exhibit typical cognitive symptoms. Hacker believes existing testing methods are not sensitive enough to detect many mild cases.
“We saw an opportunity to fit this test somewhere between 'not at all concussive' and a concussion severe enough to be detected consistently.” “This is a game changer,” said Hacker, who wrote the paper as an undergraduate at USC Viterbi and is now a master's student in the Mork Family Department of Chemical Engineering and Materials Science.
“Clinician” He added, “We don't necessarily prescribe imaging or request MRIs for patients who don't have symptoms. The idea is to do this as an adjunct to assist clinicians when a patient is exhibiting certain symptoms but cannot be definitively diagnosed with a concussion based on cognitive testing alone.”
Hacker said that his colleagues, led by Andrei Irimia, an associate professor of gerontology, biomedical engineering and neuroscience at the Leonard Davis School of Gerontology at USC, developed the model using MRI brain scan data from both healthy and people with concussions. Their classifier is based on diffusion-weighted imaging, which assesses the movement of fluid along different pathways in the brain.
“This data quantifies the directionality of diffusion between different regions in the brain, which tells us how strongly these different nodes are linked. We then used machine learning to develop a classifier.” The hacker said. “We trained this classifier on a discovery sample, teaching it how the connectivity matrices of healthy and injured individuals differ. Then, when we gave it an independent test sample, it was able to detect which of these subjects had a concussion and which were healthy, based on the patterns in their brain connectivity matrices and the strength of certain neural pathways.”
Hacker and his team found that the classification model performed extremely well, achieving 99% accuracy on both the training and testing datasets.
“This is a level of accuracy we've never seen before with this method.” The hacker said. “I think it's because no one has come up with exactly our pipeline – taking diffusion-weighted images, converting them into a connectivity matrix and leveraging customized machine learning to discover the pathways most affected by head trauma. This is certainly novel in that, until now, we haven't had a reliably accurate image-based classifier to classify concussions.”
The classifier was built using Bayesian machine learning, which Hacker described as a probabilistic system that creates classifications based on features that are least likely to be incorrect or misclassified, based on prior knowledge of the situation.
“We use the training data to determine what patterns we see in healthy people and what patterns we see in injured people.” The hacker said.
“I feel like this research will definitely have a positive impact on the field and has great potential. That's the most exciting part to me. I can't wait to see what this leads to.” The hacker said.
Journal References:
- Benjamin J Hacker, Phoebe E Imms, Ammar M Dharani, Jessica Zhu, Nahian F Chowdhury, Nikhil N Chaudhari, Andrei Irimia. Concussion Identification and Connectomics Profiling Using Bayesian Machine Learning. Journal of Neurotrauma, 2024, DOI: 10.1089/neu.2023.0509
