Matt Hoffman, Senior Editor for NeurologyLive, has covered medical news for MJH Life Sciences, NeurologyLive’s parent company, since 2017. He hosts the NeurologyLive Mind Moments podcast, as well as Second Opinion on Medical World News. Follow him on Twitter @byMattHoffman or email him at email@example.com
The platform, dubbed DystoniaNet, was able to identify 3 varieties of focal dystonia in a matter of 0.36 seconds with almost 100% accuracy.
New study findings from an assessment of artificial intelligence (AI) based deep learning platform, dubbed DystoniaNet, suggest that the platform can accurately identify patients with dystonia within a single second via magnetic resonance imaging (MRI) data. The investigators noted that the tool may help clinicians severely cut down the time to diagnosis in an otherwise challenging condition.1,2
All told, when doing a comparison using MRI data from 612 individuals—220 healthy subjects and 392 with 3 types of isolated focal dystonia (laryngeal dystonia, n = 279; cervical dystonia, n = 59; blepharospasm, n = 54)—the DystoniaNet platform diagnosed the condition with 98.8% accuracy in a matter of 0.36 seconds. Specifically, it showed 98.2% accuracy in diagnosing laryngeal dystonia, 100% in diagnosing cervical dystonia, and 98.1% in diagnosing blepharospasm, while referring 3.5% of patients (n = 6) for further examination.2
The performance of the platform lends credence to its potential to make an impact in the movement disorder field, as dystonia is an oft-misdiagnosed condition that can take up to a decade for patients to correctly receive.
“We basically leveraged the advances made in deep learning and designed an architecture that was able to look at raw structural MRI and find a biomarker for dystonia that could help with the diagnosis of this disorder,” study author Davide Valeriani, PhD, postdoctoral research fellow in the Dystonia and Speech Motor Control Laboratory at Mass Eye and Ear and Harvard Medical School told NeurologyLive. “We specifically took an approach that could be easily clinically translated, which is why we used raw structural MRI. This kind of measure is what comes out directly from the scanner and is available to many clinicians in-clinic.”
Valeriani explained that the DystoniaNet platform takes the MRI and learns from the training algorithm developed from one of the largest dystonia datasets, and then uses this data-driven approach to identify the condition. It was developed over the last several years by Valeriani and coauthor Kristina Simonyan, MD, PhD, DrMed, director, Laryngology Research, Mass Eye and Ear; associate neuroscientist, Massachusetts General Hospital; and associate professor, Otolaryngology-Head and Neck Surgery, Harvard Medical School, and was assessed in an analysis-driven marker evaluation previously.
“[The analysis-driven] results were quite encouraging, where we would have an accuracy between 71% and 81% correctly diagnosing dystonia, and research from other labs also has shown that similar analysis-driven biomarkers can be implemented for diagnosing dystonia” Simonyan said. “However, that was not good enough for us because that type of paradigm has to involve substantial data processing, which has been clinically challenging to do. Us clinicians just don’t have the luxury of several hours per patient to process their data. In a way, our study was designed to lessen the time burden on the clinician and to speed up the diagnosis.”
The performance of the biomarker and its DystoniaNet pipeline was consistently stable independent of the magnetic field strength (accuracy range, 98–100), MRI scanner vendor (accuracy range, 96.9–100), head coil (accuracy range, 95.2–100), T1-weighted image acquisition sequence (accuracy range, 98.3–100), or a data collection site (accuracy range, 97.6–100). The specificity was additionally confirmed with a third and supplementary independent dataset of 1480 healthy controls (accuracy, 96.9%; referral rate, 2.6%).
Simonyan noted that traditionally, this diagnosis is made based on clinical observations, and available literature suggests that that the agreement on dystonia diagnosis between clinicians using purely clinical assessments is as low as 34%, and with roughly 50% of the cases going misdiagnosed or underdiagnosed at a first visit.
It is estimated that 35 of 100,000 individuals have isolated or primary dystonia, though it is believed that this prevalence is underestimated due to the current challenges in diagnosis. Although Parkinson disease or stroke can lead to the development of dystonia, the majority of isolated dystonia cases have no known cause and impact only a single muscle group in the body. Focal dystonia can lead to disability and problems with physical and emotional quality of life.
“In some sense, this provided us an opportunity to train this machine-learning algorithm to diagnose dystonia because we had a large available dataset of neuroimages,” Valeriani said. “But we also made the development of DystoniaNet open to the possibility of extending it to other disorders. For example, we took away all the prognosis that’s required and just stared with raw structural MRI. Every patient could get that, and we have this data for many different disorders. We are currently testing this on some other disorders with very promising results—about 80% accuracy. We want to see how much this architecture will generalize to other disorders.”