AI detects hidden movement clues linked to brain disorders, study shows

Early detection of even the slightest motor function changes can be critical to slowing the progression of Parkinson’s disease. Yet these subtle signs often go unnoticed. 

Now, UF researcher Diego L. Guarín, Ph.D., is harnessing AI to spot these subtle changes from video recordings before clinical symptoms become evident to the clinician’s eyes. 

Guarín, an assistant professor in the UF College of Health & Human Performance’s Department of Applied Physiology & Kinesiology and an affiliate faculty member in the Herbert Wertheim College of Engineering’s J. Crayton Pruitt Family Department of Biomedical Engineering and the Norman Fixel Institute for Neurological Diseases, recently published the results of his research in Nature. 

“Video analysis is allowing us to see movement alterations that the eyes of the clinician cannot see,” Guarín said. “Early identification of these movement alterations is critical for disease management.”

In his study, Guarín analyzed videos of finger-tapping movements from 66 participants, including healthy individuals; people with idiopathic REM sleep behavior disorder, or iRBD; and people with early Parkinson’s disease. 

Idiopathic REM sleep behavior disorder involves people acting out their dreams, including talking, moving or even lashing out while sleeping. More than 80% of people with iRBD will develop Parkinson’s or a related brain disorder, making this an important subset of the population for studying early changes in motor function. 

Importantly, trial participants needed to show no visible signs of Parkinsonism – brain conditions that have similar symptoms, like slowed movements – on their finger-tapping videos. 

“An expert clinician looked at the videos and indicated that those participants were healthy,” Guarín said. “Everyone we took an analysis of looked healthy to an external observer.”

Video recordings were analyzed using VisionMD, an open-source machine learning software that Guarín’s team developed.

“When you process this video of a healthy-looking person with VisionMD, it will immediately say, ‘No, this person is moving way slower than you expect from a healthy person,’ so there are some motor alterations present in the video that cannot be detected with the naked eye,” Guarín said. “Our video analysis technique is so sensitive that it can identify things that the clinician cannot.”

The study found that even when clinicians thought a finger-tapping test appeared normal, video analysis using AI identified that people with Parkinson’s disease had smaller and slower movements than other groups, demonstrating the significance of this approach. 

Additionally, AI and video analysis could detect the sequence effect in people with iRBD and Parkinson’s disease. The sequence effect is the progressive decrease in movement amplitude and/or speed during repetitive movement, like tapping one's fingers. The origin and mechanisms of this motor sign are poorly understood, but its presence in both iRBD and Parkinson’s might indicate that the sequence effect is an early indicator of brain disorders. 

“Conducting simple, effective screening like this through standard video recordings, even those taken on a smartphone or webcam, could open the door to giving a brain diagnosis sooner and help those at greater risk of disease progression,” Guarín said.