AI Watches a Video and Spots Movement Disorders in Kids and Adults
One day, a phone video could help diagnose rare neurological conditions faster and more fairly.
Doctors who diagnose movement disorders face a frustrating problem: two experts watching the same patient often disagree on what they see and how severe it is. Those conditions also tend to overlap — a child might have both tremor and dystonia — which makes consistent diagnosis even harder. So a team of researchers asked a simple question: could AI watch a standard video and describe what it sees the same way each time?
Their approach borrowed two existing AI tools. One, called SAM 3, breaks each video frame into detailed shapes — tracking how limbs, face and body move. The other, TabICLv2, sorts those movement patterns into eight known disorder types, like a very careful librarian filing symptoms into labeled drawers. Crucially, they froze the AI after training on just 21 adults and 4 healthy controls. It then ran unchanged on a separate group of 12 children and 20 adults with tremor, with only a small per-hospital adjustment.
The results were striking in the safety direction. When doctors agreed on a label, the AI's false alarms dropped to zero in both new groups. It caught dystonia perfectly (7 of 7 children, 15 of 15 adults) and chorea in all 3 children. Its main blind spot was myoclonus, a sudden-jerk condition. It also beat an older video-tracking method, scoring 0.93 versus 0.76 when doctors had to agree on labels.
The honest catch: this involved only a few dozen people, in a handful of clinics, and the authors say larger multi-centre trials are needed before any doctor uses it. But if it holds up, it hints at a future where a routine video could make diagnosis faster, cheaper and less dependent on who happens to be in the room.
- The AI analyzes normal video and identifies eight movement disorder types — no sensors or special equipment needed
- It learned from 21 people, then worked on children and adults it had never seen, with zero false alarms on cases doctors agreed on
- It missed myoclonus (sudden jerks) and has only been tested on a few dozen patients, so clinical use is still years away
Why It Matters
Faster, more consistent diagnoses could mean earlier treatment and fewer families bounced between specialists for years.