AI model detects infant neurodevelopmental impairment from motion videos; activity holds steady when cardiology inpatients self-monitor

A pair of small hexagonal hand dumbbells lying side by side next to an amber resistance band tied in a loose loop with a small knot.

Researchers in China reported in npj Digital Medicine that a generative AI model trained on more than 1,300 infant motion videos detected neurodevelopmental impairment at a level equivalent to clinical experts. In the Netherlands, a JMIR Cardio study found that the activity of cardiology inpatients stopped declining after they could see their own accelerometer data in real time.

  • AI Oct 8, 2026
    Researchers in China trained and validated a generative AI model, DeepAD, on more than 1,300 infant motion videos collected from 2019 to 2024 at three children's hospitals in China, according to a paper published Oct. 8 in npj Digital Medicine. The videos included about 50 infants diagnosed with cerebral palsy. The authors said the model reached an equivalent level to clinical experts in detecting neurodevelopmental impairment.
  • Technology Oct 7, 2026
    At University Medical Center Utrecht in the Netherlands, daily physical activity of cardiology inpatients declined significantly, by about 6% per month, when only health care practitioners could see data from a PAM AM400 accelerometer (75 patients), according to an interrupted time series study published Oct. 7 in JMIR Cardio. After patients were given real-time feedback on their activity (84 patients), the decline was no longer observed, and the time trends differed significantly between phases (P = .03). The authors said seasonal influences cannot be excluded and that self-monitoring may support activity maintenance during hospital stays.
    Source: JMIR Cardio

The weekly innovation roundup for physical therapy, published every Sunday for the week before; every item links to its sources. Latest physical therapy news · All specialties