Automatic ML-based vestibular gait classification: examining the effects of IMU placement and gait task selection.
Safa Jabri, Wendy Carender, Jenna Wiens and 1 others
PMID 36456966WHAT IT FOUND
A wearable sensor on the left arm combined with a walking-with-eyes-closed task correctly identified 82% of people with vestibular deficits and screened out 78% of controls.
This specific placement and task yielded the highest diagnostic accuracy among the options tested.
Key findings
01The most effective screening model used a single sensor on the left arm while participants walked with their eyes closed, achieving an accuracy score of 0.88.
02Using this specific sensor and task setup, the model identified 82% of participants with a vestibular diagnosis and correctly screened out 78% of age-matched controls.
03Participants with vestibular deficits showed significantly reduced left arm swing, characterized by lower angular velocities and smaller pitch displacements while walking with eyes closed compared to controls.
STILL TO COME
How it was doneWhat they foundWhat it means for PTs
Read the rest of this summary
You get three full summaries a month, free, and we do not ask for a card. Search, the TL;DRs and your library stay unlimited either way.
What it does not show
The sample size was small (30 participants), limiting the generalizability of the findings. The study only classified participants as either having a vestibular deficit or being a healthy control; it did not test whether the model could distinguish vestibular gait from other neurological gait disorders (e.g., Parkinson's or stroke). Nine participants wore face masks during data collection due to COVID-19 safety protocols, which may have altered visual input or gait mechanics for some tasks, though the authors note this likely did not affect the eyes-closed task. The study did not examine whether the side of the vestibular lesion (left vs. right) affected arm swing asymmetry due to limited sample size in specific subgroups.
Declared interests
The authors declared no conflicts of interest. The study was supported by U.S. Government and Non-U.S. Government research support.
The easy way to misread this
Do not interpret the 82% sensitivity and 78% specificity as evidence that this system is ready for clinical diagnosis. The model was trained on a very small sample of 30 people and was only tested on a binary distinction between vestibular patients and healthy controls. It has not been validated against other common causes of gait instability, such as orthopedic or neurological conditions, which could lead to false positives in a real-world clinic.