Tracking Hyoid Bone Displacement During Swallowing Without Videofluoroscopy Using Machine Learning of Vibratory Signals.
Cara Donohue, Shitong Mao, Ervin Sejdić and 1 others
PMID 32419103WHAT IT FOUND
A neck sensor and machine learning located about half of the hyoid bone's position on each video frame during swallowing.
It could also distinguish reduced from normal hyoid movement. This is a promising non-invasive screening tool, but it is not yet accurate enough to replace imaging for diagnosis or treatment monitoring.
Key findings
01The machine learning algorithm detected the exact location of more than 50% of the bounding box containing the hyoid bone on each frame, with an average overlap of 51.6% in patients and 49.9% in healthy adults.
02Vibratory signal features, specifically standard deviation and spectral centroid, differed significantly between swallows rated as having normal versus reduced hyoid displacement on the MBSImP scale.
03Human experts performing frame-by-frame tracking of the hyoid bone themselves showed only 79.05% exact pixel-level agreement, indicating inherent variability in the gold standard method.
STILL TO COME
How it was doneWhat they foundWhat it means for PTsWhat it means for OTsWhat it means for SLPs
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 accuracy of the algorithm was approximately 50%, which the authors acknowledge is not high, though they argue it is notable given the small size of the hyoid bone. The study did not control for variables such as patient diagnosis, age, bolus characteristics, or head posture, which could affect the signal quality. The sample size for the MBSImP analysis was small (76 swallows), limiting the robustness of the findings regarding the distinction between normal and reduced displacement. The gold standard itself (human frame-by-frame tracking) has inherent variability, as evidenced by the 79% inter-rater agreement. This was a methodological feasibility study, not a clinical trial testing whether using this tool improves patient outcomes.
Declared interests
The research was supported by the National Institutes of Health (N.I.H., Extramural). The text does not mention any commercial conflicts of interest or funding from device manufacturers.
The easy way to misread this
Do not interpret the ability to distinguish 'normal' from 'reduced' hyoid displacement as clinical readiness for use. The algorithm's frame-by-frame location accuracy was only about 50%, and the authors explicitly state that further development and validation are required before this system can be deployed as a diagnostic or biofeedback tool in clinical settings.