Classification of accurate and misarticulated /ɑr/ for ultrasound biofeedback using tongue part displacement trajectories.
Sarah R Li, Sarah Dugan, Jack Masterson and 7 others
PMID 35254181WHAT IT FOUND
A simple ultrasound parameter tracking just two tongue points classified accurate versus misarticulated /r/ sounds with 88.5% accuracy.
This matches the performance of complex full-tongue tracking, suggesting a simplified visual feedback display could reduce the cognitive load currently hindering therapy outcomes.
Key findings
01Using only the displacement of the tongue dorsum and blade at the midpoint of the sound, the classifier achieved 88.5% accuracy, which was not significantly different from the 89.2% accuracy achieved by tracking the full movement trajectory of the entire tongue.
02The study defined a single parameter (delta) based on these two tongue points that correlated strongly with clinician ratings of sound accuracy, meaning it can distinguish not just correct versus incorrect, but how close an attempt was to being correct.
03The authors propose that displaying this single parameter (e.g., as a needle on a meter) would allow patients to focus on the outcome of their movement rather than interpreting complex ultrasound images, potentially reducing the high cognitive load associated with current ultrasound biofeedback.
STILL TO COME
How it was doneWhat they foundWhat 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 study did not test the simplified feedback display on patients; it only proved that the underlying math works on recorded data. Ultrasound imaging quality was poor for some speakers due to anatomy, leading to the exclusion of about 18% of the recorded productions. Clinician agreement on sound accuracy was poor for sounds that were neither clearly correct nor clearly wrong (intermediate ratings), which likely contributed to the prediction errors in the regression model. The data was collected using a head stabilizer to keep the probe still, which is not standard practice in clinical settings; it is unclear if the results would hold up with handheld ultrasound probes.
Declared interests
The study was supported by the National Institute on Deafness and Other Communication Disorders (N.I.H., Extramural). The authors declared no competing interests.
The easy way to misread this
Do not assume this simplified display is currently available or proven to improve therapy outcomes. The study validated the algorithm's ability to classify tongue movements from recorded data, but it did not test whether using this simplified feedback actually helps patients learn to produce /r/ more effectively than standard therapy.