Online Machine Learning Audiometry.
Dennis L Barbour, Rebecca T Howard, Xinyu D Song and 8 others
PMID 30358656WHAT IT FOUND
Online machine learning audiometry gave hearing threshold estimates close to online Hughson-Westlake audiometry, but repeat results were less consistent.
In about 10% of ears, 49 tones were too few for reliable extreme-frequency estimates.
Key findings
01At the 6 standard audiogram frequencies, online machine learning thresholds differed from online Hughson-Westlake thresholds by a mean signed difference of -0.969 ± 6.02 dB and a mean absolute difference of 3.24 ± 5.15 dB.
02On repeat testing, online machine learning thresholds had a mean absolute difference of 2.85 ± 6.57 dB, while online Hughson-Westlake thresholds had a mean absolute difference of 1.88 ± 3.56 dB.
03In about 10% of the ears, 49 tones were apparently too few for the current online machine learning method to achieve reliable estimates at extreme frequencies.
STILL TO COME
How it was doneWhat they found
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
Only 21 participants were tested, and they had to be fluent English speakers with no neurological disorder, so this does not show how the method works in a broad patient group. The online machine learning method is still in active development, so results may change when the algorithm is updated. In about 10% of ears, 49 tones were apparently too few for reliable estimates at extreme frequencies. The study reported threshold agreement and test-retest differences, not whether the method changes clinical decisions. The comparison was structured to favor the standard Hughson-Westlake method, so the machine learning performance reported here may be a lower bound. Variability was greater at 8 kHz and 250 Hz, and the authors did not explain that variance.
Declared interests
No author conflicts statement is included in the supplied text. The article is listed as supported by NIH and U.S. and non-U.S. government research sources.
The easy way to misread this
Do not conclude that online machine learning audiometry is ready to replace online Hughson-Westlake testing. The study showed close threshold agreement in 21 participants, but repeat machine learning estimates were less consistent than online Hughson-Westlake estimates, and extreme frequencies were sometimes undersampled.