In-situ hearing threshold estimation using Gaussian process classification

Abstract

One in six Americans suffers from hearing loss. While treatment with amplification is possible for many, the acceptance rate of hearing aids is low. Poor device fitting is one of the reasons. The hearing aid fitting starts with a detailed hearing assessment by a trained audiologist in a sound-controlled environment, using standard equipment. The hearing aid is adjusted step-by-step, following well-described procedures based on the audiogram. However, for many patients in rural settings, considerable travel time to a hearing center discourages them from receiving a hearing test and treatment. We hypothesize that hearing assessment with the patient's hearing aid can reliably substitute the hearing test in the clinic. Over-the-counter hearing aids could be programmed from a distance and fine-tuned by the hearing aid wearer. This study shows that a patient-controlled hearing assessment via a hearing aid in a non-clinical setting is not statistically different from an audiologist-controlled hearing assessment in a clinical setting. The differences in hearing obtained with our device and the Gaussian Process are within 3 dB of the standard audiogram. At 250 Hz, the sound delivery with the hearing aid used in this study added an additional reduction of sound level, which was not compensated.

Figure 1

The figure shows audiograms obtained from two study participants, one with normal hearing (panel (a) through (f)) and one with hearing loss (panel (h) through (l)). The traces in the top row with the “x” markers show the behavioral hearing thresholds at discrete frequencies, at which hearing was assessed during standard audiometry by a certified audiologist in a sound-reduced chamber with standard equipment. The green traces in the top row with the “o” markers show pure tone thresholds of the same patient obtained with a self-guided in-situ testing procedure using hearing aids in a sound-reduced environment. The second row shows the thresholds for the different methods at the same frequencies. The third row shows the difference between the methods. Both methods provided similar results for the hearing thresholds of the patients. The broken red line indicates the zero line or normal hearing.

Figure 2

The panels show plots of all audiograms captured with M1 and M2 in a study participant's left and right ear. In panels (a–c), the results for M1 at the standard frequencies are shown. Panels (d–f) show the corresponding results obtained with M2. Panels (g–i) provide the difference between the results obtained with M1 and M2. The green lines are the individual difference; the red circles show the average ± one standard deviation. The broken red line indicates the zero line.

Figure 3

The figure shows the differences for all data sets as a cumulative plot. The 50% difference is shifted to the right by 3.8 ± 7.2 (N = 35) on the right and 2.8 ± 7.9 dB (N = 35) on the left side.