Automated audiogram image recognition feature in Phone innovative approach for Personalized Sound Amplification: Study
In August 2022, the US Food and Drug Administration issued the final rule to establish a new category of over-the counter (OTC) hearing aids. Over-the-counter hearing aids are intended for use by adults with mild-to-moderate hearing loss without the need for an audiologist to adjust and program the device for the user’s level of hearing as is currently done for most conventional...
In August 2022, the US Food and Drug Administration issued the final rule to establish a new category of over-the counter (OTC) hearing aids. Over-the-counter hearing aids are intended for use by adults with mild-to-moderate hearing loss without the need for an audiologist to adjust and program the device for the user’s level of hearing as is currently done for most conventional hearing aids. One key aspect of the effectiveness of these future OTC hearing aids is provision of the correct level of sound amplification based on the individual’s level of hearing. Thus, a feature to import prior hearing test (audiogram) results into these devices may be an important consideration to ensure optimal amplification of sound.
A 2021 Apple iOS 15 update (Apple Inc, Cupertino, California) included features termed “headphone accommodation” and “conversation boost” that allow the sound output from compatible earphones to be customized based on a user’s prior audiogram and may be the prelude to future OTC hearing aids. Users have the option of manually inputting audiogram results or relying on an automated image recognition system that extracts hearing data from an image of the audiogram.
The study by Janet S et al examined the accuracy of the automated audiogram image recognition in the iPhone using a large sample of audiogram reports representing various degrees of hearing loss. This study was approved by the institutional review board at University of Minnesota. Authors collected over 1000 deidentified audiogram images from patients aged 18 years or older who had hearing test data available for bilateral ears from January 2021 to March 2021 and their corresponding pure-tone threshold raw data from the health system clinical hearing database. Five hundred audiograms were randomly sampled based on the worse hearing ear speech-frequency pure-tone average (PTA, average of air-conduction thresholds at 0.5, 1.0, 2.0, and 4.0 kHz) to attain equal distribution of various levels of hearing. Images in PDF format of the 500 sampled audiograms were uploaded to an iPhone (iOS 15.4.1) with enabled headphone accommodation. Air conduction hearing thresholds based on the automated audiogram image recognition were compared with the raw data. Accurate identification of the thresholds was defined as correct recognition of the speech-frequency pure-tone average within 5 dB HL. Authors further explored the hearing characteristics (laterality, asymmetry, severity, and type) associated with accuracy of the audiogram image importation.
The accuracy of the automated audiogram image recognition was 7.0% and 11.8% for right and left ears, respectively. The feature was unable to detect adequate thresholds to calculate PTA for over 80% of ears, prompting the user to manually enter the audiogram data. Accuracy was significantly higher when hearing was asymmetric (OR, right ear 2.26 and left ear 2.27). Type, degree of hearing loss, and number of missing thresholds were not associated with accuracy. Additional analyses of accuracy were performed for an alternative audiogram format presenting the audiogram for left and right ears separately. The accuracy remained low at 7.4% and 4.6% for right and left ears, respectively.
The automated audiogram image recognition feature in the iPhone is an innovative approach to potentially allow for easier importation of audiogram data and customization of sound output for OTC hearing aids. However, the accuracy of automated recognition of hearing thresholds was low and in 80% to 90% of cases the user is prompted to manually input audiogram data.
This study has limitations including the use of digitally generated audiograms from a single institution. The accuracy of the image recognition in the clinical setting accounting for factors such as poor image quality, hand-written symbols, and various audiogram report formats may be lower. Future efforts will be needed to improve the accuracy of automated audiogram image recognition if this approach is used for the customization of future OTC hearing aids. Other approaches to customization of OTC hearing aids such as insitu audiometric testing of the user’s hearing should be explored.
Source: Janet S. Choi, Tyler J. Gathman, Frank R. Lin, Meredith E. Adams; JAMA Otolaryngology–Head & Neck Surgery
doi:10.1001/jamaoto.2022.4383
