Robust Data-Driven Auditory Profiling Towards Precision Audiology

Abstract

The sources and consequences of a sensorineural hearing loss are diverse. While several approaches have aimed at disentangling the physiological and perceptual consequences of different etiologies, hearing deficit characterization and rehabilitation have been dominated by the results from pure-tone audiometry. Here, we present a novel approach based on data-driven profiling of perceptual auditory deficits that attempts to represent auditory phenomena that are usually hidden by, or entangled with, audibility loss. We hypothesize that the hearing deficits of a given listener, both at hearing threshold and at suprathreshold sound levels, result from two independent types of “auditory distortions.” In this two-dimensional space, four distinct “auditory profiles” can be identified. To test this hypothesis, we gathered a data set consisting of a heterogeneous group of listeners that were evaluated using measures of speech intelligibility, loudness perception, binaural processing abilities, and spectrotemporal resolution. The subsequent analysis revealed that distortion type-I was associated with elevated hearing thresholds at high frequencies and reduced temporal masking release and was significantly correlated with elevated speech reception thresholds in noise. Distortion type-II was associated with low-frequency hearing loss and abnormally steep loudness functions. The auditory profiles represent four robust subpopulations of hearing-impaired listeners that exhibit different degrees of perceptual distortions. The four auditory profiles may provide a valuable basis for improved hearing rehabilitation, for example, through profile-based hearing-aid fitting.

Audiometric profiles and patterns of benefit: a data-driven analysis of subjective hearing difficulties and handicaps

“…using factor analysis. Second, the participants were stratified into four groups based on their degree of low-(HL LF ) and high-frequency (HL HF ) hearing loss as an approximation of the audiometric profiles of Sanchez-Lopez et al (2020). Third, to identify archetypal patterns of benefit, the overall data as well as the data belonging to each of the four subgroups were processed using an archetypal analysis.…”

“…This approach can be useful when applying the stratification to studies without supra-threshold measures retrospectively. However, this pre-classification in audiometric groups does not guarantee that the listener is correctly classified and supra-threshold tests can confirm the listener's auditory profile (Sanchez-Lopez et al 2020).…”

“…Therefore, information about supra-threshold auditory deficits, such as speech intelligibility in noise, might help to better understand the scale and scope of an individual's sensory impairments. Recently, Sanchez-Lopez et al (2020) proposed a stratification of hearing-impaired individuals into four clinically relevant subgroups, referred to as "auditory profiles." The auditory profiles were the result of a data-driven analysis that allowed the identification of four archetypal patterns of perceptual deficits along two largely independent dimensions, speech-intelligibility deficits and loudness perception deficits, respectively.…”

“…In the present study, subjective data from a questionnaire of hearing difficulties (SSQ) and a questionnaire of hearing handicap (HHQ) were analysed using a data-driven approach and following the principles of the "knowledge discovery from databases" (KDD; Frawley, Piatetsky-Shapiro, and Matheus 1992;Mellor, Stone, and Keane 2018). A data set of a clinical population of hearing-aid users was analysed in an attempt to uncover archetypal "benefit profiles" reflected in the subjective data that can be associated to specific audiometric groups based on the average audiometric thresholds of the four auditory profiles according to Sanchez-Lopez et al (2020). The goals of the study were (1) to identify the patterns of benefit associated with different audiometric groups, (2) to identify the priorities of hearing rehabilitation in terms of particular hearing difficulties and handicaps that need to be improved, and (3) to examine the possibilities of implementing a personalised post-fitting rehabilitation strategy.…”

“…Since the dataanalysis involved stratification of the participants in audiometrybased auditory profiles, the audiometric thresholds were also retrieved from the dataset but not used in the analysis. The audiometric thresholds were grouped into low-frequency ( 1 kHz; LF) and high-frequency (>1 kHz; HF) averages; for simplicity and comparability with Sanchez-Lopez et al (2020), only the better-ear averages were used for further analysis. Sanchez-Lopez et al (2020) did not include participants with average low-frequency hearing thresholds above 65 dB hearing level (HL LF ) in their data-driven profiling approach.…”