Evaluation of pupillometry as a diagnostic tool

Pupillometry is a widely used tool for measuring listening effort in hearing sci­ence. Changes in pupil dilation during speech-in-noise tasks have been linked to differences in speech intelligibility, masker type, hearing status and noise reduction schemes. Recent technological progress has allowed for pupillom­etry's use on a larger scale, thus opening its potential for clinical application where it can be relevant to measure hearing-impaired (HI) listeners' effort ex­penditure. However, the link between effort and the pupil response has not yet been validated for the individual listener, nor has the method's test-retest relia­bility been thoroughly evaluated. Moreover, an understanding of the impact of listener factors on the variation of the pupil response observed during speech recognition is still missing and little is known about the relationship between the evoked pupil response and the subjective effort investment perceived by the listener. This thesis assessed the individual pupillary response as an outcome measure of listening effort by investigating its reliability and sensitivity within speech-in-noise tasks.
The first study of this thesis assessed the reliability of a broad range of pupil features in normal-hearing (NH) and hearing-impaired (HI) individuals while performing a speech-in-noise test. It was found that some features of the pupil response (the rise and fall around the peak and the mean pupil dilation) showed high reliability independent of the listener group, while other pupil features' reliability varied depending on the listener group. Furthermore, a cluster analysis performed on the temporal characteristics of the pupil response showed that the signal-to-noise ratio (SNR) was, contrary to expectation, not a good predictor to cluster these pupil features.
The second study expanded the reliability analysis to include more SNRs, multiple visits and different normalization procedures. The results showed that data normalization procedures have a strong impact on the reliability of the pupil features. In particular, subtractive baseline correction in combination with a range normalization applied to the individual pupil response across all visits resulted in the highest reliability. Furthermore, the results suggested that the SNR and the number of visits only have a minor impact on the reliability of the pupil response. The most reliable pupil features were the traditional mean pupil dilation (MPD) and peak pupil dilation (PPD). The outcome of the first and second studies helped to identify test conditions and parameters as well as the pre-processing data analysis under which highly reliable pupil features can be obtained.
The third study explored the impact of individual listener factors, such as age, hearing status, cognitive abilities, motivation, and fatigue, on PPD and MPD and their variation across multiple visits. Furthermore, this study examined the effect of these listener factors on the dynamic range of the pupil response measured in several tasks (a speech-in-noise task, a cognitive task and at rest). The results identified motivation as the main listener factor affecting PPD and MPD. In addition, PPD was modulated by daily-life fatigue and age. At the same time, MPD was highly affected by the interaction of cognitive abilities with visits, resulting in changes in MPD across visits based on the listeners' cognitive abilities.
The final study investigated the sensitivity of the pupil response to changes in SNR and its relation to the perceived listening effort in a novel paired-sentence paradigm. The concept of a 'just noticeable difference in effort' (JND in effort) was introduced, reflecting the minimum increase in SNR necessary for a person to perceive a difference in perceived effort. The results were related to corre­sponding pupil responses at the JND in effort and two additional behavioural JNDs, the 'JND in clarity' and the 'JND in meaning' that have been reported in earlier studies. The results showed that, on average, the JND in effort was between the JND in clarity and JND in meaning but varied substantially across individuals. The pupil responses showed a difference between the pairs of sen­tences at the SNRs corresponding to the JND in effort and the JND in meaning for particular time-windows (i.e., retention period and listening time, respectively), whereas no difference between the pairs was found at the SNRs corresponding to the JND in clarity.
Together, the findings of this thesis suggest that pupillometry has potential for future applicability as a clinical measure of individual listening effort. More specifically, depending on the test conditions (e.g., the SNR) and the normal­ization procedures, highly reliable pupil features can be obtained, which is a prerequisite for a clinically feasible measure. However, listener factors have been shown to contribute to the variability in the pupil response, meaning that such factors need to be considered when interpreting the pupil response. Finally, the assessment of the behavioral JND in effort appears very relevant for the interpretation of the individual's pupil response as a marker of effort investment. Overall, this work may provide a valuable basis for developing a clinical tool to assess listening effort, which will facilitate more comprehensive evaluations of speech communication that extend beyond audibility and speech intelligibility.