Sparse machine learning of resting-state fMRI reveals brain-wide dysconnectivity in hyperacusis.
Authors
Affiliations (9)
Affiliations (9)
- Neuroscience Program, University of Illinois Urbana-Champaign, Champaign, IL, USA. [email protected].
- Beckman Institute for Advanced Science & Technology, University of Illinois Urbana-Champaign, Champaign, IL, USA. [email protected].
- Neuroscience Program, University of Illinois Urbana-Champaign, Champaign, IL, USA.
- Beckman Institute for Advanced Science & Technology, University of Illinois Urbana-Champaign, Champaign, IL, USA.
- Department of Speech and Hearing Science, University of Illinois Urbana-Champaign, Champaign, IL, USA.
- Department of Psychology, University of Illinois Urbana-Champaign, Champaign, IL, USA.
- Neuroscience Program, University of Illinois Urbana-Champaign, Champaign, IL, USA. [email protected].
- Beckman Institute for Advanced Science & Technology, University of Illinois Urbana-Champaign, Champaign, IL, USA. [email protected].
- Department of Speech and Hearing Science, University of Illinois Urbana-Champaign, Champaign, IL, USA. [email protected].
Abstract
Loudness hyperacusis may alter brain function beyond the discomfort elicited by regular sound levels. Yet, the neuroscientific literature of hyperacusis has largely focused on the sensory neural components, and often in the context of other comorbid conditions. Our goal was to investigate brain-wide neural interactions associated with loudness hyperacusis using resting-state fMRI and machine learning classification. Fourteen young, healthy adults experiencing hyperacusis were recruited and compared to twenty-five age-, gender-, and education-matched control individuals. All participants had normal hearing thresholds and they were classified as having hyperacusis based on having a score greater than 22 on the Hyperacusis Questionnaire (HQ). Functional connectivity measures were used in a machine learning model that distinguished participants with hyperacusis from controls. Model weights were further analyzed systematically to reveal the cognitive brain networks and regional hubs where functional coupling was significantly altered in hyperacusis. Here, we observe that participants with hyperacusis are distinguishable from control individuals using a functional connectivity-based classification model, which yields a classification F1-score of 0.679. Owing to optimized feature selection, the model coefficients capture highly specific neural connectivity differences between the groups, including brain regions and networks implicated in semantic processing, working memory, emotion processing, and self-regulation. Furthermore, network connectivity measures, scaled by model-informed coefficients, explain up to 53% of the variance in individual HQ scores. Through rigorous data-driven modeling, we characterize the reduced sound tolerance condition of loudness hyperacusis as being associated with atypical spontaneous connectivity across cognitive networks that extend beyond the auditory system. Such improved knowledge of the condition validates patient experiences and has implications for future treatments and assessments.