Prediction of acoustic tinnitus suppression using resting state EEG: An explainable AI approach

Payam S. Shabestari,Stefan Schoisswohl,Zino Wellauer,Adrian Naas,Tobias Kleinjung,Martin Schecklmann,Berthold Langguth,Patrick Neff
DOI: https://doi.org/10.1101/2024.04.16.589690
2024-04-19
Abstract:Tinnitus, characterized by the perception of sound without an external source, affects a significant portion of the population and can lead to considerable individual suffering, yet understanding of its suppression remains limited. Understanding neural traits of tinnitus suppression may be crucial for developing accurate predictive models in tinnitus research and treatment. This study aims to classify individuals capable of brief acoustic tinnitus suppression (BATS; also known as residual inhibition) based on their independent resting state EEG (n=102), exploring the classification’s robustness on various sample splits, and the relevance of resulting specific EEG features in the spirit of explainable AI. A comprehensive set of EEG features, including band power in standard frequency bands, spectral entropy, aperiodic slope and offset of the power spectrum, and connectivity, was included in both sensor and source space. Binary classification of the BATS status was performed using a comprehensive set of standard classifiers and Pearson correlation for feature selection, which addresses multicollinearity, avoiding complex dimensionality reduction techniques. Feature importance was assessed using Gini impurity metrics, allowing interpretation of the directionality of identified neural features. The Random Forest model showed the most consistent performance, with its majority voting mechanism effectively reducing overfitting and providing reliable predictions, and was therefore chosen for subsequent feature interpretation analysis. Our classification task demonstrated high accuracy across the various BATS split thresholds, suggesting that the choice of threshold does not significantly influence the underlying pattern in the data. We achieved classification accuracies of 98% for sensor and source models and 86% for the connectivity model in the main split. Looking at identified important features, our findings align with and extend existing neuroscience research in tinnitus by discovering highly specific and novel neural features in naive resting-state data predictive of BATS. Gamma power is identified as the most important feature in the sensor model, followed by alpha power, which fits current models of sensory processing, prediction, and updating (gamma) as well as inhibitory (alpha) frameworks. The overall spectral shape of the EEG power spectrum tends to be more normal in +BATS individuals, as reflected in the aperiodic offset and slope features. In the source model, important features are lateralized in that the gamma feature is more prominent in the left core auditory network, whereas the alpha feature is distributed more sparsely over the right hemisphere in line with auditory attention data. Furthermore, we identified several hotspots in the temporal, insular, parietal, parahippocampal, medial prefrontal, and (posterior) cingulate cortex implicated in sensory processing, gating, attention, and memory processes. Relevant network features were found in a hyperconnected bilateral auditory network (within the network), while the full auditory network was hyperconnected to limbic regions (between networks), which may reflect an intact sensory gating mechanism aiding tinnitus suppression. This study’s implications extend to improving the understanding and prediction of tinnitus loudness perception and tinnitus distress as well as its (acoustic) suppression. Furthermore, our approach underscores the importance of careful feature selection, model choice, and validation strategies in analyzing complex neurophysiological data.
Neuroscience
What problem does this paper attempt to address?
The main problem that this paper attempts to solve is: **How to use resting - state electroencephalogram (resting state EEG) to predict whether an individual can briefly suppress tinnitus through acoustic stimulation (Brief Acoustic Tinnitus Suppression, BATS)**. Specifically, the research aims to: 1. **Classify individuals**: Based on resting - state EEG data, distinguish between individuals who can perform BATS and those who cannot. 2. **Explore feature robustness**: Evaluate the robustness of classification under different sample partitions and determine which specific EEG features have an important impact on classification. 3. **Interpret AI models**: Understand the neuroscientific significance of these features through Explainable AI methods. ### Research Background Tinnitus is a common auditory hallucination phenomenon, that is, perceiving sound without an external sound source. Although tinnitus causes a significant psychological burden on many people, its suppression mechanism has not been fully understood. Understanding the neural characteristics of tinnitus suppression is crucial for developing accurate prediction models and treatment regimens. ### Research Methods - **Data sources**: The research used datasets from two independent laboratories at the University of Zurich in Switzerland and the University of Regensburg in Germany, which included resting - state EEG data from a total of 102 subjects. - **Feature extraction**: A series of frequency - domain features were extracted from the EEG data, including power spectra in standard frequency bands, spectral entropy, aperiodic slope and offset, etc. - **Classification models**: Multiple standard classifiers (such as random forest, gradient boosting, support vector machine, etc.) were used, and features were selected through Pearson correlation coefficients to avoid the problem of multicollinearity. - **Feature importance assessment**: The importance of features was evaluated through the Gini impurity index, and the directional influence of features was explained using SHAP values. ### Main Findings - **High classification accuracy**: In the sensor - space model, the random forest model achieved a 98% classification accuracy; in the source - space model, the accuracy was 86%. - **Key features**: Gamma power and alpha power are the most important features. Gamma power is more prominent in the core auditory network of the left hemisphere, while alpha power is more distributed in the right hemisphere. - **Neural mechanisms**: The study shows that individuals with BATS ability exhibit a more "normal" EEG power - spectrum shape at rest, which is reflected in aperiodic offset and slope features. ### Significance This research not only helps to deepen the understanding of the neural mechanisms of tinnitus suppression but also provides new ideas for future diagnosis and personalized treatment. Through automatic classification algorithms and Explainable AI methods, the research reveals specific brain - activity patterns related to BATS, laying the foundation for further research and clinical applications.