Improving AI Interpretability for Multilingual Parkinson’s Disease Classification through Voice Analysis

Daria Hemmerling, Michal Zakrzewski, Marek Wodzinski, Milosz Dudek, Filip Gaciarz, Magdalena Wojcik-Pedziwiatr, Juan Rafael Orozco-Arroyave, Elmar Noth, David Sztaho, Taras Rumezhak
Proceedings of The First AAAI Bridge Program on AI for Medicine and Healthcare, PMLR 281:49-55, 2025.

Abstract

Addressing the imperative need for interpretability in medical models based on machine learning and artificial intelligence, our study focuses on the crucial task of Parkinson’s disease detection. In this paper, we introduce a vision transformer incorporating multilingual vowel phonations, achieving a classification accuracy of 89%. To enrich the input representation for vision transformer, we utilized images of melspectrograms and regular spectrograms. The success of our model goes beyond performance metrics, as we strategically integrate explainable artificial intelligence techniques. The synergy between robust classification results and explainability underscores the effectiveness of our approach in opening the black-box nature of neural networks. This, in turn, contributes to enhanced medical decision-making and reinforces the potential of artificial intelligence in advancing diagnostic methodologies for Parkinson’s disease.

Cite this Paper

BibTeX
@InProceedings{pmlr-v281-hemmerling25a, title = {Improving AI Interpretability for Multilingual Parkinson’s Disease Classification through Voice Analysis}, author = {Hemmerling, Daria and Zakrzewski, Michal and Wodzinski, Marek and Dudek, Milosz and Gaciarz, Filip and Wojcik-Pedziwiatr, Magdalena and Orozco-Arroyave, Juan Rafael and Noth, Elmar and Sztaho, David and Rumezhak, Taras}, booktitle = {Proceedings of The First AAAI Bridge Program on AI for Medicine and Healthcare}, pages = {49--55}, year = {2025}, editor = {Wu, Junde and Zhu, Jiayuan and Xu, Min and Jin, Yueming}, volume = {281}, series = {Proceedings of Machine Learning Research}, month = {25 Feb}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v281/main/assets/hemmerling25a/hemmerling25a.pdf}, url = {https://proceedings.mlr.press/v281/hemmerling25a.html}, abstract = {Addressing the imperative need for interpretability in medical models based on machine learning and artificial intelligence, our study focuses on the crucial task of Parkinson’s disease detection. In this paper, we introduce a vision transformer incorporating multilingual vowel phonations, achieving a classification accuracy of 89%. To enrich the input representation for vision transformer, we utilized images of melspectrograms and regular spectrograms. The success of our model goes beyond performance metrics, as we strategically integrate explainable artificial intelligence techniques. The synergy between robust classification results and explainability underscores the effectiveness of our approach in opening the black-box nature of neural networks. This, in turn, contributes to enhanced medical decision-making and reinforces the potential of artificial intelligence in advancing diagnostic methodologies for Parkinson’s disease.} }
Endnote
%0 Conference Paper %T Improving AI Interpretability for Multilingual Parkinson’s Disease Classification through Voice Analysis %A Daria Hemmerling %A Michal Zakrzewski %A Marek Wodzinski %A Milosz Dudek %A Filip Gaciarz %A Magdalena Wojcik-Pedziwiatr %A Juan Rafael Orozco-Arroyave %A Elmar Noth %A David Sztaho %A Taras Rumezhak %B Proceedings of The First AAAI Bridge Program on AI for Medicine and Healthcare %C Proceedings of Machine Learning Research %D 2025 %E Junde Wu %E Jiayuan Zhu %E Min Xu %E Yueming Jin %F pmlr-v281-hemmerling25a %I PMLR %P 49--55 %U https://proceedings.mlr.press/v281/hemmerling25a.html %V 281 %X Addressing the imperative need for interpretability in medical models based on machine learning and artificial intelligence, our study focuses on the crucial task of Parkinson’s disease detection. In this paper, we introduce a vision transformer incorporating multilingual vowel phonations, achieving a classification accuracy of 89%. To enrich the input representation for vision transformer, we utilized images of melspectrograms and regular spectrograms. The success of our model goes beyond performance metrics, as we strategically integrate explainable artificial intelligence techniques. The synergy between robust classification results and explainability underscores the effectiveness of our approach in opening the black-box nature of neural networks. This, in turn, contributes to enhanced medical decision-making and reinforces the potential of artificial intelligence in advancing diagnostic methodologies for Parkinson’s disease.
APA
Hemmerling, D., Zakrzewski, M., Wodzinski, M., Dudek, M., Gaciarz, F., Wojcik-Pedziwiatr, M., Orozco-Arroyave, J.R., Noth, E., Sztaho, D. & Rumezhak, T.. (2025). Improving AI Interpretability for Multilingual Parkinson’s Disease Classification through Voice Analysis. Proceedings of The First AAAI Bridge Program on AI for Medicine and Healthcare, in Proceedings of Machine Learning Research 281:49-55 Available from https://proceedings.mlr.press/v281/hemmerling25a.html.

Related Material