ki:elements

Using acoustic speech biomarkers to detect isolated REM sleep behavior disorder

Johannes Tröger 1, Elisa Mallick 1, Sinah Röttgen 2, Ebru Baykara 1, Nicklas Linz 1, Tabea Thies 2, Michael Barbe 2, Michael Sommerauer 2

1) ki elements GmbH, Germany. 2) Department of Neurology, Faculty of Medicine and University Hospital Cologne, University of Cologne, Cologne, Germany.

* Poster presented at the AD/PD™ Alzheimer’s disease and Parkinson’s disease Conference 2024, Portugal

Abstract

Aims

Isolated REM sleep behavior disorder (iRBD) is an early alpha-synucleinopathy and can precede overt motor Parkinson’s Disease (PD) for decades. Speech changes have proven to be a robust and sensitive indicator of motor-function and coordination impairment in PD. Recently automatic speech analysis has shown to be an early biomarker of incipient motor dysfunction in individuals with iRBD (Hlavnička et al., 2017). The goal of this research is to evaluate the feasibility of using speech biomarkers to automatically detect iRBD as compared to healthy controls (HC).

Methods

This feasibility research is based on German pilot data with 27 iRBD (4F, mean age = 61,43 ± 8,59) and 30 HC (10F, mean age = 64,67 ± 5,71) with recordings from maximum phonation of the vowel /a/ and fast syllable repetitions of the sequence /pataka/. From the recordings, acoustic features were extracted using the Dysarthria Analyser Software (DYSAN) (Hlavnicka, 2019). We normalized acoustic features for age and gender effects and subsequently trained extra trees machine learning classifiers using leave one out cross-validation and feature selection.

Results

The extra trees model was able to differentiate between iRBD and HC with an AUC of .71 (sensitivity and specificity both around .67).

Conclusions

The results show that acoustic speech biomarkers could be potentially utilized to screen for iRBD. Future research has to show whether this result persists in a study with a larger population and different languages. Results might impact future applications both in clinical trials as well as healthcare.

References

  • Hlavnička, J. (2019). Automated analysis of speech disorders in neurodegenerative diseases (Doctoral dissertation, Czech Technical University).
  • Hlavnička, J., Čmejla, R., Tykalová, T., Šonka, K., Růžička, E., & Rusz, J. (2017). Automated analysis of connected speech reveals early biomarkers of Parkinson’s disease in patients with rapid eye movement sleep behaviour disorder. Scientific reports, 7(1), 12.
Share this article