Tabea Thies, Felix Dörr, Andreas Rouvalis, Nicklas Linz, Alexandra König, Raffaele Dubbioso, Anja Schneider, Andreas Hermann, Jan Rusz & Johannes Tröger
* Poster presented at the 35th International Symposium on ALS/MND, Montreal (Canada)
Abstract
Background: Recent studies indicate that automatic speech analysis can detect acoustic changes in people with Amyotrophic Lateral Sclerosis (pwALS). Speech symptoms can be identified through acoustic analysis before speech intelligibility declines significantly. This enables noninvasive remote monitoring and early identification of pwALS.
Objectives: The aim is to build an automatic speech-based composite score for dysarthria based on sustained phonation of the vowel /a/ and validate it across different languages in separating pwALS from healthy controls (HC).
Methods: Three different datasets were used: 1) Czech data containing 21 pwALS (mean age 65.0 ± 0.0; 15 female; mean ALSFRS-R 35.48 ± 6.35) and 44 HC (mean age 61.68 ± 10.87; 17 female) [1], 2) Italian data containing 102 pwALS (mean age 62.76 ± 11.5; 37 female; mean ALSFRS-R 34.46 ± 8.43) and 51 HC (mean age 62.98 ± 10.45; 19 female) [2] and 3) German data containing 49 pwALS (mean age 63.29 ± 9.34; 14 female; no ALSFRS-R available) and 47 HC (mean age 59.51 ± 13.46; 30 female) [3]. The ki:elements SIGMA library was used to extract acoustic features from the audio recordings. Using the Italian dataset, Spearman Rank-Sum partial correlations were computed between these features and the ALSFRS-R score and its speech item, controlling for age and sex. Features with a p-value < .05 were selected to build the dysarthria score. To assess generalizability, Kruskal-Wallis tests were conducted between ALS and HC in all samples partialling out the effects of age and sex.
Results: A combination of 32 features was used to determine the dysarthria score. In the Italian dataset, pwALS (Mdn = – 0.11) differed significantly from HC (Mdn = 0.23) with regards to their dysarthria score (H = 13.179, p < .001, η2 = 0.081). Similarly, the Czech pwALS (Mdn = -0.54) differed significantly to the Czech HC (Mdn = 0.26; H = 16.435, p < .001, η2 = 0.245). In the German dataset, the difference was non-significant (H = 0.326, p = .568, η2 = 0.010).
Discussion: The results show that an automatic speech-based composite score for dysarthria based on sustained vowel phonation recordings can identify pwALS across different languages. Future research should extend clinical validation to tracking disease stages and change over time.
References: [1] Chiaramonte, R., & Bonfiglio, M. (2020). Acoustic analysis of voice in bulbar amyotrophic lateral sclerosis: a systematic review and meta-analysis of studies. Logopedics Phoniatrics Vocology, 45(4), 151-163. [2] Novotny, M., Melechovsky, J., Rozenstoks, K., Tykalova, T., Kryze, P., Kanok, M., … & Rusz, J. (2020). Comparison of automated acoustic methods for oral diadochokinesis assessment in amyotrophic lateral sclerosis. Journal of Speech, Language, and Hearing Research, 63(10), 3453-3460. [3] VOC-ALS Database, VOiCe signals acquired in Amyotrophic Lateral Sclerosis Patients and Healthy Control. DOI: https://doi.org/10.7303/syn53009474 [4] Tröger, J., Baltes, J., Baykara, E., Kasper, E., Kring, M., Linz, N., … & Hermann, A. (2023). PROSA—a multicenter prospective observational study to develop low-burden digital speech biomarkers in ALS and FTD. Amyotrophic Lateral Sclerosis and Frontotemporal Degeneration, 24(7-8), 589-598.
Acknowledgments: This work has been partially funded by the Target ALS foundation under grant agreement number BM-2022-C2-L1, by the Czech Ministry of Health (grant no. NW24-04-00211) and by the Italian Ministry of University and Research (MUR) (grant no. E53D23011330006).
