Hali Lindsay (2), Elisabeth Kasper(3), Judith Baltes(4), Martha Kring(1,8), Anja Schneider(4,5), Stefan Vielhaber(6,7), Judith Machts(6,7), Susanne Vogt(6), Johannes Prudlo(3,8), Johannes Tröger (2), Prof. Dr. Andreas Hermann (1,8)
1. Translational Neurodegeneration Section “Albrecht Kossel”, Dept. Neurology, University Medical Center Rostock, Universitätsmedizin Rostock + Magdeburg. 2. ki elements GmbH, Saarbrücken, Germany. 3. Department of Neurology, University Medical Centre, Rostock, Germany. 4. German Center for Neurodegenerative Diseases (DZNE), Bonn, Germany. 5. Department of Neurodegenerative Diseases and Geriatric Psychiatry, UniversityHospital Bonn, Bonn, Germany. 6. Department of Neurology, Otto-von-Guericke University, Magdeburg, Germany. 7. German Centre for Neurodegenerative Diseases (DZNE), Magdeburg, Germany. 8. German Centre for Neurodegenerative Diseases(DZNE), Rostock/Greifswald, Rostock, Germany.
* Poster presented at the ENCALS Meeting 2024, Sweden
Abstract
Speech biomarkers have emerged as promising tools in Amyotrophic Lateral Sclerosis (ALS), as alterations in speech can reflect the motor neuron degenerative patterns. Employing speech features for monitoring ALS is non-invasive and objective, facilitating early detection and longitudinal tracking. By making use of speech features, uncover novel methods for diagnosis, prognosis, accelerated drug development, and evaluation of treatment response.
The Cookie Theft Picture (CTP) assesses speech quality and linguistic ability. Previous research identified noise and spectral features as ALS speech indicators. Perturbation features measure voice irregularities, including Jitter for pitch instability and Shimmer for vocal intensity fluctuations, crucial for ALS assessment.
This analysis considered 31 German ALS patients (mean age = 58.48 years, SD = 11.40; mean total ALSFRS= 35.97, SD = 7.07; mean ALSFRS-R Bulbar = 10.00, SD = 2.24; mean ALSFRS-R Respiratory = 10.71, SD = 1.60; mean MoCA = 25.77, SD = 2.72, excluding Frontotemporal Dementia) from Universitätsmedizin Rostock in Magdeburg, Germany. Verbal responses to the CTP were recorded and manually transcribed. Subsequently linguistic and acoustic features were extracted using the ki:elements proprietory speech processing pipeline, SIGMA. We calculated partial correlations between speech features and the ALSFRS-R bulbar, respiratory and total score, adjusting for age and cognitive function (MoCA).
The bulbar score correlated with features related to vocal intensity (shimmer: r(29) = .33 to .36, p <.05); rate of loudness peaks: r(29) = .51, p = .00), and vocal fold instability (jitter: r(29) = .35 to .39, p < .03). Furthermore, rate of adverbs (r(29) = .4, p = .02), and the verb phrase with auxiliary rate (r(29) = .43, p = .01), declined as the bulbar score deteriorated. Moreover, acoustic features exhibited correlations with both the respiratory score and ALSFRS-R score: features denoting the change in vocal tract and articulation (F1 relative mean energy: r(29) = .33-.34, p = .04; F2 relative energy standard deviation: r(29) = .33-.39, p < .04). Lastly, the total score displayed correlations with acoustic features related to articulation (r(29)=.32-.39, p < 0.05).
Automatically extracted speech features are significantly correlated with the bulbar, respiratory, and total ALSFRS-R scores. These methods offer a systematic approach for non-invasive, continuous monitoring of ALS disease progress.
