Objective: We propose a machine learning-based method for inferring the clinical status of patients with Huntington’s Disease (HD) from raw voice recordings.

Background: Assessing HD patients usually requires a battery of tests that are both time-consuming and challenging to implement. To facilitate this assessment, a number of works propose to use speech, an emerging biomarker of HD, to predict the clinical scores traditionally used for the disease characterisation [Nunes et al., 2024, PMID: 38322583; Riad et al., 2022, PMID: 35567614]. While achieving honourable accuracy in clinical score regression (cUHDRS, TMS, TFC, …), these approaches require the potentially incomplete selection and costly extraction of hand-crafted features, mostly derived from human annotations, making these difficult to deploy in a care setting.

Method: Data were collected as part of the BioHD (NTC01412125) and RepairHD (NCT03119246) studies. They involved 124 speakers, 33 healthy and 91 with HD (age = 51.9 ± 10.5, TFC = 10.1 ± 2.1). The data includes both scripted and spontaneous speech that is segmented in speech segments of up to 10s length. In contrast to previous approaches, we make no a priori assumptions about the speech HD correlates. We compute mel spectrograms to holistically represent time-frequency variations in speech. We designed a regressor based on an Attentive Convolutional Recurrent Neural Network (ACRNN) that takes advantage of the supra-segmental disease correlates conveyed through speech. We trained it in a 5-fold cross validation paradigm to predict the motor, dependence and functional scores of the UHDRS using a mean accuracy error loss.

Results: Preliminary results show that our approach predicts the UHDRS motor, functional and dependency scores with an accuracy at least comparable to the best models published to date [Riad et al., 2022, PMID: 35567614].

Conclusion:

By leveraging the ability of neural networks to capture information within the speech signal, our approach overcomes the need for human-annotated features, bringing the clinical assessment of HD patients one step closer to automation. The next step would be to improve this performance by feeding the model with other informative inputs, such as cognitive data from the SelfCog [Lunven et al., 2023, PMID: 36938527].

References: Nunes AS, Pawlik M, Mishra RK, Waddell E, Coffey M, Tarolli CG, Schneider RB, Dorsey ER, Vaziri A, Adams JL. Digital assessment of speech in Huntington disease. Front Neurol. 2024 Jan 23;15:1310548. doi: 10.3389/fneur.2024.1310548. PMID: 38322583; PMCID: PMC10844459.

Riad R, Lunven M, Titeux H, Cao XN, Hamet Bagnou J, Lemoine L, Montillot J, Sliwinski A, Youssov K, Cleret de Langavant L, Dupoux E, Bachoud-Lévi AC. Predicting clinical scores in Huntington’s disease: a lightweight speech test. J Neurol. 2022 Sep;269(9):5008-5021. doi: 10.1007/s00415-022-11148-1. Epub 2022 May 14. PMID: 35567614; PMCID: PMC9363375.

Lunven M, Hernandez Dominguez K, Youssov K, Hamet Bagnou J, Fliss R, Vandendriessche H, Bapst B, Morgado G, Remy P, Schubert R, Reilmann R, Busse M, Craufurd D, Massart R, Rosser A, Bachoud-Lévi AC. A new approach to digitized cognitive monitoring: validity of the SelfCog in Huntington’s disease. Brain Commun. 2023 Mar 6;5(2):fcad043. doi: 10.1093/braincomms/fcad043. PMID: 36938527; PMCID: PMC10018460.

« Back to 2024 International Congress

MDS Abstracts - https://www.mdsabstracts.org/abstract/predicting-clinical-status-of-patients-with-huntingtons-disease-from-raw-speech-samples-a-lightweight-machine-learning-approach/