Category: Huntington's Disease
Objective: This study aims to identify key features relevant for predicting the clinical status of patients with Huntington’s Disease (HD) using machine learning techniques applied to raw voice recordings.
Background: Traditional assessment of HD patients involves time-consuming and complex tests. Speech has emerged as a potential biomarker for HD assessment, with recent studies proposing its use for predicting clinical scores [Nunes et al., 2024, PMID: 38322583; Riad et al., 2022, PMID: 35567614]. Machine learning methods offer a solution but raise concerns regarding model interpretability. Therefore, the investigation of features involved in prediction becomes a critical field of study that could significantly assist clinicians in interpreting results.
Method: Data from the BioHD (NTC01412125) and RepairHD (NCT03119246) studies were utilized, comprising 124 speakers (33 healthy, 91 with HD; mean age = 51.9 ± 10.5, Total Functional Capacity (TFC) = 10.1 ± 2.1). We replicated the study by Riad et al. to validate results across a larger cohort and to confirm feature’s weight in the regression (ElasticNetCV).
Results: Preliminary findings indicate reproducibility of Riad et al.’s results and regression performances are improved when trained on a larger data set. The weights of features in the regression are consistent with those reported in the initial study. [Riad et al., 2022, PMID: 35567614].
Conclusion: Understanding features used in prediction is crucial for clinicians to interpret machine learning predictions in relation to perceptible speech characteristics, such as pauses, silences, and pronunciation errors. While machine learning algorithms offer promising results, their interpretability remains a challenge. This study, by identifying key features in predicting clinical scores might be used to reverse-engineer features learned by an Attentive Convolutional Recurrent Neural Network (ACRRN) trained on spectrograms. This approach promises to enhance our comprehension and interpretation of the underlying mechanisms of the machine learning algorithm, facilitating more informed clinical decision-making.
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.
To cite this abstract in AMA style:
T. Le Ludec, C. Le Moine, A-C. Bachoud-Lévi, R. Massart. Indentifying Key Features for Predicting Clinical Status in Huntington’s Disease Patients Using Raw Speech Samples: a Lightweight Machine Learning Approach [abstract]. Mov Disord. 2024; 39 (suppl 1). https://www.mdsabstracts.org/abstract/indentifying-key-features-for-predicting-clinical-status-in-huntingtons-disease-patients-using-raw-speech-samples-a-lightweight-machine-learning-approach/. Accessed October 15, 2024.« Back to 2024 International Congress
MDS Abstracts - https://www.mdsabstracts.org/abstract/indentifying-key-features-for-predicting-clinical-status-in-huntingtons-disease-patients-using-raw-speech-samples-a-lightweight-machine-learning-approach/