Publicação
Towards the Automatic Diagnosis of Amyotrophic Lateral Sclerosis from Speech
| Resumo: | Amyotrophic Lateral Sclerosis (ALS) is an incurable neurodegenerative disease affecting motor neurons. ALS diagnosis requires extensive clinical examination, often leading to diagnostic delay and causing a considerable burden to patients and their caregivers. No objective biomarkers for ALS have been established so far as indicators for early diagnosis. Speech has recently emerged in the literature as a promising biomarker for neurodegenerative diseases capable of being integrated into telemonitoring solutions. This dissertation focuses on speech degeneration in ALS patients and explores its po- tential to support the development of medical decision support systems for ALS diagnosis. We present a comprehensive study with several phonatory tasks and speech features to evaluate the generalisation potential of the proposed classification models. We use a public dataset with sustained vowels (N=64) and data with ALS and healthy volunteers being collected from ongoing research trials (N=22). We studied several supervised and unsupervised learning models with general-purpose features from temporal, statistical, and spectral domains calculated using the Time Series Feature Extraction Library (TSFEL) along with a dedicated feature set for speech analysis. Two approaches for supervised classification were considered: i) sample-based, where the signals were divided into fixed- length windows, and ii) patient-based, where a voting system was implemented based on the sample-based classification of each patient. We achieved a mean diagnostic performance with an F1-score over 80 % . The best scores for the sample and patient-based classifications being 96 % and 100 % for vow- els, 96 % and 95 % for sentences and 82 % and 87 % for cough. Our results support that speech-dedicated features improve the models’ performance when combined with general-purpose features. Our findings support the utility of speech as a promising digi- tal biomarker and pave the way for remote examination at patients’ residences, increasing the data available for clinicians for better diagnosis and prognosis of ALS. |
|---|---|
| Autores principais: | Cebola, Ricardo Alexandre Silvestre Matos |
| Assunto: | Amyotrophic Lateral Sclerosis Speech Automatic Speech Analysis Signal Processing Machine Learning Feature Selection |
| Ano: | 2022 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade Nova de Lisboa |
| Idioma: | inglês |
| Origem: | Repositório Institucional da UNL |
| Resumo: | Amyotrophic Lateral Sclerosis (ALS) is an incurable neurodegenerative disease affecting motor neurons. ALS diagnosis requires extensive clinical examination, often leading to diagnostic delay and causing a considerable burden to patients and their caregivers. No objective biomarkers for ALS have been established so far as indicators for early diagnosis. Speech has recently emerged in the literature as a promising biomarker for neurodegenerative diseases capable of being integrated into telemonitoring solutions. This dissertation focuses on speech degeneration in ALS patients and explores its po- tential to support the development of medical decision support systems for ALS diagnosis. We present a comprehensive study with several phonatory tasks and speech features to evaluate the generalisation potential of the proposed classification models. We use a public dataset with sustained vowels (N=64) and data with ALS and healthy volunteers being collected from ongoing research trials (N=22). We studied several supervised and unsupervised learning models with general-purpose features from temporal, statistical, and spectral domains calculated using the Time Series Feature Extraction Library (TSFEL) along with a dedicated feature set for speech analysis. Two approaches for supervised classification were considered: i) sample-based, where the signals were divided into fixed- length windows, and ii) patient-based, where a voting system was implemented based on the sample-based classification of each patient. We achieved a mean diagnostic performance with an F1-score over 80 % . The best scores for the sample and patient-based classifications being 96 % and 100 % for vow- els, 96 % and 95 % for sentences and 82 % and 87 % for cough. Our results support that speech-dedicated features improve the models’ performance when combined with general-purpose features. Our findings support the utility of speech as a promising digi- tal biomarker and pave the way for remote examination at patients’ residences, increasing the data available for clinicians for better diagnosis and prognosis of ALS. |
|---|