Classification of Amyotrophic Lateral Sclerosis Patients Using Speech Signals

EasyChair Preprint 10558

Abstract

The neurological condition known as Amyotrophic lateral sclerosis (ALS), which progresses and is irreversible, starts with early signs including speech and swallowing difficulties. The early acoustic presentation of speech and voice problems can be difficult to identify for both human experts and automated systems. In order to address this challenge we developed a non-invasive machine learning model for ALS diagnosis, this study proposes a voice assessment approach for an automatic system that can distinguish between healthy individuals and ALS patients. Specifically, our work focuses on analyzing continuous production of the vowel sounds /a/ and /i/using a feature extraction technique known as the wavelet time scattering transform which is not yet explored for ALS disease detection. In order to select the most relevant features from the extracted set of 84 features, we employed a correlation-based feature selection approach, using which we identified features with a correlation coefficient lower than 0.75. The selected features were then subjected to principal component analysis (PCA) to reduce the dimensionality of the dataset, resulting in a final set of 10 features.. The resulting PCA-based model achieved an accuracy of 84.2% using support vector machine (SVM) for classification, with a sensitivity and specificity of 77.8% and 90%, respectively.

Keyphrases: Amyotrophic lateral sclerosis(ALS), non-invasive, sustained vowel phonation, wavelet time scattering transform

@booklet{EasyChair:10558,
  author    = {Om Prakash Kurmi and Manasi Gyanchandani and Nilay Khare and Arvind Pillania},
  title     = {Classification of Amyotrophic Lateral Sclerosis Patients Using Speech Signals},
  howpublished = {EasyChair Preprint 10558},
  year      = {EasyChair, 2023}}