Using LSTM-Based Model on Vocal Signal Analysis for the Classification of Multiple Sclerosis

Multiple Sclerosis (MS) is a chronic, neurodegenerative, and autoimmune disease of the central nervous system (CNS) that affects the brain and spinal cord. It impacts motor, sensory, cerebellar, cognitive, and language functions, with more than 2.5 million people affected worldwide. In such a scenario, early detection is crucial for effective management and treatment of MS. Quantitative analysis of MS-related speech disorders could significantly aid physicians in diagnosis and patient monitoring. Neurological voice disorders, including those seen in MS, arise from disruptions in the nervous system’s interaction with the larynx. This study investigates the use of classification techniques on speech signal analysis for MS identification by considering acoustic and mel-frequency cepstral coefficient (MFCC) features to differentiate MS patients from healthy controls within a dedicated dataset. The performance of Long Short-Term Memory (LSTM), Bidirectional LSTM (BiLSTM), and Support Vector Machine (SVM) models was evaluated across different classification experiments, including those performed on augmented datasets, to explore pathological speech characteristics. These findings provide preliminary information that may support physicians in characterizing the vocal features of MS patients.