Advancing the Diagnosis, Prognostic Assessment, and Speech Communication for ALS: A Machine Learning Approach

Abstract: Amyotrophic lateral sclerosis (ALS), also known as Lou Gehrig’s disease, is a fatal, neurodegenerative disease that is characterized by the rapid progression of muscle weakness. The patients’ quality of life is severely impacted due to the loss of motor functions, including speech. Currently, there is no definite procedure for the diagnosis of ALS, and objective measures are highly needed to assess the disease progression. This research aims to (1) advance the diagnosis of ALS using speech acoustic samples, (2) automatically predict the bulbar score of ALS functional rating scaling - revised (ALSFRS-R), the clinical assessment for disease progression, using speech acoustic and kinematic samples, and (3) decode imagined speech from non-invasive neural (Magnetencephalography or MEG) signals with potential to resume a level of communication for patients in late-stage ALS (fully paralyzed). Machine learning techniques were used to analyze speech acoustic, kinematic, and neural signals collected from the patients with ALS. Although the current focus is on ALS, this research is relevant to a broader range of neurological speech impairments, including those related to stroke, traumatic brain injury, or Parkinson’s disease.