Using Speech Biomarkers to Identify and Track Neuromuscular and Neurodegenerative Disorders


Clinical Trials

Poster Number: M195


Ashkan Vaziri, PhD, Biosensics LLC, Adonay Sastre Nunes, PhD, BioSensics LLC, Newton, MA, USA, Ram kinker Mishra, PhD, BioSensics LLC, Newton, MA, USA, Andrew Geronimo, MD, Department of Neurology, Penn State College of Medicine, PA, USA, Zachary Simmons, MD, Department of Neurology, Penn State College of Medicine, PA, USA, Amanda Guidon, MD, MPH, Massachusetts General Hospital/Harvard, Anne-Marie Wills, MD, MPH, Department of Neurology, Massachusetts General Hospital, Harvard Medical School, Boston, MA, USA, Alexander Pantelyat, MD, Johns Hopkins University School of Medicine, Department of Neurology, Baltimore, MD, 21287, USA, Jamie Lynn Adams, MD, Department of Neurology, University of Rochester Medical Center, Rochester, NY, USA

Speech impairment is a prevalent symptom in neuromuscular and neurodegenerative disorders, however the mechanism and extent of speech impairment varies significantly in different disorders, and even within patients with the same disease. Thus, there is a need for scalable and cost-effective solutions to assess speech impairment as a potential tool for diagnosing and tracking neurological and neuromuscular disorders.
To assess and compare speech features across multiple neuromuscular and neurodegenerative disorders using BioDigit Speech.
Passage reading speech assessments were performed in four different studies using BioDigit Speech: 1) amyotrophic lateral sclerosis (ALS) (n=11), 2) progressive supranuclear palsy (PSP) (n=11), Parkinson’s disease (PD) (n=10), 3) Huntington’s Disease (HD) (n=41), and 4) Myasthenia gravis (n=20, screening). Group differences and correlations with clinical scores were explored. A machine learning classifier was trained to automatically differentiate different patient populations based on their speech features.
Machine learning classification models achieved a weighted accuracy of 90% in identifying the disease from simple speech tasks with a sensitivity of 79% for PSP, 90% for PD, 100% for ALS, 86% for HD and 95% for controls. In ALS, bulbar dysfunction as measured by the ALSFRS-R was associated with reduced articulatory rate and intelligibility. Similar observations were made in PSP and HD. In addition, multiple speech measures correlated with the MoCA, including similarity and intelligibility in PD, HD and PSP. Machine learning models demonstrated strong capabilities in predicting clinical diagnoses and outcomes with high accuracy and sensitivity.

Our findings highlight the potential of BioDigit Speech as a valuable tool to aid in identifying and tracking multiple neuromuscular and neurodegenerative disorders.