Using one’s voice involves the coordination of a variety of anatomical systems. When the lungs push air through the vocal cords, these vocal cords vibrate and produce sounds, which the tongue, lips, and nasal cavities shape into words. The brain controls the processes and regulates the words we use to communicate. Someone's speech can provide indicators of a disease that affects one of these systems. Using voice AI can identify these indicators and distinguish comorbidity. Comorbidity can be a barrier to forming an accurate diagnosis. Comorbidity is the presence of two or more medical conditions in a patient. Deconstructing diagnoses with biomarkers will help predict and improve responses to treatments. This technology is also noninvasive, inexpensive, and effective at detecting cardiovascular, neurodegenerative, respiratory, and mental health disorders.
Parkinson's, one such neurodegenerative disease, has no biomarkers, which makes it difficult to detect and diagnose. However, Max Little, an applied mathematician, created an AI database that processes audio recordings of people speaking and finds specific features in their speech patterns to identify and diagnose people with Parkinson’s. Little’s goals are to use this voice-based screening to monitor a patient’s state frequently, especially when check-ups for Parkinson's patients happen every 3-6 months. His other goals include using the database to find large numbers of people to be recruited for treatment trials and completing a large-scale screening program to detect early signs of Parkinson's before it becomes too late to start treatment. Early intervention and treatment are crucial as they slow down disease progression and reduce symptoms.
Winterlight Labs, a startup that detects and monitors speech patterns to detect signs of Alzheimer's, conducted a similar initiative. Using machine learning and AI, Winterlight’s system identifies biomarkers and early indicators of Alzheimer's through patterns in speech from audio recordings. Winterlight Labs conducted a study with 250 participants where the participants' linguistic and acoustic features of their speech were analyzed. Winterlight's AI system identified 35 vocal features that distinguished the patients with Alzheimer's from the control patients with 92% accuracy. The start-up's voice technology is now being used in drug trials for pharmaceutical companies such as Johnson & Johnson, Cortexyme, and Alector.
Another company, Sonde Health, has developed software that uses voice samples to identify mental and physical medical conditions. Sonde identifies acoustic features in voices that could find out the chances of someone having respiratory symptoms such as coughing, shortness of breath, chest tightness. These are common symptoms that occur in asthma, but they can also be used to detect COVID-19. Sonde can also detect vocal biomarkers for mental health problems. Since the early 20th century, voice changes were considered important biomarkers of depression. Using voice biomarkers to diagnose mental health and psychiatric disorders can promote early diagnosis before they seek care and monitor symptoms after seeking care. If individuals cannot seek mental health treatment, then they could use the voice software themselves to track their progress.
The potential for and capability of diagnostic voice AI makes it an emerging technology that needs to be implemented in hospitals to improve healthcare for all.
Cover Images: "Designed by Freepik" - https://www.freepik.com/
Comments