Adding Voice To Respiratory And Cardiovascular Disease Diagnosis: Interview With Prof. Elad Maor

The field of telehealth is growing thanks to the steady growth in supportive technology and the need for remote monitoring, assessment, diagnosis, and testing. Voice is unique to every individual due to people’s anatomical differences, which makes for a powerful tool when working remotely.

Vocalis Health is a company that uses machine learning and artificial intelligence to identify biomarkers in voice recordings. The company’s technology has been successfully used to identify mortality and hospitalization among heart failure patients. Vocalis is now adapting their technology to understand the association between voice biomarkers and COVID-19 infected individuals.

We recently had the opportunity to interview Prof. Elad Maor, an interventional cardiologist at Sheba Medical Center, Israel and an Associate Professor at Tel Aviv University. He is also a consultant at Vocalis Health and the lead in a recent study using their voice analysis technology.

Rukmani Sridharan, Medgadget: How can voice be used to identify disease? What signatures can you identify from voice recordings?

Prof. Elad Maor, Vocalis Health: Voice analysis is sophisticated and complex. It goes far beyond what we can “hear.” Therefore, it can be used to identify disease states with the help of machine learning and artificial intelligence tools.

The signatures are actually mathematical functions that are used to analyze voice data, and we prefer to use the term “voice features.” We used more than 200 voice features that were extracted from 20 second of speech. Based on these features we developed an algorithm that successfully identified high risk patients. Those patients were high risk for hospital re-admission and death.

Medgadget: Can you describe your recently published study in the Journal of the American Heart Association and its main findings?

Prof. Elad Maor, Vocalis Health: In the study we used voice recordings of patients with chronic disease states. These patients were registered to a call center of a large HMO in Israel (Maccabi Health Services). By combining electronic health care data with their voice recordings, we developed a unique database.

The database was used to identify patients at-risk for re-admission or mortality using machine learning and artificial intelligence tools. We then validated our tool on a specific sub population of the cohort – patients with congestive heart failure. Our main finding was that voice is independently associated with re-admission and mortality, and that it can assist healthcare providers with better identifying high risk patients.



Medgadget: Can you comment on non-cardiovascular diseases that can also be diagnosed with voice analysis?

Prof. Elad Maor, Vocalis Health: There is data to suggest an association between voice and neurologic disorders such as Parkinson’s disease. We have shown data to suggest a correlation with obstructive coronary artery disease as well as with pulmonary hypertension.

Vocalis Health also has data suggesting a role for voice in the diagnosis of COPD and shortness of breath in general. There are ongoing clinical studies at Mayo Clinic, as well as at Sheba Medical Center and in other places around the world looking for additional non-cardiovascular diseases.

Medgadget: How can your system be adapted to analyze the health of COVID patients?



Prof. Elad Maor, Vocalis Health: This is being looked into at the moment. We are recruiting both COVID-19 positive and negative patients at Sheba Medical Center. Their voice will undergo similar analysis with artificial intelligence and machine learning tools in order to try and identify patterns associated with COVID-19. We also have an online platform where we are asking the general public to donate their voices to help us fine-tune our analysis tools.

Medgadget: What are the challenges and confounding factors in using voice analysis for COVID diagnostics?

Prof. Elad Maor, Vocalis Health: Challenges include non-specificity of the lung injury (which can also occur with other viruses.) Confounding factors include noise and other disease states associated with lung injury (COPD or smoking for example.)

Medgadget: What is the future of voice analysis and telehealth in healthcare?

Prof. Elad Maor, Vocalis Health: Voice is transparent and available in every patient interaction. If it proves to be accurate, I believe the future is bright for voice as an additional complimentary clinical tool.

Source