Skip to main content

Is AI the Future for COVID-19 Screening?

April 19, 2021

By Shady Hassan, MD, Co-Founder, Chief Medical Officer and Chief Operating Officer of Vocalis Health

Shady Hassan, MD, Co-Founder, Chief Medical Officer and Chief Operating Officer of Vocalis HealthFor life to return to anything close to ‘normal’ post-COVID, people need to be confident they can go back to their work environment and other public areas like buses, stores, cinemas and airports.  There are currently two strategies to build that confidence—screening and vaccinations. Both have limitations. 

The current approach to screening relies on PCR or antigen tests—to detect if the person is currently infected—or antibody (serology) testing to detect if the person previously had the virus. These tests are costly, time consuming and dependent on competent testers and adequate supplies of test kits. Because of the cost, they have to be used sparingly—normally on people who have some symptoms or have been exposed to someone who has them.

Alternatively, standalone screening solutions, including temperature and symptom checks, are insufficient. They cannot detect people who are asymptomatic, and in the case of temperature, can easily be masked by a person taking an OTC medication like Tylenol. Survey-based symptom checks also have limitations as they rely on the honesty of the person answering the questions. Current symptom-based screening options only detect a fraction of cases–and are completely ineffective at detecting asymptomatic patients. With the high contagion levels and mortality associated with COVID-19, erring on the side of caution has been critical, so the current suite of testing options is probably keeping more people isolated at home than is needed.

The vaccination program is also not a short-term solution. Assuming that Dr Anthony Fauci, Chief Medical Advisor to the US President, is correct, it will only be possible to achieve ‘herd immunity’ with 70-85% of the US population vaccinated, by the end of 2021.

We have to accept that the current approach to screening is not sustainable or effective. For people to get back to a near-normal life, gathering in larger numbers, we will need a system of screening that is non-invasive–and provides an effective first-line of defense. A simple, highly scalable and low-cost test that can assess the risk before people go to the office, gym or a restaurant is needed.

Voice-based screening to detect COVID-19-sufferers

One screening tool which meets all the necessary requirements of affordability, availability, and accuracy is voice-based screening. Since COVID impacts the respiratory system, it is natural that the voice will be impacted and reveal tell-tale signs, and it has demonstrated the ability to identify asymptomatic patients as well.

In February 2021, the Municipal Corporation of Greater Mumbai (MCGM) shared initial results of a clinical study conducted at their NESCO COVID-19 Center to validate voice-based screening. The study included over 2,000 participants who spoke numerous languages including English, Hindi, Marathi and Gujarati. The test used vocal biomarkers–think of fingerprints for the voice–that achieved 80+ percent accuracy in assessing COVID-19 infection.

Previously vocal analysis has proved to be effective in detecting and monitoring patients with chronic obstructive pulmonary disease (COPD) and other voice-affecting diseases. COVID patients also have distinctive voice characteristics that can be recorded on their own smartphone–without the risk of contagion–and can be presented as spectrograms or ‘voiceprints’. These are complex color plotted digital ‘maps’ of the way they speak, identifying hundreds of biomarkers or features in each spectrogram. These are produced in real-time like the undulations on a sound-system. Using AI-powered software and machine learning the spectrograms of confirmed COVID patients are then used as a benchmark to compare to the biomarkers of people who are being screened.

A voice-based screening process would simply require a patient to record their voice on an app and recite a simple sequence of words such as counting from 50 to 70. The software correlates the spectrogram against the database of thousands of confirmed COVID patients. If there is a low correlation, the person can confidently go about their daily activities; if there is a high correlation, the person is advised to stay home or get tested.

As voice analysis can be conducted at home or on the move via apps with virtually no resource implications, they can be used at scale on a regular basis for ongoing monitoring. So, they could be used by office workers, outpatients or university students at the start of their day or even at points during the day.

AI-guided vocal analysis is hard to fool and has even been used effectively with people who are asymptomatic.  Voice analysis studies have been conducted on people with no outward symptoms in a range of facilities including, hospitals, outpatient clinics and on volunteers, and they were able to detect COVID in 70-80% of people who were carrying the virus without any reported symptoms. Once screened, institutions then have data to guide and prioritize PCR testing, or simply recommend that a person remains at home until they are screened again and can prove they have recovered.

The current approach to screening is expensive and resource intensive. Vaccination may be a mid-to-long-term solution, but on its own will not reassure people that it is safe to return to large offices, auditoriums or feel safe in public sites. Voice analysis based on apps that are driven by AI and machine learning can screen for COVID in a non-invasive way and provide a highly effective first line of defense to instill confidence that people may safely return to the public sphere.

Shady Hassan, MD, is an experienced internal medicine physician focused on innovation and disruption of healthcare. He is a co-founder of Vocalis Health, an AI healthtech company pioneering the development of vocal biomarkers to screen, detect, monitor and predict health symptoms, conditions and diseases. Shady received his MD from the Technion, completed a residency in internal medicine and worked as an attending physician in the internal medicine and emergency departments.

Agree or disagree with an article? Share your professional thoughts on an article you read.

Your Name
7 + 3 =
Solve this simple math problem and enter the result. E.g. for 1+3, enter 4.
Back to Top