A team from the Massachusetts Institute of Technology (MIT) hypothesized that people with coronavirus, including asymptomatic cases, could be accurately detected by a recording on cell phones using artificial intelligence (AI). The findings were published in the IEEE Open Journal of Engineering in Medicine and Biology. Out of the audio samples from 5,320 subjects, 2,500 of the recordings were from people who tested positive for the virus between April and May.

Their MIT Open Voice model was trained through a collection of around 70,000 recordings. The neural network called ResNet50 could distinguish various sounds such as "mmmm" which determine vocal strength. The neural network was also trained to tell the difference between the words them, the, and then by screening more than 1,000 hours of speech.

Another neural network was trained to differentiate speech associated with emotional states. For example, people diagnosed with Alzheimer's disease have certain negative expressions such as frustration more than happy and calm expressions.

Combining AI Neural Networks

The third neural network was trained with cough audio samples to distinguish lung and respiratory states. The combination of vocal cord strength, expressions of sentiment, lung/respiratory performance, was overlaid with an algorithm to detect muscular degradation to differentiate a healthy from an unhealthy cough.

Brian Subirana said that talking and coughing affect the vocal cords and surrounding organs such as the lungs. Aside from detecting the cough, gender, mother tongue, and emotional state of people, the team analyzed if Alzheimer's biomarkers were relevant for coronavirus.

The AI speech processing framework detected positive cases with 98.5% sensitivity and 94.2% specificity. The program also successfully detected asymptomatic cases with a sensitivity of 100% and a specificity of 83.2%.

Read Also: Researchers Turn to AI to Protect Orchids and Other Threatened Species

Developing a Diagnostic App

Subirana said that developing the diagnostic tool, like creating a mobile app, could diminish the spread of the virus if it was used before going to public places such as campuses and restaurants or help contain a sudden outbreak. The AI algorithm could detect the subtle differences between a healthy cough those with severe or asymptomatic cases, shared the team.

The team shared that the AI method "can produce a free, non-invasive, real-time, any-time, instantly distributable, large-scale COVID-19 asymptomatic screening tool" to contain the spread of the virus. However, the team would have to wait for the approval of the Food and Drug Administration for the technology to be integrated into a mobile app.

The team clarifies that the AI model cannot diagnose people with lung issues such as coronavirus, flu, or asthma. Rather, AI technology can discern the difference between an asymptomatic and a healthy cough.

Currently, the team has partnered with a company to develop an app as well as various hospitals to collect more cough audio samples to improve their initial findings. "Pandemics could be a thing of the past if pre-screening tools are always on in the background and constantly improved," the researchers wrote.

Read Also: Disney's Robot Can Perform Realistic Eye Movements & Social Interaction 

Check out more news and information on Artificial Intelligence on Science Times.