AI Talk: Cough screening, monitoring behavior and mobility data

Nov. 13, 2020 / By V. “Juggy” Jagannathan, PhD

This week’s blog focuses on COVID-19, highlighting stories that shed light on how AI is helping folks deal with the pandemic.

COVID-19 Cough Sensor

A team at MIT has developed an AI model that can detect COVID-19 from a forced cough. This model of detection could be a potential game changer as it is basically free. You cough into your smartphone and presto: It can tell you if you have COVID-19, even if you are asymptomatic. So far, the model has recognized 100 percent of asymptomatic cases and showed a rate of 98.5 percent sensitivity in correctly identifying COVID-19 cases and a rate of 94.2 percent specificity in correctly identifying people without COVID-19. This is quite incredible.

How did the MIT researchers decide to pursue this model? Well, they were already researching how to detect Alzheimer’s using an Open Voice program, which has been collecting voice and various metadata from people. They were able to detect gender and mother tongue from three-second cough samples! Who knew our coughs were that specialized? The deep learning approach they used in their research of early Alzheimer’s detection involved identifying various biomarkers. They adopted the same strategy for COVID-19 detection. Specifically, they trained four separate models to detect biomarkers related to muscular degradation, changes in vocal cords, changes in sentiment/mood and changes in the lungs and respiratory tract. They used this data to create a composite model that predicts the presence or absence of COVID-19.  

So, what are the next steps? It appears the researchers have partnered with an undisclosed Fortune 100 company to further demonstrate the value of this tool. More power to them! The sooner this kind of tool proves its efficacy, the better its usefulness in the current pandemic.

Face mask usage and social distancing

I saw this article in AI in Healthcare about a Canadian health system’s efforts to keep its patients and staff safe. The health system did this by monitoring video feeds to determine if patients and staff were wearing facial masks and social distancing in highly-trafficked areas of their facilities. The pilot prototype was developed through a joint collaboration between Microsoft and Providence Health Care of Canada. The Canadian health system has utilized various AI toolkits and visualization solutions from Microsoft to build a live dashboard of their most used areas like emergency department and radiology waiting rooms.

The health system was quick to point out that the cameras are not collecting any personally identifiable information. Admittedly, if you cover half of someone’s face it is difficult to recognize them, though not impossible. It is not clear what actions are taken if an area is overcrowded or someone is not wearing a mask. Perhaps they make an announcement over a loudspeaker? Nevertheless, this is an interesting use of advanced AI technology. In addition to monitoring COVID-19 policy adherence, they also plan to monitor hand hygiene practices. That technology, once developed, will probably become a permanent fixture!

Opening restaurants and social inequities

A recent blog post in MIT Technology Review highlights research conducted by a collaborative research team from Stanford and Northwestern universities. The preview of the research paper has been published in Nature. The research? Figuring out how to use smartphone mobility data (from a company called SafeGraph and Google) to predict where people are contracting COVID-19. The verdict? Restaurants are one of the main locations where the infection is spreading. Places in economically disadvantaged areas are also a main contributor to the overall infection rate.

How did they manage to arrive at these conclusions? They use mobility data as their source data as it provides anonymized location data from the smartphones of 98 million people in the 10 biggest cities. With this data they were able to plot how many people spent what amount of time at various points of interest. Then they predicted how many people were likely to get infected and matched those predictions to real data on known infection rates in these cities.

The interesting work reported here, using their validated model for prediction, is to determine the impact of various policy stances that one may take. For instance, the data shows if you reduce the maximum occupancy of a minority of establishments to say 20 percent, the infection rate is reduced by 80 percent in that area. Establishing food distribution centers in economically disadvantaged areas will reduce the infection rate in such areas as well. Essentially, this study is providing a blueprint on what knobs to tweak to keep the economy open while reducing the infection rate.

Acknowledgement

My student Noble Nkwocha sent me the COVID-19 cough detector story.

I am always looking for feedback and if you would like me to cover a story, please let me know. “See something, say something!” Leave me a comment below or ask a question on my blogger profile page.

V. “Juggy” Jagannathan, PhD, is Director of Research for 3M M*Modal and is an AI Evangelist with four decades of experience in AI and Computer Science research.