#CloseUp interview with VoiceMed
Pandemic crisis was and still is a serious challenge for everyone around the world. Nevertheless, it brought people together to create many amazing solutions helping to fight the crisis. One of these solutions is VoiceMed that was created during Hack the Crisis Lithuania online hackathon. VoiceMed is a technology created to identify COVID-19 related illness symptoms. It has a voice recognition software powered by artificial intelligence (AI), allowing to record your coughing and breathing, and giving instant results if a person might have COVID-19 virus.
We decided to learn more about VoiceMed and interviewed Jonė Pukėnaitė, MD who has been working with the solution.
How is VoiceMed solution helping to fight COVID-19 crisis?
The COVID 19 pandemic has become substantial repercussions on global economies which leads to an unprecedented burden on the healthcare ecosystem. Lack of adequate care triaged testing facilities and protective gear supply along with inequities within universal healthcare systems, increases the enormity of the pandemic especially in low- and middle-income countries. The need for difficult evidence-based decisions made has never become more crucial and the speed for front-line carers to make these decisions likely to have a negative effect. AI and machine learning in VoiceMed solution helps us to improve consistency and efficiency to manage the limited resources within existing clinical settings in prioritizing the patients with high risk for COVID19 in real-time. In an effort to manage both increased demand (e.g., access) by patients and patients (appropriately) nudged into clinic settings, telehealth and its component pieces provide needed access to healthcare, possible cost savings, and, in many cases, improved continuity of care/care management via patient monitoring and adherence to treatment protocols. This will ensure a smooth workflow for the healthcare ecosystem.
Why is VoiceMed important to the community, how will it help people?
Our mission is to provide people with the capacity to test COVID-19 through their phones from anywhere, anytime and get instant result.
VoiceMed will be used to identify high-risk patients with COVID-19 by implementing deep learning methods and classify both coronavirus-positive and negative sound streams. We are developing an AI technology intending to identify a healthy person from an infected with COVID-19 through the use of their voice.
How did you decide to create such a solution?
We have launched the initiative couple of months ago during the Hack the Crisis Lithuania online hackathon. Since then, we brought 40+ motivated people together with different backgrounds (tech, ML, Compliance, Marketing, MD etc.) from 10 different countries.
We were searching for a solution to at least partly solve the crisis. Our researchers were coming across the articles that were shaping our idea and giving a glimpse of how possible it actually was.
Most of the literature on clinical characteristics of COVID-19 patients have reported the involvement of respiratory pathophysiology (with cough, sore throat and breathing difficulty) as the main hallmark. Acoustic analysis has the potential to expedite the detection and diagnosis of voice disorders.
From the various literature on clinical symptoms of COVID-19 that has been published, cough and shortness of breath are among the main clinical characteristics, indicating involvement of respiratory pathophysiology. The pathophysiological changes caused by different respiratory conditions modulate the sound quality. Associated with breathing difficulty and cough, these sound streams can be used to detect COVID-19.
After collecting all the research documents and making sure our idea can really become a solution to fight COVID-19, we dug into work making this solution our priority.
How this solution could facilitate the work of the public sector?
We expect that the public sector could use our designed solution for pre-diagnostics of COVID-19. There will be a possibility to integrate our product into the phone lines and websites of the hospitals, therefore patients who have COVID-19 symptoms could be checked out without going to the facility and get instant results.
Some more solutions could be created as well, as needing to cough into the microphone before entering the hospital, so doctors and patients would go through the door to the facility only if the answer of the test is negative.
Any other interesting facts/elements about the creation of VoiceMed?
Thus far, as a first approach, we have managed to implement a cough classification model based on cough sounds samples that were previously labelled wet/dry by one of our doctors in the team. This data was then pre-processed and used for classification model. The original model used for classification is a logistic regressor with stochastic gradient descent (SGD) algorithm and convolutional neural networks (CNN), for classification. Prior to training the model, the training set of observations was normalized by fitting a scaler function. The same fit scaler function (to centre the data) was applied to the validation set prior to using it for predictions. The final average accuracy obtained was 84% with the method described.