The Coronavirus pandemic (Covid-19) is wreaking havoc all over the world and at the same time it has focused everyone to work together to be able to overcome this crisis. And it is that people from all areas have begun to contribute and in the case of programmers and the area of technology, they are also focusing all their efforts to contribute with that small grain of sand.
And it is that with the number of people affected and deceased by the new coronavirus continues to increase around the world. At the moment, the fight against this disease also involves the use of artificial intelligence.
In fact, The personnel within this sector are managing to develop various tools that allow, in particular, to optimize the monitoring of patients or the disinfection of contaminated areas, pharmaceutical research or detection.
On this last point, a team of researchers from the universities of Oklahoma (U.S), Kharkiv (Ukraine) and Michigan (U.S) they were able to create a mobile app that supposedly helps with the diagnosis of coronavirus thanks to the analysis of the user's cough sound.
It is important to mention that this app is in internal testing and a trial version has not been released to the public, so they merely released a report on what it has accomplished thus far.
The inability to perform tests at scale has become the Achilles heel in humanity's ongoing war against the COVID-19 pandemic.
An agile, scalable and cost-effective test, deployable on a global scale, can act as a game changer in this war. To address this challenge, building on the promising results of our previous work on cough-based diagnosis of a variety of respiratory diseases, we developed an Artificial Intelligence (AI) -based test for the preliminary diagnosis of COVID-19.
The test can be implemented at scale through a mobile application called AI4COVID-19. The AI4COVID-19 application requires 2-second cough recordings of the subject. By analyzing cough samples through an AI engine running in the cloud, the app returns a preliminary diagnosis in one minute.
Unfortunately, coughing is a common symptom of more than two dozen medical conditions not related to COVID-19. This makes the diagnosis of COVID-19 from coughing alone an extremely challenging problem.
We solve this problem by developing a novel multi-mediator-focused risk-averse AI architecture that minimizes misdiagnosis. At the time of writing, our AI engine can distinguish between the cough of COVID-19 patients and various types of cough without COVID-19 with more than 90% accuracy. AI4COVID-19 performance is likely to improve as more and better data becomes available.
If the app detects the sound of coughing, it is transmitted to three rating systems parallel and independent: the multi-class classifier based on deep learning (DL-MC), the machine learning based multiple class classifier classical (CML-MC) and the deep learning-based binary classifier (DL-BC).
The researchers integrated a cough detector into their AI engine, which is capable of distinguishing sound from 50 other common ambient noises.. To train and test the diagnostic system, they collected various logs: 48 COVID-19, 102 bronchitis, 131 whooping cough and 76 normal cough attacks.
The results of the three classifiers are transmitted to a mediator. If the classifiers disagree, the app shows that the test is inconclusive. Otherwise, the three classifiers will use different approaches that will be cross-validated, leading to two results (positive or negative).
Its developers mention that AI4COVID-19 is not designed to compete with the tests clinics that are being carried out in almost all countries.
Instead, they mention that their intention is to offer a complementary tele-test tool Deployable anytime, anywhere, by anyone, so clinical testing and treatment can be channeled to those who need it most, saving more lives.