Artificial intelligence taught to detect COVID-19 in asymptomatic carriers with 97% accuracy

One of the main problems of the coronavirus pandemic is the identification of carriers of the infection who have no symptoms or are mild. Researchers at the Massachusetts Institute of Technology (MIT) have developed a diagnostic method using artificial intelligence that identifies carriers of the virus by coughing with an accuracy of 97%.

This technique builds on the results of a previous study to identify Alzheimer's disease based on analysis of cough and speech. As it turned out, similar changes occur in patients with coronavirus, since the nasopharynx and vocal cords are primarily affected.

The updated technique uses the ResNet50 neural network, which has been trained on thousands of hours of recordings of human speech, on a set of words pronounced in various emotional states, as well as cough variations entered in the database, which helps to detect pathological changes in the work of the respiratory system.

When combining all three models, the scientists used a noise layer overlay to separate a strong cough from a weak one. After analyzing 2, 500 cough records of confirmed COVID-infected people, the AI ​​identified 97.1% of patients out of 100% of asymptomatic cases.

The results of this study were published in the IEEE Open Journal of Engineering in Medicine and Biology.