In this research, we introduce our methods for creating a COVID-19 diagnostic tool based on the artificial intelligence to analyze the data of COVID-19 external symptoms such as cough, respiratory sounds, and other clinical symptoms such as fever, muscle pain, cold, sore throat, asthma, etc. to detect the virus without the need of any chemical or clinical tests which are expensive, slow and not available everywhere. Our diagnostic tool can be used publicly in crowded places such as shops, schools, or any human gatherings to detect the patients in their early stages, reduce the virus spread and forward the suspected people to clinical examination.
For creating our tool, we used deep learning-based models to analyze and learn from the collected sounds and the clinical features of confirmed COVID-19 cases and other normal cases. By using those models as classifiers they could distinguish the positive cases from the negatives. And we found that using simple binary classifiers trained with small samples of COVID-19 data collected early in 2021 can be trustworthy to detect COVID-19 in the recently collected samples regardless of the changes that occurred to the virus. And by testing the samples collected from 313 cases after several months of training our models, we could achieve an average accuracy of 91% to prove the proficiency of our tool in diagnosing COVID-19 and detecting the virus in the long term after several mutations.