الفهرس 
chapter 1: IntroductionOnly 14 pages are availabe for public view
 |  | from | 106 |  |  |
| | | from | 106 | | |
Abstract
The aim of this thesis is to introduce the required methods for building an Artificial Intelligence-based tool for detecting and diagnosing COVID-19 depending on the external symptoms only of the disease without the need for any clinical or chemical tests.
The main motivation of this study was the limitations of the traditional COVID-19 tests such as the PCR test which is expensive and requires about 2-48 hours for results to be available. Also, it was impossible to apply chemical tests on a large scale in crowded places. So it was important to find an alternative fast, and accurate diagnostic tool.
The proposed tool works on monitoring and analyzing the external symptoms of the disease such as sounds (cough, breath, and speech) and other clinical symptoms that could be felt or described by patients without any clinical test. This study is based on recent researches that confirm the existence of COVID-19 signature in the cough and respiratory sounds of patients. The virus effect might not be detectable by the human ear but well-trained machine-learning models could detect it accurately.
For building the detector, several machine learning models were trained by different types of COVID-19 sounds such as cough, breaths, and human speech besides other clinical symptoms. And by testing those models, It was found that models trained with few samples collected early in 2021 were able to detect the virus in samples collected after several months of the training process regardless of the mutations and changes that occurred to the virus. Also, those models could achieve an accuracy of 91% on testing 313 samples that contain 236 positive cases.
As those promising results indicate how efficient and reliable are the proposed models, The mentioned models could be embedded in an IOT device, which will help unlimited numbers of users to make COVID-19 tests instantly at any time or any place.