Abstract:
"
COVID-19 is a highly contagious infectious disease that was declared a pandemic on
March 11th, 2020. It has caused both healthcare and financial crisis all around the
world by being spread out in 220 counties. COVID-19 is caused by SARS-CoV-2
whose nucleic acids are detected by the standard diagnostic method of detecting
COVD-19. Early detection of the disease diagnosis is essential to control the
transmission of the disease from an infected person. Medical imaging modalities have
been commonly utilized to detect various lung pathologies effectively. Pure ground glass opacities and consolidation opacities are the characteristic lung lesions in a
COVID-19 patient. Due to the limitations of the standard diagnostic method (RT PCR) of detecting COVID-19, an alternative diagnostic method (chest CT imaging)
has been proposed by various existing works. Furthermore, it is found that CT (98%)
imaging has higher sensitivity compared to RT-PCR (71%). On the other hand, deep
learning methods including Convolutional Neural Networks (CNNs) are extensively
applied in medical imaging to detect pathologies. In this project, the deep learning
approach of CNN has been applied for detecting COVID-19 from chest CT images of
the patients. The approach has been tested using the largest existing CT imaging
database of COVID-19. Experimental results depict that the proposed model can
achieve 93.89% accuracy of detecting COVID-19 through CT images with an F1
score, precision, and recall over 0.93, 0.93, and 0.92 respectively. Moreover, the false
positive and false negative results are very low having a very high level of prediction
value. The experimental observations suggest that this deep CNN-based approach
would have a very high potential of being applied to detect COVID-19 faster with
higher accuracy. The observations and the evaluations further suggest that the system
has the potential to be improved to detect the severity of COVID-19 and various other
lung diseases during the same attempt."
