Abstract:
Since the outbreak of the global COVID-19 pandemic in Wuhan, China, in 2019, its impact has been seen worldwide. Early
detection of COVID-19 is very important, as it keeps the infected people isolated from other people, thus minimizing the risk of
further transmission. The standard diagnostic approach is based on RT-PCR. However, due to the scarcity of PCR kits in some regions
and the costs associated with this technique, there is a growing demand for alternative solutions. Recently, diagnosis of COVID-19
by medical imaging has been recognized as a valid clinical practice. Meanwhile, the massive increase in COVID-19 cases has put
considerable pressure on radiologists responsible for interpreting these scans. This paper introduces an automated detection approach
as a rapid alternative for COVID-19 diagnosis. We present a deep CNN model to differentiate between normal and pneumonia cases,
as well as patients with COVID-19. Our approach is based on EfficientNet-B7 architecture and improved with Squeeze and Excitation
block as an attention mechanism. In addition, we propose an innovative architecture that combines CNN with SVM to achieve the best
performance. Experimental results show that the proposed framework provides better performance than existing SOTA methods, with
an average accuracy of 97.50%, while the precision and recall of COVID-19 are both 100% without any pre- or post-processing.
