ABSTRACT
People suspected of having COVID-19 need to know quickly if they are infected, so they can receive appropriate treatment, self-isolate, and inform those with whom they have been in close contact. Currently, the formal diagnosis of COVID-19 requires a laboratory test (RT-PCR) on samples taken from the nose and throat. The RT-PCR test requires specialized equipment and takes at least 24 h to produce a result. Chest imaging has demonstrated its valuable role in the development of this lung disease. Fast and accurate diagnosis of COVID-19 is possible with the chest X-ray (CXR) and computed tomography (CT) scan images. Our manuscript aims to compare the performances of chest imaging techniques in the diagnosis of COVID-19 infection using different convolutional neural networks (CNN). To do so, we have tested Resnet-18, InceptionV3, and MobileNetV2, for CT scan and CXR images. We found that the ResNet-18 has the best overall precision and sensitivity of 98.5% and 98.6%, respectively, the InceptionV3 model has achieved the best overall specificity of 97.4%, and the MobileNetV2 has obtained a perfect sensitivity for COVID-19 cases. All these performances have occurred with CT scan images.