A practical Deep Learning approach to assist COVID-19 detection based on Chest X-ray images

Given the large number of COVID-19 cases around the world, a practical solution to decrease and relieve the queue of patients in the hospitals and in the health care systems is welcome. Fast and reliable diagnosis based on technological tools can support medical professionals to manage this bottleneck situation, such as the diagnostic based on image techniques, which allows non-intrusive procedures. In this paper, we propose a practical methodology using deep learning to detect and classify lungs affected by COVID-19 using Chest X-ray radiography. RetinaNet architecture is considered here. This architecture is an one-stage object detection using focal loss often applied with dense, small and imbalance objects. We consider a dataset with 2500 images for model training and 1000 images to validate the model. Besides, a set of 1000 images from two different datasets are applied to test the pipeline approach. The obtained results show a specificity score of 0.54, precision of 0.68, recall of 0.994, and mAP of 0.913. The high recall score explains that a patient with COVID-19 will be classified correctly.