CNN for the detection of COVID-19 from chest X-ray images

COVID-19 can be detected by molecular tests or medical imaging. Doctors use medical images such as chest X-rays of patients to detect COVID-19, however, it is not so easy to detect the disease just by looking at an X-ray, in this work a system based on Convolutional Neural Networks (CNN) capable of classifying X-ray images of healthy patients, pneumonia patients and COVID-19 patients. The images were taken from the Kaggle open access datasets, 6279 images were taken in total, 4175 for training, 949 for validation and 1155 for testing. All the images were transformed at a resolution of 224x224 pixels and each pixel value was normalized, then the images were equalized by histogram. In the training stage, a CNN with MobileNet architecture was used, which has 50176 inputs (pixels), a hidden layer of 1000 neurons and an output layer of three neurons corresponding to the classes healthy lung, lung with pneumonia and lung with pneumonia by coronavirus. The Grad-CAM tool was used for visualization. For the test set, a batch size of 1 was used, this for the model to evaluate and classify each image, in this set a accuracy of 99.48 % was obtained on average for all categories. Where the objective is to contribute with an auxiliary tool for the detection of COVID-19 based on CNN. Thus, the proposed system is capable of differentiating and classifying chest X-ray images of healthy patients, with pneumonia and with COVID-19, so the system is capable of detecting COVID-19 with good precision.