In this project we deal with the massive COVID-19 diagnosis problem. Since nations need to test the highest amount of people, the number of diagnostic kits is not enough to make massive diagnosis. This massive diagnosis could lead governments to apply the correct public policies to avoid community-acquire infection. To help with this problem we propose the use of x-ray images as a diagnosis method using Deep Neural Networks to help with the findings and to make accurate diagnosis. X-Ray is already a widespread image method used to detect many other diseases. It is known that it's hard to find artifacts on chest X-Ray, but Artificial Intelligence could help doctors with possible findings on the image. With the large amount of data created by the current diagnosis process, it is possible to create a real time computer-aided diagnosis system.
Our proposal was inspired in developing technologies to help doctors to test and differentiate patients with COVID-19 in order to prescribe an accurate diagnosis and an efficient treatment. We believe that accurate and early diagnosis could lead to a better prognosis, helping saving lives and improve patient’s life quality. For that, we used the most recent Artificial Intelligence technique to detect patterns and classify images. We used the World Health Organization website, in the link provided by Nasa SpaceApps, to find a dataset of images which could be used for our propose. We found an image dataset with over 262 x-ray images from patients with covid-19. In this dataset, we used part of it to train our Deep Neural Network and the rest to evaluate the method. To develop the code, we used Tensorflow 2.1, a Google Library for Deep Learning implementation, and OpenCV, a Computer Vision Library for image enhancement and processing. Our software is based on an online application and our website mockup was developed using React Native. The architecture of our Neural Network was based on Convolutional Neural Network with two convolutional layers, followed by an MaxPooling layer each. We’ve got an accuracy of 93% on classifying X-Ray images between sick and healthy patients.
In project we used an image dataset provided by World Health Organization, accessed by the first Resource Example Link of this challenge.