Wolfram and other image classifier deep neural networks (DNNs) were adapted to differentiate COVID-19 from normal and other viral pneumonia chest x-rays, giving up to 100% accuracy on 3x 1,200 training images. Another DNN gave an initial COVID-19 clinical staging accuracy of 80% on just 80 images (work in progress). Methods to improve the training data by employing routine and AI-assisted quality assurance are discussed and assessed.