Abstract:
A chest radiograph is a crucial tool to diagnose numerous diseases. With an increase in workload
of radiologists, the time to treat patients has excessively increased. With-the-development-of AI
and deep neural networks, it has become possible to assist different professionals with their
tasks. Deep neural networks provide highly accurate results for the classification of extremely
heterogeneous images if the dataset is large enough. This report discusses the process of training
a Convolutional-Neural-Network (CNN) on the dataset containing thousands of radiographs and
then testing it on a different dataset. CheXpert, a large chest radiograph dataset, from Stanford
Machine Learning Group, was acquired and used for training the model. Results including
accuracy, F1 scores, and histograms of balanced and unbalanced data are reported.
