Classification of chest radiography from general radiography using deep learning approach

Classifying x-ray images into individual classes of body parts is needed, when they are mixed without proper labels. This paper proposes a hierarchical training of convolutional neural network (CNN)-based framework, for classifying chest posterior–anterior (PA) x-ray images from other 12 classes. The first model is constructed for filtering chest PA from the other classes, before constructing the second model to separate the rest of the 12 classes. This is beneficial to address class-imbalanced and overfitting problems, with assists of class weighting and data augmentation. The proposed method achieves promising performances with precision and recall of 100% and F0.5 of 99%.