Chest Radiograph Pathology Categorization via Transfer Learning

The goal of this chapter is to give an overview of the research we have been conducting in automated X-ray pathology detection for the past 10 years, from bag-of-visual-words (BoVW) models to the Convolutional Neural Network (CNN) Deep Learning schemes. Our study was one of the first to suggest the possibility of using non-medical training, using transfer learning from the general imagery to the medical domain. In this chapter we explore deep features that are extracted from intermediate CNN layers in comparison to a set of classical shallow features, including GLCM, PHOG, GABOR, GIST and the more recent, state-of-the-art BoVW model. We investigate the possible benefits of using feature selection techniques on the Deep CNN feature layers. Average AUC results of close to 90% are shown for categorization of 6 different pathologies in a dataset of more than 600 radiographs. This study shows the strength and robustness of the CNN features. We conclude that deep learning with large scale non-medical image databases may be a good substitute, or addition, to domain specific representations which are yet to be available for general medical image recognition tasks. With the BoW schemes, our method won first place in ImageClef competitions. With the DL architectures, we are now able to use the system in real clinical settings.