Identifying Cardiomegaly in ChestX-ray8 Using Transfer Learning

Recently, the National Institutes of Health (NIH) published a chest X-ray image database named "ChestX-ray8", which contains 108,948 X-ray images that are labeled with eight types of diseases. Identifying the pathologies from the clinical images is a challenging task even for human experts, and to develop computer-aided diagnosis systems to help humans identify the pathologies from images is an urgent need. In this study, we applied the deep learning methods to identify the cardiomegaly from the X-ray images. We tested our algorithms on a dataset containing 600 images, and obtained the best performance with an area under the curve (AUC) of 0.87 using the transfer learning method. This result indicates the feasibility of developing computer-aided diagnosis systems for different pathologies from X-rays using deep learning techniques.