A Pseudo Lesion Generation Method for Deep Learning Based Chest X-Ray Lung Disease Detection

摘要

Lung diseases remain to be fatal even in the modern and technologically advanced world. When using artificial intelligent (AI) methods for lung disease diagnosis, it is better to give the location of the lesion areas, which makes the AI diagnosis more convincing. However, compared with general objects detection or segmentation, the annotation of medical images requires professional knowledges and more time, which limited the application of deep learning technology in medical image analysis. In this work, we proposed a pseudo lesion generation method, which can use annotated lesion CXR and normal CXR to create new annotated lesion CXRs. Two publicly available datasets, i.e. RSNA and ChestX-Det10 were employed for performance evaluation. The experimental results showed that the proposed pseudo lesion generation method can improved about 4% of the network performances.