Chest radiograph registration and its application based on deep learning segmentation masks

Medical image registration is of vital importance for clinical diagnosis and treatment. The registration performance based on deep learning algorithms has been found to be more accurate when compared to that of conventional registration methods. In order to apply deep learning algorithms to the registration of serial chest radiographs, the current research conducted preprocessing on original chest radiographs by using a mask, then the mask was used for ResUnet registration network training, and finally the evaluation of the registration model was performed. Results showed that the model based on a deep learning mask and deep learning registration was able to approach good registration performance on chest radiographs, and the model can be used as a potential tool on temporal subtraction of sequential chest radiographs.