Medical wireless capsule endoscopy is an effective method for diagnosis and evaluation of gastrointestinal diseases. However, due to energy and size limitations, it produces low-resolution images, which makes it difficult to detect and diagnose the abnormality and may even lead to an incorrect diagnosis. Recently, endoscopy methods with improved resolution have been shown to be more effective than conventional endoscopy approaches in disease detection and characterization, and it is expected that they will have the same success in the field of capsule endoscopy. In this study, a novel method based on deep learning techniques is proposed that can generate high-resolution counterparts of low-resolution endoscopic images. Conditional GANs and spatial attention blocks are combined to increase the resolution by 8x, 10x and 12x. Extensive qualitative and quantitative analyses show that the proposed method is more successful than the recent deep super-resolution technologies, such as Deep Back Projection Network (DBPN) and Residual Channel Attention Networks (RCAN).
Capsule endoscopy, deep super resolution, spatial attention blocks, generative adversarial networks