Image Super-Resolution Reconstruction of Pancreatic Carcinoma Based on Edge Repair Generative Adversarial Network
Yuchen Geng, Weimin Zhou · 2022 41st Chinese Control Conference (CCC) · 2022
High-resolution medical imaging of pancreatic carcinoma is of great significance for the early diagnosis of the disease. Despite breakthroughs in accuracy and the capable of generating realistic textures, there are still some problems remain unsolved in current medical image super-resolution reconstruction methods: the edge details are often accompanied with unpleasant artifacts, how do we recover finer textures? To achieve this, we propose an Edge Repair Generative Adversarial Network (ERGAN). The model adds an edge repair network on the basis of the generator and the discriminator network, performs edge detection and repair on the input low-resolution images, then fuses the output edge feature map with the shallow feature map generated by the generator. Finally, the reconstructed images are sent to the discriminator through the up-sampling layer to evaluate the reconstruction effect. In this paper, the model is evaluated on the pancreatic carcinoma data set CPTAC-PDA published on the website of the Cancer Imaging Archive (TCIA). The experimental results show that the images reconstructed by the network model proposed in this paper not only improve the evaluation indicators, but also has a clearer cross-sectional outline and more obvious detailed features.