CT Image Quality Assessment Based on Denoising Auto-Encoder

Due to the restriction of lacking reference image, medical image evaluation is still a hot and challenging problem, in that we proposed a new evaluation strategy for CT images based on denoising auto-encoder (DAE). Firstly, pseudo-reference map of the CT image is obtained based on the DAE network, and then derived from the difference between the original CT and the new map, the global assessment of the CT can be achieved. An adaptive feature fusion module is added in the encoding process and we introduced a countermeasure network in the basic DAE network to reduce cross artifacts in the reconstructed image. In addition, we made up for the evaluation deviation caused by local artifacts by combining global and local results. Experiments on the CT dataset exhibit good result. The Pearson linear correlation coefficient (PLCC) of our method and learned perceptual image patch similarity (LPIPS) is 0. 905656, the Spearman rank correlation coefficient (SRCC) index is 0. 871139. The accuracy of binary classification based on real clinical data is 0. 947891, and the F1 score is 0. 956521.