LDCT Image Reconstruction on Edge Attention Graph Convolutional Network with Perceptual Loss and LoDoPaB-CT

Abstract

Image reconstruction performs a protruding role in medical image analysis. Low-Dose CT (LDCT) scan images are a common diagnostic procedure to identify diseases in the human body. Recent scanners follow deep learning-based post-processing methods for low-dose imaging. Low-dose CT image reconstruction techniques deteriorate image quality, which has an impact on a physician's diagnosis. Therefore, this paper introduces a novel LDCT image reconstruction method based on the edge attention technique utilized in graph convolutional neural networks. The quality of the outcomes is measured through the perceptual loss function. Experimental assessments are shown on the LoDoPaB-CT benchmark dataset. It is demonstrated that the proposed method produced an improved high-quality image compared to both traditional and deep learning-based reconstruction methods qualitatively and quantitatively.