Ke Chen, Professor, Department of Mathematics and Statistics, University of Strathclyde, UK

Chong Chen, Associate Professor

A Self-supervised Dual-Domain Network for Low-Dose CT Denoising

15:30-16:30 March 13 (Wednesday), N202

A recent trend in deep learning research is to develop effective networks that can overcome the usual difficulty of lacking ground truth labels. This talk presents our recent work on self-supervised learning methods as applied to low dose Computed Tomography (LDCT) denoising, with the advantage of not requiring any labelled data. Previous and related self-supervised methods operate only in the image domain, ignoring valuable priors in the sinogram domain. We propose a dual-domain method that addresses limitation of related work. Specifically, we split an input sinogram into two subsets based on the positions of detector cells to generate paired training data with high similarity and independent noise. These sub-sinograms are then restored to their original size using 1-D interpolation and learning-based correction. To achieve adaptive and moderate smoothing in the sinogram domain, we integrate Dropblock, a type of convolution layer with regularization, into our network, and set a weighted average between the denoised sinograms and their noisy counterparts, leading to a well-balanced dual-domain approach.   Numerical experiments will show that our method outperforms popular non-learning and self-supervised learning methods, demonstrating its effectiveness and superior performance, and yet produces results close to supervised learning.  

This is joint work with Dr. Ran An and Prof. Hongwei Li of Capital Normal University in Beijing.