Multi-contrast MR image denoising for parallel imaging using multilayer perceptron

Abstract

For clinical diagnosis in MRI, multiple examinations are commonly performed to acquire various contrast images. This article presents a learning-based denoising method for parallel imaging to enhance the quality of multi-contrast images so that the imaging time can be accelerated highly. Multi-contrast images share structural information and coil geometry. The proposed method adopts the multilayer perceptron (MLP) model to save the sharable and redundant information among the multi-contrast images. The images are divided into patches, which are used as the input and output of MLP. A geometry factor map is additionally used to provide noise amplification information of the accelerated MR images. Computer simulation demonstrates that the use of multi-contrast images and geometry factor contributes to the quality of the reconstructed images. The proposed method reconstructs high-quality images without impairing details from the subsampled intermediate images, and it shows better results than previous denoising methods