Fast imaging method and system based on synthetic analysis deconvolution network

Abstract

A fast imaging method based on a synthetic analysis deconvolution network includes the steps of performing filter learning using an image library on the synthetic analysis deconvolution network integrated with DN and MRF prior information; in the process of filter training, updating the filter and sparse coefficients by the rotation technique, the semi-secondary separation technique and the iterative reweighted least squares method; on the basis of the trained filter, obtaining an initial image from an undersampled K space, and establishing an image reconstruction model; updating the sparse coefficient and a target image on the reconstruction model by using the rotation technique, the semi-quadratic-element separation technique and the iterative reweighted least squares method; and satisfying a convergence condition to obtain a reconstructed image. The invention improves the accuracy of image reconstruction to a certain extent, and also provides a fast imaging system based on a synthetic analysis deconvolution network using the fast imaging method of the synthetic analysis deconvolution network, which can obtain the reconstructed image with a high accuracy.

Description

technical field [0001] The invention belongs to the technical field of medical imaging, and in particular relates to a fast imaging method based on a synthesis analysis deconvolution network Background technique [0002] Inverse problems such as medical magnetic resonance (MRI) imaging are usually transformed into minimizing a suitable energy function, and this minimization provides a balance between data consistent terms and prior information terms. Early algorithms for deriving prior information included total variation, wavelet sparsity, and more generally Markov random fields. Traditional predefined dictionary / transformation methods may not be able to sparsely represent the objective function. To address this issue, adaptive learning methods such as dictionary updating have attracted enormous attention. In general, there are two classical learning strategies to represent image feature structures: models based on synthetic learning and models based on analytical learnin...

AI Technical Summary

Problems solved by technology

[0003] In the sparse model of synthetic learning, the dictionary learning method uses the traditional block-based sparse representation for image reconstruction, so it has an essential shortcoming: since the image blocks subdivided into blocks are independent of each other, meaningful Important spatial structure will be lost

The existing technology usually uses the analysis operator or the synthesis operator alone and has achieved certain results, but there is no technology that combines the synthesis operator and the analysis operator into the same learning process and constraints

Images