Compressed sensing magnetic resonance image reconstruction method based on ultra-deep dense network

Abstract

The invention relates to a compressed sensing magnetic resonance image reconstruction method based on an ultra-deep dense network. According to the method, a deep network is trained to learn a corresponding relation between a plurality of undersampled k-space data and high-quality magnetic resonance images in an image training database, and the corresponding relation is generalized to an image testing database, so that reconstruction of testing data is realized. The deep network has two characteristics: firstly, the number of network layers is large, the deep network comprises more complex sub-networks, and each sub-network can pre-reconstruct a magnetic resonance image; secondly, the connection between the sub-networks is dense connection, and the sub-networks can generate an image with better reconstruction quality by analyzing a plurality of pre-reconstructed images; and as long as the undersampled k space data in the image test database is used as the input of the deep network, the obtained output is the generated high-quality magnetic resonance image. According to the method, compressed sensing magnetic resonance image reconstruction can be rapidly carried out, and a magnetic resonance image with high definition is generated.

Description

technical field [0001] The invention belongs to the field of magnetic resonance image reconstruction, in particular to a compression sensing magnetic resonance image reconstruction method based on an ultra-deep dense network. Background technique [0002] Magnetic resonance imaging (MRI) is a relatively safe medical tomographic imaging technique that can provide high soft tissue contrast. In the past 20 years, magnetic resonance imaging has made great strides in the fields and research of life science and medicine, and has become an important tool for clinical diagnosis and basic research. However, in clinical medical imaging, the further application and development of magnetic resonance imaging are limited by factors such as long scanning time of magnetic resonance imaging and poor quality of reconstructed images caused by the movement of sampling objects. How to quickly obtain high-quality magnetic resonance images has always been one of the most concerned frontier resear...

AI Technical Summary

Problems solved by technology

First, each sub-network only uses the results of its previous sub-network, but does not fully utilize the information of all previous sub-networks; second, most of these networks are shallow networks with few network layers and relatively simple network structures. , which limits the reconstruction performance of the entire network

Images