A fast magnetic resonance imaging method and device based on deep convolutional neural network

Abstract

A rapid magnetic resonance imaging method and apparatus based on a deep convolutional neural network. The method comprises: step S1 (S101). constructing a deep convolutional neural network; step S2 (S102), acquiring offline magnetic resonance image data, training the deep convolutional neural network, and learning a mapping relationship between an undersampled magnetic resonance image and a fully sampled image; and step S3 (S103), reconstructing a magnetic resonance image by using the deep convolutional neural network learned in step S2 (S102). In the rapid magnetic resonance imaging method and apparatus based on a deep convolutional neural network, a large amount of collected magnetic resonance data is used to train an offline deep convolutional neural network and learn a mapping relationship between an undersampled magnetic resonance image and a fully sampled image, so as to fully use a large quantity of offline magnetic resonance images and develop prior information thereof, such that the offline deep convolutional neural network may restore more fine structures and image features from undersampled magnetic resonance data, and an undersampling factor and imaging precision of magnetic resonance imaging are improved.

Description

technical field [0001] The invention relates to the technical field of magnetic resonance imaging, in particular to a fast magnetic resonance imaging method and device based on a deep convolutional neural network. Background technique [0002] The successful application of compressed sensing theory must meet the following three conditions: ① the signal is sparse, ② the artifacts caused by undersampling are incoherent in the transform domain, and ③ the reconstruction results have good consistency with the sampled data. In magnetic resonance images, these three conditions can be well met. In the classic fast magnetic resonance imaging model based on compressed sensing, there are usually two components: data fitting item and sparse regularization item. Suppose the reconstructed MRI image is m, ψ represents the sparse transformation from the pixel domain to the sparse domain, and F u Represents the undersampling operator of K space, y is the K space data measured in the scan, ...

AI Technical Summary

Problems solved by technology

[0005] However, most of the traditional fast MRI reconstruction methods are based on the compressed sensing framework, which only uses part of the acquired K-space data and develops image sparsity to constrain the imaging model for MRI image reconstruction, while a large number of offline MRI Data are underutilized and development of prior information remains limited

Images