Magnetic resonance image reconstruction method based on regularized depth image prior method

Abstract

The invention discloses a magnetic resonance image reconstruction method based on a regularized depth image prior method, and relates to the technical field of magnetic resonance imaging. The invention aims to solve the problem of limitation of a conventional magnetic resonance image reconstruction algorithm based on deep learning, and aims to improve the quality of a reconstructed image and shorten the reconstruction time. The method comprises the following steps: (1) constructing a neural network model; (2) constructing a loss function containing a regular term; (3) acquiring partial k spacedata; (4) obtaining a reconstructed reference image; (5) constructing network input; (6) setting the maximum number of iterations; (7) reconstructing an image by using a network; (8) obtaining a degraded image of the network output image, calculating a loss function in combination with the reference image, and optimizing network parameters; (9) storing the output image with the highest index; and(10) judging whether the number of iterations reaches the maximum number of iterations, if so, outputting the optimal reconstructed image, and otherwise, returning to the step (7). Compared with a convolutional neural network, the method has the advantages that the dependence on data is small, a high-quality reconstructed image can be obtained, and the reconstruction speed is increased.

Description

technical field [0001] The invention relates to the technical field of magnetic resonance imaging, in particular to a method for reconstructing a regularized depth image prior magnetic resonance image using a deep learning network. Background technique [0002] Magnetic resonance imaging has been widely used in medical diagnosis due to its advantages of no ionizing radiation, high imaging resolution, and multiple parameters. However, problems such as too long scanning time hinder the further development and application of MRI technology. In recent years, with the successful application of deep learning in natural image processing, the accelerated magnetic resonance imaging method based on deep learning has received widespread attention. This method first constructs a convolutional neural network, and uses a large number of magnetic resonance image datasets to train the network, and obtains The parameters of the network are optimized to reconstruct the input undersampled dat...

AI Technical Summary

Problems solved by technology

However, deep learning methods need to use a large amount of magnetic resonance data for training to obtain network parameters. Compared with natural images, it is difficult to obtain a large number of medical magnetic resonance image data sets. Therefore, there are certain limitations in the application of deep learning methods in the field of accelerated magnetic resonance imaging. limitation

[0003] In 2017, Dmitry Ulyanov and others proposed a deep image prior method [1] , use the neural network itself to obtain the prior information of the image, and process the image. This method overcomes the dependence of traditional deep learning methods on large data sets, and can achieve image denoising, repair, and deblurring in the case of small sample image data. etc., its performance is better than the traditional compressed sensing method, but due to the lack of prior information provided by a large amount of image data, when the degree of image degradation is high, the reconstruction effect of this method and the method based on convolutional neural network (CNN) still exist to some extent. gap

Images