CN110276725A - A fast method for removing artifacts in free-breathing lower abdomen MRI - Google Patents

Abstract

The present invention relates to a method for rapidly removing artifacts in MRI of the lower abdomen under free breathing, which includes collecting image data with artifacts, preprocessing to obtain several subimages, eliminating artifacts through a correction model, and splicing L subimages to obtain Output image parts. The invention can remove the streak artifact in the high-magnification sub-sampling image in real time, and has clinical potential application value.

Description

A fast method for removing artifacts in free-breathing lower abdomen MRI

technical field

The invention relates to a method for quickly removing artifacts of magnetic resonance imaging of the lower abdomen during free breathing, and belongs to the technical field of image processing.

Background technique

Currently, magnetic resonance imaging plays a vital role in the clinical diagnosis and treatment of abdominal lesions. However, MRI takes a long time, and the abdomen will be affected by breathing movements. Respiratory movement will cause movement and deformation of organs, which will lead to respiratory movement artifacts in the image, which will reduce the resolution and signal-to-noise ratio of the image. In the process of image-guided interventional therapy, there will be inconsistencies between the static guidance information and the position of the moving structure. The phenomenon. In recent years, some researchers have proposed a sequence that can form abdominal images under the subject's free breathing, that is, the 3D golden angle radial stack-of-stars (SOS) sequence. However, in the case of high subsampling, the reconstructed image has obvious streak artifacts. A method based on compressed sensing is usually used to eliminate such artifacts. However, since this method requires iterative calculations, the reconstruction speed is slow, and it is not feasible for clinical application. Therefore, there is an urgent need for a rapid removal method for MRI artifacts in the freely breathing lower abdomen.

Contents of the invention

In order to be able to solve the deficiencies of slow elimination of streak artifacts and difficulty in clinical application during the subject’s abdominal magnetic resonance imaging under free breathing, the purpose of the present invention is to provide a magnetic resonance image streak that can be down-sampled during free breathing A quick way to remove artifacts.

In order to realize the object of appeal, the technical scheme of the present invention is as follows:

A method for rapidly removing artifacts in MRI of the freely breathing lower abdomen, consisting of the following steps:

1) Collect the image data I0 of the subject with streak artifacts in the free breathing state of the abdomen; preferably, the magnetic resonance scan sequence collected is a 3D golden angle radial stack-of-stars (SOS) scan sequence ; Preferably, the image data I0 is obtained by downsampling by 5-20 times.

2) Preprocessing the collected I0 to obtain L sub-images, so that the down-sampled image data I0 is input into the correction model M; The square is a sliding window, L sub-regions are obtained by traversing the image I0, and the overlapping degree between adjacent sub-regions is 70-85%.

3) performing artifact removal on the L sub-images respectively through a pre-trained correction model M, preferably, the correction model M is a non-linear mapping obtained through supervised learning, and is constructed by a stacked convolutional autoencoder network; wherein The correctly labeled data (ground true) of the corrected model M is an artifact-free image obtained by sampling the full K-space, and the training sample of the model M comes from an artifact-free image obtained by sampling 5%-20% of the corresponding correctly labeled data in the K-space ; Preferably, the magnetic resonance scanning sequence in which the sample is correctly marked with data is a 3D golden angle radial stack-of-stars (SOS) scanning sequence to complete the full sampling of K space.

4) The splicing of L subimages after artifact removal is called image I1, which is used as the output after artifact removal.

To sum up, a fast removal method for MRI artifacts in free breathing lower abdomen, the main innovation is: by using the full sampling data as the correct standard data (ground true) and the down sampling data as the sample input , build a stacked convolutional auto-encoder network (Stacked Convolutional Auto-encoders, SCAE), and get the nonlinear mapping between the input and output image data, and then through this network, the streak artifacts caused by downsampling can be removed, and the Fast Removal of Streak Artifacts in Free-breathing Downsampled Magnetic Resonance Images.

Description of drawings

Fig. 1 is the overall flowchart of the present invention;

Fig. 2 is a reconstruction result figure among the present invention

Detailed ways

The present invention will be described below through specific examples, but the present invention is not limited thereto.

As shown in Fig. 1, a kind of fast removal method of the MRI artifact of the free breathing lower abdomen of the present invention, at first collect the image data I0 of the subject with the streak artifact under the free breathing state of the abdomen; and then collect the collected I0 is preprocessed to obtain L sub-images; then the L sub-images are respectively removed from the artifacts through the pre-trained correction model M; finally, the L sub-images after the artifact removal are spliced and called image I1, which is used as the post-artifact removal Output.

In this embodiment, the magnetic resonance scanning sequence used by the subject in the free breathing state of the abdomen is a 3D golden angle radial stack-of-stars (SOS) scanning sequence, and then the image with artifacts is obtained by downsampling by 5 times Data I0.

In the preprocessing method of this embodiment, a square with a side length of 7 pixels is used as a sliding window, and L sub-areas are obtained after traversing the image I0, and the overlapping degree between adjacent areas is 80%.

The correction model M in this embodiment adopts a stacked convolutional autoencoder network. The correctly labeled data of the sample of the network is an artifact-free image obtained by sampling in the full K space, and the training samples of the network come from 20% of the corresponding correctly labeled data. The image with artifacts obtained from K-space sampling, and the MRI scan sequence in which the sample is correctly labeled with data is a 3D golden angle radial stack-of-stars (SOS) scan sequence to complete the full sampling of K-space. In addition, the network consists of 9 hidden layers, including:

1) 1 convolutional layer with a filter size of 128×5×5;

2) One 2×2 maximum pooling layer;

3) 64 convolutional layers with a filter size of 5×5;

4) One 2×2 maximum pooling layer;

5) 64 deconvolution layers with a filter size of 5×5;

6) A 2×2 upsampling layer;

7) 128 deconvolution layers with a filter size of 5×5;

8) A 2×2 upsampling layer;

9) 1 deconvolution layer with filter size 5×5.

Fig. 2 shows to adopt the present invention to obtain the removal image of MRI artifact of free breathing lower abdomen; The image provides the result figure of three typical subjects, and the left column is the training image with streak artifact of input, and the middle The column is the result image output by the stacked convolutional autoencoder network (SCAE), and the right column refers to the gold standard image (ground true). The results show that this method can remove streak artifacts well, and the error with the gold standard only appears in the abdominal border, concentrated in the fat highlight signal area.

Claims (5)

1. a kind of quick minimizing technology for freely breathing lower abdomen magnetic resonance imaging artifact, which is characterized in that by following steps group At:

1) acquisition subject has the image data I0 of jail-bar artifacts under the free respiratory state of abdomen；

2) collected I0 is pre-processed to obtain L subgraph；

3) L subgraph is subjected to artifact removal by the good calibration model M of precondition respectively；

4) the L subgraph splicing after removing artifact is known as image I1, as the output for going pseudo- movie queen.

2. a kind of quick minimizing technology for freely breathing lower abdomen magnetic resonance imaging artifact as described in claim 1, feature Be, when image data I0 of the acquisition subject under the free respiratory state of abdomen used magnetic resonance imaging sequence For 3D Golden Angle radial direction stack-of-stars (SOS) scanning sequence；The image data I0 is adopted by 5-20 times of drop Sample obtains.

3. a kind of quick minimizing technology for freely breathing lower abdomen magnetic resonance imaging artifact as described in claim 1, feature It is, the preprocess method includes: that use side length for the square of 5-10 pixel be sliding window, and traversal image I0 is obtained L sub-regions, the degree of overlapping between adjacent subarea domain are 70-85%.

4. a kind of quick minimizing technology for freely breathing lower abdomen magnetic resonance imaging artifact as described in claim 1, feature It is, the calibration model M is the Nonlinear Mapping obtained by supervised learning, by stacking convolution autoencoder network structure It builds；Wherein the correct labeled data of the sample of calibration model M is the artifact free image that full K spatial sampling obtains, the training sample of model M What this part 5%-20% K spatial sampling from corresponding correct labeled data obtained has artifacts.

5. calibration model M as claimed in claim 4, which is characterized in that the magnetic resonance imaging sequence of the correct labeled data of sample For it is fully sampled to complete the space K with 3D Golden Angle radial direction stack-of-stars (SOS) scanning sequence.