CN104280705B - magnetic resonance image reconstruction method and device based on compressed sensing - Google Patents

Abstract

The present invention provides a magnetic resonance image reconstruction method and device based on compressed sensing, wherein the method includes: obtaining under-sampled original K-space data through magnetic resonance scanning; Image reconstruction; when the reconstruction result of the image reconstruction tends to converge, adding a disturbance to the original K-space data, and counting the variance of the reconstruction result after adding the disturbance to obtain a first variance map; extracting the first variance A region of interest in the figure; image masking is performed on the region of interest to obtain a second variance map; a nonlinear function is used to add the second variance map to the image reconstruction result to obtain a reconstructed image. The above method and device improve the accuracy of the magnetic resonance image reconstruction results.

Description

Magnetic resonance image reconstruction method and device based on compressed sensing

technical field

The present invention relates to the technical field of magnetic resonance imaging, in particular to a method and device for reconstructing magnetic resonance images based on compressed sensing.

Background technique

The theory of compressed sensing takes advantage of the sparsity of the signal, and only needs to collect a small number of samples to reconstruct the original data with high quality. In recent years, the theory of compressed sensing (CS) has been rapidly developed and applied in fast magnetic resonance imaging. Using this theory, the original image can be reconstructed from the under-sampled k-space, thereby reducing the number of acquisition lines of k-space and reducing the Scanning time, to achieve the purpose of fast imaging.

In the traditional MRI image reconstruction method based on compressed sensing, the detail information of the low-contrast image is easy to lose during the image reconstruction process, and the information loss will be more serious with the increase of the acceleration multiple, which will affect the accuracy of the MRI image reconstruction results.

Contents of the invention

Based on this, it is necessary to provide a magnetic resonance image reconstruction method and device based on compressed sensing that can improve the accuracy of magnetic resonance image reconstruction results.

A method for reconstructing magnetic resonance images based on compressed sensing, the method comprising:

Obtain undersampled raw K-space data through magnetic resonance scanning;

performing image reconstruction on the original K-space data by means of compressed sensing;

After the reconstruction result of the image reconstruction tends to converge, adding a disturbance to the original K-space data, and counting the variance of the reconstruction result after adding the disturbance to obtain a first variance map;

extracting a region of interest in the first variogram;

performing image mask processing on the region of interest to obtain a second variance map;

The second variance map is added to the image reconstruction result by using a nonlinear function to obtain a reconstructed image.

In one of the embodiments, the step of adding a disturbance to the original K-space data, and counting the variance of the reconstruction result after adding the disturbance to obtain the first variance map includes:

Randomly select a preset number of sampling points in the original K-space data, and set the sampling points to zero so that the original K-space is converted into new K-space data;

performing image reconstruction on the new K-space data using a compressed sensing method to obtain a reconstruction result image;

Repeating the above two steps for a preset number of times to obtain a reconstruction result image equal to the preset number of times;

The variance in the reconstruction result image is counted to obtain a first variance map.

In one of the embodiments, the step of extracting the region of interest in the first variance map includes:

A foreground area and a background area in the first variance map are extracted, and the foreground area is used as an area of interest.

In one of the embodiments, the step of using a nonlinear function to add the second variance map to the image reconstruction result to obtain the reconstructed image includes:

Performing normalization processing on the magnitude of the image reconstruction result to obtain a processed reconstruction magnitude map;

Using a nonlinear function to perform nonlinear transformation on each pixel in the reconstructed magnitude map to obtain a transformed image;

adding the second variance map to the transformed image to obtain an added image;

A nonlinear inverse transformation is performed on each pixel in the added image to obtain a reconstructed image.

In one of the embodiments, the step of normalizing the magnitude of the reconstruction result image to obtain the processed reconstruction magnitude map includes:

Taking a modulus value for each pixel in the reconstruction result image to obtain an amplitude map;

Acquiring the maximum grayscale value in the magnitude map, dividing each pixel in the magnitude map by the maximum grayscale value to obtain the reconstructed magnitude map.

A magnetic resonance image reconstruction device based on compressed sensing, the device comprising:

The spatial data acquisition module is used to obtain undersampled original K-space data by magnetic resonance scanning;

The first image reconstruction module is used to perform image reconstruction on the original K-space data using a compressed sensing device;

The first variance map acquisition module is used to add disturbance to the original K-space data when the reconstruction result of the image reconstruction tends to converge, and count the variance of the reconstruction result after adding the disturbance to obtain the first variance map;

A region of interest extraction module, configured to extract a region of interest in the first variance map;

A second variance map acquisition module, configured to perform image mask processing on the region of interest to obtain a second variance map;

The second image reconstruction module is configured to use a nonlinear function to add the second variance map to the image reconstruction result to obtain a reconstructed image.

In one of the embodiments, the first variance map acquisition module includes:

A spatial data update module, configured to randomly select a preset number of sampling points in the original K-space data, and set the sampling points to zero so that the original K-space is converted into new K-space data;

The third image reconstruction module is used to reconstruct the image of the new K-space data by using the compressed sensing device to obtain the reconstruction result image;

A repeat execution module, configured to repeatedly execute the above two steps for a preset number of times, to obtain a reconstruction result image equal to the preset number of times;

A variance statistics module, configured to count the variance in the reconstruction result image to obtain a first variance map.

In one of the embodiments, the region-of-interest extracting module is further configured to extract the foreground region and the background region in the first variance map, and use the foreground region as the region-of-interest.

In one of the embodiments, the second image reconstruction module includes:

A normalization processing module, configured to perform normalization processing on the magnitude of the image reconstruction result to obtain a processed reconstruction magnitude map;

A nonlinear transformation module, configured to use a nonlinear function to perform nonlinear transformation on each pixel in the reconstructed magnitude map to obtain a transformed image;

An image addition module, configured to add the second variance map to the transformed image to obtain an added image;

The inverse transform module is used to perform non-linear inverse transform on each pixel in the added image to obtain a reconstructed image.

In one of the embodiments, the normalization processing module includes:

A module for obtaining a modulus value, which is used to obtain a modulus value for each pixel in the reconstruction result image to obtain an amplitude value map;

The reconstructed amplitude map acquisition module is configured to obtain the maximum gray value in the amplitude map, and divide each pixel in the magnitude map by the maximum gray value to obtain the reconstructed amplitude map.

In the method and device for reconstructing magnetic resonance images based on compressed sensing, in the process of reconstructing magnetic resonance images by compressive sensing methods, disturbances are added to the original K-space data, so that the reconstruction results are slightly changed, and the variance of the reconstruction results after adding disturbances is counted The information in the variance map is used to supplement the information lost in the reconstruction results, which improves the accuracy of the MRI image reconstruction results.

Description of drawings

Fig. 1 is a schematic flow chart of a magnetic resonance image reconstruction method based on compressed sensing in an embodiment;

Fig. 2 is a schematic flow chart of adding disturbance to the original K-space data in one embodiment, and statistically adding the variance of the reconstruction result after disturbance to obtain the first variance map step;

Fig. 3 is a schematic flow chart of the step of obtaining a reconstructed image by adding the second variance map to the reconstruction result by using a nonlinear function in one embodiment;

Fig. 4 is a display diagram of images generated by each step in the process of image reconstruction based on the compressed sensing magnetic resonance image reconstruction method in an embodiment;

Fig. 5 is a schematic structural diagram of a magnetic resonance image reconstruction device based on compressed sensing in an embodiment;

Fig. 6 is a schematic structural diagram of the first variance map acquisition module in an embodiment;

Fig. 7 is a schematic structural diagram of a second variance map acquisition module in an embodiment;

Fig. 8 is a schematic structural diagram of a normalization processing module in an embodiment.

detailed description

In order to make the object, technical solution and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described here are only used to explain the present invention, not to limit the present invention.

As shown in Figure 1, in one embodiment, a kind of magnetic resonance image reconstruction method based on compressive sensing is provided, and this method comprises the following steps:

Step 101 , obtain undersampled original K-space data through magnetic resonance scanning.

Step 102, image reconstruction is performed on the original K-space data by using the compressive sensing method.

In this embodiment, the image reconstruction equation performed by the compressed sensing method is: min||ψx|| ₁ , st||F _p xy|| ₂ ≤ε, where Ψ is called a fixed sparse transformation, and y is the measured K space signal, F _p is the under-sampled Fourier operator, and ε is a parameter related to the noise level of the signal. Commonly used sparse transforms include wavelet transform, principal component analysis, finite difference transform, etc. x is the reconstructed image to be solved, and the reconstruction equation is solved by the conjugate gradient descent method to obtain the reconstruction result x

Step 103, when the reconstruction result of the image reconstruction tends to converge, adding disturbance to the original K-space data, and counting the variance of the reconstruction result after the disturbance is added to obtain a first variance map.

In this embodiment, converting min||ψx|| ₁ , st||F _p xy|| ₂ ≤ ε into Lagrangian is expressed as: Among them, λ is the regularization coefficient, which is selected according to experience. The above formula is solved by the conjugate gradient descent method, so that the reconstruction result tends to converge.

In one embodiment, step 103, adding disturbances to the original K-space data, and counting the variance of the reconstruction results after adding the disturbances to obtain the first variance map includes:

Step 201, randomly select a preset number of sampling points in the original K-space data, and set the sampling points to zero to convert the original K-space data into new K-space data. The value range of the preset number is: greater than or equal to 1 and less than or equal to 20, and the default number is 10 in priority.

Step 202, using the compressed sensing method to perform image reconstruction on the new K-space data to obtain a reconstruction result image.

Step 203 , repeating the above two steps for a preset number of times to obtain the reconstructed result images equal to the preset number of times. Repeat step 201 and step 202 for N times to obtain _N reconstructed result images {x ₁ , x ₂ .

In step 204, the variance in the reconstructed image is counted to obtain a first variance map.

In this embodiment, the obtained N reconstruction result images have the same spatial position, that is, any spatial point of the reconstruction result image has N reconstruction values, and the variance is calculated for each spatial point in the reconstruction result image, and finally the first Variance plot.

Step 104, extract the region of interest in the first variance map.

Specifically, the foreground area and the background area in the first variance map are extracted, and the foreground area is used as the region of interest. In one embodiment, a threshold method is used to distinguish the foreground area from the background area in the first variance map, and by selecting a threshold, the pixels greater than the threshold in the image are marked as the foreground area, and the pixels less than the threshold are marked as background area.

Step 105, performing image mask processing on the region of interest to obtain a second variance map.

Step 106, using a nonlinear function to add the second variance map to the reconstruction result to obtain a reconstructed image.

In this embodiment, the low-contrast information lost in the reconstruction process of the magnetic resonance image is restored through the second variance map, and the low-contrast information is restored by adding the second variance map to the reconstruction result by using a nonlinear function. In the above-mentioned magnetic resonance image reconstruction method based on compressed sensing, in the process of magnetic resonance image reconstruction by the compressed sensing method, disturbance is added to the original K-space data, so that the reconstruction result changes slightly, and the variance map of the reconstruction result after adding the disturbance is counted, The information in the variance map is used to supplement the information lost in the reconstruction result, which improves the accuracy of the reconstruction result of the magnetic resonance image.

As shown in FIG. 3, in one embodiment, step 106, using a nonlinear function to add the second variance map to the reconstruction result, and obtaining the reconstructed image includes:

Step 301 , performing normalization processing on the magnitude of the image reconstruction result to obtain a processed reconstruction magnitude map. Specifically, the modulus value is taken for each pixel in the reconstruction result image to obtain the magnitude map; the maximum gray value in the magnitude map is obtained, and each pixel in the magnitude map is divided by the maximum gray value, Obtain the reconstructed magnitude map.

Step 302, using a nonlinear function to perform nonlinear transformation on each pixel in the reconstructed magnitude map to obtain a transformed image. In one embodiment, the nonlinear function is a power function, f(x)=x ^a , x is a pixel in the reconstructed magnitude map, and a is a preset value, preferably, a=5.

Step 303, adding the second variance map to the transformed image to obtain an added image. The added image is obtained by adding the pixel point values in the second variance map to the pixel point values in the transformed image.

Step 304, performing non-linear inverse transformation on each pixel in the added image to obtain a reconstructed image.

A non-linear inverse transformation is performed on each pixel in the added image through the non-linear function f(x)=x ^1/a . Wherein, x is a pixel in the added image, and a is a preset value, preferably, a=5.

In one embodiment, as shown in FIG. 4 , it is a display diagram of images generated by each step in the image reconstruction process based on the compressed sensing-based magnetic resonance image reconstruction method. As shown in Figure 4, 4a is the original image, 4b is the image obtained after image reconstruction by the compressed sensing method, wherein the low contrast information of the white circle marked areas 40 and 42 in 4a disappears in 4b after being reconstructed by the compressed sensing method;

4c is the first variance map obtained by counting the variance in 4b. It can be seen from 4c that the map contains part of the low-contrast information of 40 and 42 in 4a. 4d is the second variance map obtained after performing image mask processing on the region of interest in 4c. 4e is the reconstructed magnitude map obtained after normalizing the magnitude of 4b. 4f is a transformed image obtained by performing nonlinear transformation on each pixel in 4e. 4g is an added image obtained by adding the 4f and 4d images. 4h is the reconstructed image obtained after inverse transforming 4g. It can be clearly seen that the lost low-contrast information 40 and 42 in 4b reappeared in 4h. Contrast information is restored to improve the accuracy of MRI image reconstruction.

As shown in Figure 5, in one embodiment, a kind of magnetic resonance image reconstruction device based on compressive sensing is provided, and this device comprises:

The spatial data acquisition module 50 is configured to obtain under-sampled original K-space data through magnetic resonance scanning.

The first image reconstruction module 51 is configured to perform image reconstruction on the original K-space data using a compressed sensing device.

The first variance map acquisition module 52 is configured to add disturbance to the original K-space data when the reconstruction result of the image reconstruction tends to converge, and calculate the variance of the reconstruction result after adding the disturbance to obtain the first variance map.

A region of interest extracting module 53, configured to extract a region of interest in the first variance map.

In one embodiment, the ROI extracting module 53 is further configured to extract the foreground area and the background area in the first variance map, and use the foreground area as the ROI.

The second variance map acquisition module 54 is configured to perform image mask processing on the region of interest to obtain a second variance map.

The second image reconstruction module 55 is configured to use a nonlinear function to add the second variance map to the image reconstruction result to obtain a reconstructed image.

As shown in Figure 6, in one embodiment, the first variance map acquisition module 52 includes:

The spatial data update module 520 is configured to randomly select a preset number of sampling points in the original K-space data, and set the sampling points to zero to convert the original K-space data into new K-space data.

The third image reconstruction module 521 is configured to perform image reconstruction on the new K-space data by using the compressed sensing device to obtain a reconstruction result image.

The repeat execution module 523 is configured to repeatedly execute the above two steps for a preset number of times to obtain the reconstructed result image equal to the preset number of times.

The variance statistics module 524 is configured to calculate the variance in the reconstructed image to obtain a first variance map.

As shown in Figure 7, in one embodiment, the second image reconstruction module 55 includes:

The normalization processing module 550 is configured to perform normalization processing on the magnitude of the image reconstruction result to obtain a processed reconstruction magnitude map.

The nonlinear transformation module 551 is configured to use a nonlinear function to perform nonlinear transformation on each pixel in the reconstructed magnitude map to obtain a transformed image.

The image addition module 552 is configured to add the second variance map to the transformed image to obtain an added image.

The inverse transform module 553 is configured to perform non-linear inverse transform on each pixel in the added image to obtain a reconstructed image.

As shown in Figure 8, in one embodiment, the normalization processing module 550 includes:

The modulus value acquisition module 5500 is configured to acquire a modulus value for each pixel in the reconstruction result image to obtain a magnitude map.

The reconstructed magnitude map acquisition module 5501 is used to obtain the maximum gray value in the magnitude map, and divide each pixel in the magnitude map by the maximum gray value to obtain the reconstructed magnitude map.

The above-mentioned embodiments only express several implementation modes of the present invention, and the descriptions thereof are relatively specific and detailed, but should not be construed as limiting the patent scope of the present invention. It should be pointed out that those skilled in the art can make several modifications and improvements without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent for the present invention should be based on the appended claims.

Claims (10)

1. A magnetic resonance image reconstruction method based on compressed sensing, said method comprising: Obtain undersampled raw K-space data through magnetic resonance scanning; performing image reconstruction on the original K-space data by means of compressed sensing; When the reconstruction result of the image reconstruction tends to converge, adding a disturbance to the original K-space data, and counting the variance of the reconstruction result after adding the disturbance to obtain a first variance map; extracting a region of interest in the first variogram; performing image mask processing on the region of interest to obtain a second variance map; The second variance map is added to the reconstruction result of the image reconstruction by using a nonlinear function to obtain a reconstructed image. 2. The method according to claim 1, wherein the step of adding disturbance to the original K-space data, and statistically adding the variance of the reconstruction result after disturbance to obtain the first variance map includes: Randomly select a preset number of sampling points in the original K-space data, and set the sampling points to zero so that the original K-space is converted into new K-space data; performing image reconstruction on the new K-space data using a compressed sensing method to obtain a reconstruction result image; Repeating the above two steps for a preset number of times to obtain a reconstruction result image equal to the preset number of times; The variance in the reconstruction result image is counted to obtain a first variance map. 3. The method according to claim 1, wherein the step of extracting the region of interest in the first variance map includes: A foreground area and a background area in the first variance map are extracted, and the foreground area is used as an area of interest. 4. The method according to claim 1, wherein the step of adding the second variance map to the reconstruction result of the image reconstruction by using a nonlinear function to obtain a reconstructed image comprises: Performing normalization processing on the magnitude of the reconstruction result of the image reconstruction to obtain a processed reconstruction magnitude map; Using a nonlinear function to perform nonlinear transformation on each pixel in the reconstructed magnitude map to obtain a transformed image; adding the second variance map to the transformed image to obtain an added image; A nonlinear inverse transformation is performed on each pixel in the added image to obtain a reconstructed image. 5. The method according to claim 4, wherein the step of normalizing the magnitude of the reconstruction result of the image reconstruction to obtain a processed reconstruction magnitude map includes: Taking a modulus value for each pixel in the reconstruction result of the image reconstruction to obtain an amplitude map; Acquiring the maximum grayscale value in the magnitude map, dividing each pixel in the magnitude map by the maximum grayscale value to obtain the reconstructed magnitude map. 6. A magnetic resonance image reconstruction device based on compressed sensing, characterized in that the device comprises: The spatial data acquisition module is used to obtain undersampled original K-space data by magnetic resonance scanning; The first image reconstruction module is used to perform image reconstruction on the original K-space data using a compressed sensing device; The first variance map acquisition module is used to add disturbance to the original K-space data when the reconstruction result of the image reconstruction tends to converge, and count the variance of the reconstruction result after adding the disturbance to obtain the first variance map; A region of interest extraction module, configured to extract a region of interest in the first variance map; A second variance map acquisition module, configured to perform image mask processing on the region of interest to obtain a second variance map; The second image reconstruction module is configured to use a nonlinear function to add the second variance map to the reconstruction result of the image reconstruction to obtain a reconstructed image. 7. The device according to claim 6, wherein the first variance map acquisition module comprises: A spatial data update module, configured to randomly select a preset number of sampling points in the original K-space data, and set the sampling points to zero so that the original K-space is converted into new K-space data; The third image reconstruction module is used to reconstruct the image of the new K-space data by using the compressed sensing device to obtain the reconstruction result image; The repeated execution module is used to repeatedly execute the steps performed by the above-mentioned spatial data update module and the third image reconstruction module for a preset number of times to obtain a reconstruction result image equal to the preset number of times; A variance statistics module, configured to count the variance in the reconstruction result image to obtain a first variance map. 8. The device according to claim 6, wherein the ROI extraction module is further configured to extract a foreground area and a background area in the first variogram, and use the foreground area as an ROI. 9. The device according to claim 6, wherein the acquisition module of the second image reconstruction module comprises: A normalization processing module, configured to perform normalization processing on the magnitude of the reconstruction result of the image reconstruction to obtain a processed reconstruction magnitude map; A nonlinear transformation module, configured to use a nonlinear function to perform nonlinear transformation on each pixel in the reconstructed magnitude map to obtain a transformed image; An image addition module, configured to add the second variance map to the transformed image to obtain an added image; The inverse transform module is used to perform non-linear inverse transform on each pixel in the added image to obtain a reconstructed image. 10. The device according to claim 9, wherein the normalization processing module comprises: A module for taking a modulus value, which is used to take a modulus value for each pixel in the reconstruction result of the image reconstruction to obtain an amplitude map; The reconstructed amplitude map acquisition module is configured to obtain the maximum gray value in the amplitude map, and divide each pixel in the magnitude map by the maximum gray value to obtain the reconstructed amplitude map.