CN106898000A - A kind of automatic division method of magnetic resonance imaging cerebral gray matter core group - Google Patents

Abstract

The invention discloses a method for automatically segmenting brain gray matter nuclei in magnetic resonance imaging. The method uses standard reference images and matching atlases to automatically obtain seed points located in each brain gray matter nuclei through calculation; using the seed points The image to be segmented is preprocessed to eliminate the blurred area between adjacent nuclei, and combined with the level set segmentation method, the segmentation contour of the target nuclei is automatically obtained. The present invention can more effectively deal with the variability of nuclei in different images, and the result of segmentation is more accurate. The invention can accurately segment specific nuclei in the brain, such as substantia nigra and red nucleus, and can help the diagnosis and pathological research of diseases such as Parkinson's disease to a large extent.

Description

An Automatic Segmentation Method of Brain Gray Matter Nuclei in Magnetic Resonance Imaging

technical field

The invention relates to the technical field of magnetic resonance imaging, in particular to a method for segmenting brain gray matter nuclei in magnetic resonance imaging.

Background technique

Magnetic resonance imaging technology has been widely used in the field of medical diagnosis. As an important means of diagnosis and treatment, its biggest feature is that it can provide rich image contrast information. With the continuous improvement of the spatial resolution of magnetic resonance imaging, magnetic resonance imaging technology can show the living fine gray matter tissue in the deep part of the human brain that cannot be observed before, such as the red nucleus, substantia nigra, dentate nucleus and some other located in the human brain. Brain nuclei of the basal ganglia, etc. Many studies have shown that these nuclei are closely related to some neurodegenerative diseases (such as Parkinson's disease and Wilson's disease), especially the substantia nigra has an important relationship with Parkinson's disease. Therefore, the accurate segmentation of these specific nuclei can provide important assistance for effective surgical treatment (such as deep brain stimulation) and the study of the pathological mechanism of related diseases. However, traditional magnetic resonance T1-weighted images cannot clearly display these fine nuclei. Magnetic resonance quantitative susceptibility map is a new and important development in magnetic resonance imaging technology in recent years. Based on the calculation of spatial magnetic susceptibility, it can provide new image contrast information. At the same time, since the quantitative magnetic susceptibility imaging method is more sensitive to the quantitative determination of changes in histopathological structures, it is more meaningful to segment specific nuclei in the quantitative magnetic susceptibility map.

However, due to the variable morphology of these tissues, the currently commonly used atlas-based segmentation methods cannot achieve accurate and effective segmentation, and the results are quite different from the results of the gold standard segmentation map. The level set segmentation method is a classic and effective automatic segmentation method, which can cope with variable tissue forms. However, due to the close space between these small nuclei located in the deep brain area, the classic level set method cannot effectively distinguish such as The substantia nigra and red nuclei are adjacent nuclei. This affects both diagnostic and research outcomes.

Contents of the invention

The purpose of the present invention is to provide a method for automatic segmentation of gray matter nuclei in magnetic resonance imaging brain, which overcomes the defects in the existing segmentation technology, and proposes the segmentation of image preprocessing means combined with seed point disconnection and level set method.

The concrete technical scheme that realizes the object of the invention is:

A method for automatically segmenting brain gray matter nuclei in magnetic resonance imaging, characterized in that the method includes the following specific steps:

Step 1: Register the image to be segmented with a standard reference image to obtain a transformation matrix for image registration;

Step 2: Transform the original atlas matching the standard reference image according to the transformation matrix to obtain a new atlas matching the image to be segmented;

Step 3: Linearly map the obtained new atlas to the image to be segmented to obtain the region of interest of the gray matter nuclei;

Step 4: Obtain the positions of different nuclei seed points by calculating the center of gravity in each region of interest;

Step 5: Preprocessing the image to be segmented by using the disconnection between seed points in different nuclei in the magnetic resonance image;

Step 6: Adjust the parameters, use the level set segmentation model to segment the preprocessed image, and draw the segmented contour map; where:

Described step 5 specifically comprises:

i) Set the number of iterations N;

ii) Find the shortest spatial path connecting the two sub-points;

iii) Draw the intensity profile of the corresponding grayscale image in this spatial path;

ⅳ) Using a threshold-based method to determine whether each pixel point passed by this spatial path belongs to one of the nuclei; remove those pixels that do not belong to any nuclei from the image, that is, set the gray value to 0;

v) Repeat step ii), step iii), step iv) N times, and end the iteration.

In the iterative process, the spatial shortest path of each search cannot pass through the pixel points that were eliminated in the previous iteration, and the spatial path connecting the two sub-points changes with the number of iterations.

The level set segmentation model is expressed as follows:

In the formula, Ω ₁ and Ω ₂ represent the inner and outer regions of the contour C, respectively; λ ₁ , λ ₂ and υ represent positive weight factors, respectively. x and y represent the positions of pixels in the image respectively; f ₁ (x) and f ₂ (x) are two functions that approximate the gray value of the inner and outer areas of the contour C; K _σ is a Gaussian function, and its scale coefficient is σ; the process of solving the minimization of the energy function ε is the process of obtaining the outline of the nuclei to be segmented. The invention utilizes the split Bregman iterative method to solve the minimum value of the energy function ε, and the contour C can evolve to the boundary of the tissue. In the iterative calculation, the level set function φ(x) is referenced to represent the contour C.

The invention utilizes the disconnectivity between multiple seed points to preprocess the image, and then combines the method of level set segmentation to segment these specific nuclei in the human brain. Compared with the results of classical level set segmentation, the present invention can effectively distinguish those nuclei that are closely adjacent in space. Compared with the currently commonly used atlas-based segmentation method, the present invention can more effectively deal with the variability of nuclei in different images, and the segmentation results are more accurate. The invention can accurately segment specific nuclei (such as substantia nigra, red nucleus, etc.) in the brain, which can greatly help the diagnosis and pathological research of diseases such as Parkinson's disease.

Description of drawings

Fig. 1 is a flowchart of the present invention;

Fig. 2 is the specific flowchart of image preprocessing steps of the present invention;

3 is a specific schematic diagram of an image preprocessing step in an embodiment of the present invention;

Fig. 4 is the comparison figure to the segmentation result of red nucleus and substantia nigra and the result with existing segmentation method in the embodiment of the present invention;

Fig. 5 is the comparative figure of the segmentation accuracy that the present invention adopts different segmentation methods to red nucleus;

Fig. 6 is a comparison diagram of the segmentation accuracy of substantia nigra using different segmentation methods according to the present invention.

detailed description

The present invention will be described in further detail in conjunction with the following specific embodiments and accompanying drawings. The process of implementing the present invention, conditions, experimental scheme methods, etc., except the content specifically mentioned below, are common knowledge and common knowledge in the art, and the present invention has no special limitation content.

Based on the quantitative magnetic susceptibility map with very good contrast of deep brain nuclei, the present invention adopts an image preprocessing method aiming at the disconnection of seed points and combines the level set method to segment specific gray matter nuclei deep in the brain.

The present invention will be described in detail below in conjunction with the accompanying drawings and embodiments.

Example

Segmentation of the red nucleus and substantia nigra of the brain's deep nuclei

The collected magnetic resonance image data is acquired by a multi-echo gradient echo sequence, and the data comes from a 3T magnetic resonance imaging equipment system (Siemens MAGNETOM Trio a Tim 3T), and the number of echoes used is 8.

The complex data collected by the gradient echo sequence undergoes steps such as phase fitting, phase unwrapping, background field removal, and a morphology-based dipole inversion algorithm (Morphology Enabled Dipole Inversion, MEDI) to reconstruct the brain transection magnetization rate graph.

Referring to FIG. 1 , it is a flow chart of the present invention. After the quantitative magnetic susceptibility map is obtained by calculation, it is first necessary to register the calculated images of each layer with the standard template image. Using the same registration mode, the segmentation atlas corresponding to the standard template space can be deformed to obtain the atlas corresponding to the image space to be segmented. In this embodiment, the above steps use the software "Diffeomap" provided by Issel Anne L. Lim to register the images. The second step is to use the atlas to obtain regions of interest through the graph search method, and each region of interest represents the approximate location of different nuclei. The selection of the seed point in this embodiment is based on the calculated barycenter position of each region of interest, and its barycenter position is obtained by calculating the geometric center of the region according to the gray value of each pixel in the region as a weighting factor.

Since the region of interest can roughly find out the location of the nuclei, the various sub-points found by the above method are located in the red nucleus or the substantia nigra. The level set profile cannot distinguish two adjacent nuclei, but because of the segmentation, it is known that the seed points located in the two adjacent nuclei are not connected, so this disconnection can be used as prior knowledge to perform segmentation on the image to be segmented. preprocessing. The flow chart of the preprocessing steps is shown in Figure 2, and the specific description is as follows:

1) In the process of each iteration, use the A-STAR algorithm to find the shortest spatial path between the two sub-points. It should be noted that the shortest path searched during each iteration is constantly changing, because the subsequent processing will remove some pixels that passed through the previous path (points in the black circle on the left of Figure 3), and search again every time The shortest path between the two subpoints of will bypass the previously removed pixel locations.

2) Draw the intensity profile of the corresponding grayscale image in this path (as shown in the right figure of Figure 3).

3) Due to the influence of the uneven distribution of the image gray level, the simple threshold method cannot be judged. A method similar to full width at half maximum is used to judge whether each pixel in this path belongs to the substantia nigra or the red nucleus or neither the red nucleus nor the substantia nigra. These pixels that do not belong to the red nucleus or the substantia nigra are removed from the image (the gray value is set to 0).

4) Repeat the above step 1) to step 3); when enough image pixels that neither belong to the substantia nigra nor the red nucleus are removed, the blurred area between the red nucleus and the substantia nigra disappears.

For the preprocessed image, the RSF model proposed by Chan and Vese is used to solve the model by introducing the level set function to obtain the final segmentation contour. The specific expressions of the model and solution are as follows:

Here H _∈ is a smooth step function,

In the formula, λ ₁ , λ ₂ and υ represent positive weight factors respectively. x and y respectively represent the position of the pixel in the image. f ₁ (x) and f ₂ (x) are two functions that approximate the gray value of the inner and outer regions of the contour C. K _σ is a Gaussian function whose scale coefficient is σ. The level set function φ(x) is used to represent the contour, and the present invention uses the split Bregman iterative method to solve the minimum value of the energy function ε(φ, f ₁ , f ₂ ), and the contour will gradually evolve to the edge of the tissue with the iteration, and the completion segmentation.

This embodiment is the brain data of a healthy and normal volunteer. The data comes from a Siemens 3.0T magnetic resonance imaging system. The data acquisition uses a 12-channel head coil, and the scan uses an 8-echo gradient echo sequence. The parameters are: TR=60ms, ΔTE =6.8ms, flip angle 15°, field of view (FOV) 240x180mm ² , image resolution 384*288, pixel size 0.625*0.625mm ² , 96 layers in total, layer thickness 2mm. The comparison of the results after image segmentation is shown in Figure 4 (the outline represents the segmentation results of substantia nigra and red nucleus): (a) is the result of the segmentation method based on the atlas; (b) the segmentation result of the classic level set method; (c) Segmentation results of the present invention; (d) gold standard for segmentation. Among them, the segmentation gold standard was manually drawn by an expert using the software ITK-SNAP 3.2 to draw the deep nuclei ROI on the quantitative magnetic susceptibility map, so as to evaluate the segmentation accuracy.

Figure 5 and Figure 6 show the quantitative evaluation statistics for segmentation. Among them, the Dice coefficient is used as the quantitative index of segmentation accuracy, which is obtained by the following formula:

The similarity Dice coefficient refers to the ratio of the number of pixels of the correct segmentation result to the entire segmented area (including all areas of manual segmentation and automatic segmentation), which is sensitive to the difference in the size and position of the two areas, and the value range is [0, 1], 1 means completely consistent.

It can be seen from the segmentation results in Fig. 4, Fig. 5 and Fig. 6 that the nuclei segmentation accuracy obtained by the present invention is better than that based on the atlas registration method and the classic level set segmentation method.

The protection content of the present invention is not limited to the above embodiments. Without departing from the spirit and scope of the inventive concept, changes and advantages conceivable by those skilled in the art are all included in the present invention, and the appended claims are the protection scope.

Claims (3)

1. an automatic segmentation method of magnetic resonance imaging brain gray matter nuclei, is characterized in that, the method comprises the following concrete steps: Step 1: Register the image to be segmented with a standard reference image to obtain a transformation matrix for image registration; Step 2: Transform the original atlas matching the standard reference image according to the transformation matrix to obtain a new atlas matching the image to be segmented; Step 3: Linearly map the obtained new atlas to the image to be segmented to obtain the region of interest of the gray matter nuclei; Step 4: Obtain the positions of different nuclei seed points by calculating the center of gravity in each region of interest; Step 5: Preprocessing the image to be segmented by using the disconnection between seed points in different nuclei in the magnetic resonance image; Step 6: Adjust the parameters, use the level set segmentation model to segment the preprocessed image, and draw the segmented contour map; where: Described step 5 specifically comprises: i) Set the number of iterations N; ii) Find the shortest spatial path connecting the two sub-points; iii) Draw the intensity profile of the corresponding grayscale image in this spatial path; ⅳ) Using a threshold-based method to determine whether each pixel point passed by this spatial path belongs to one of the nuclei; remove those pixels that do not belong to any nuclei from the image, that is, set the gray value to 0; v) Repeat step ii), step iii), step iv) N times, and end the iteration. 2. automatic segmentation method according to claim 1, is characterized in that, in iterative process, the spatial shortest path of each search can not pass through the pixel point that is rejected in last iteration, the space that connects between two kinds of sub-points The path changes with the number of iterations. 3. automatic segmentation method according to claim 1, is characterized in that, described level set segmentation model is as follows expression: $\mathrm{<span\; class="notranslate">\; \&epsiv;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; C</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 2</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}{\mathrm{<span\; class="notranslate">\; \&lambda;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; \&Integral;</span><span\; class="notranslate">\; \&lsqb;</span>{<span\; class="notranslate">\; \&Integral;</span>}_{{\mathrm{<span\; class="notranslate">\; \&Omega;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}{\mathrm{<span\; class="notranslate">\; K</span>}}_{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span>{<span\; class="notranslate">\; |</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}\mathrm{<span\; class="notranslate">\; d</span>}\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; \&rsqb;</span>\mathrm{<span\; class="notranslate">\; d</span>}\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; v</span>}<span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; C</span>}<span\; class="notranslate">\; |</span>$ In the formula, Ω ₁ and Ω represent the inner and outer regions of the contour C, respectively; λ ₁ , λ ₂ and υ represent positive weight factors, respectively. x and y represent the positions of pixels in the image respectively; f ₁ (x) and f ₂ (x) are two functions that approximate the gray value of the inner and outer areas of the contour C; K _σ is a Gaussian function, and its scale coefficient is σ; the process of solving the minimization of the energy function ε is the process of obtaining the outline of the nuclei to be segmented.