Brain region segmentation method based on nuclear magnetic resonance image

Abstract

The invention discloses a brain region segmentation method based on a nuclear magnetic resonance image, and the method comprises the steps: carrying out the brain tissue pre-segmentation processing of the nuclear magnetic resonance image according to a standard brain mask, and obtaining a roughly separated brain tissue image; segmenting and screening according to the brightness value of each tissue in the brain tissue image to obtain brain tissue points, and searching and screening the obtained brain tissue points by using a breadth-first search algorithm to obtain a brain single-connected region image; performing convolution processing on the brain single-connected region image by adopting a convolution kernel, and performing traversal search on the image after convolution processing by utilizing a breadth-first search algorithm to identify whether a plurality of independent connected regions exist in the image, if so, performing screening according to the area of each connected region, and if not, performing screening according to the area of each connected region; the area of the connected region exceeds a preset screening threshold value, and a brain region image is obtained.

Description

Technical Field

[0001] The present invention relates to the technical field of nuclear magnetic resonance image processing, and more particularly to a brain region segmentation method based on nuclear magnetic resonance images. Background Art

[0002] In the medical field, determining the condition of a patient's internal organs is often necessary to diagnose conditions and develop treatment plans. Magnetic resonance imaging (MRI) utilizes the principles of nuclear magnetic resonance (NMR). It detects electromagnetic waves emitted by an external gradient magnetic field, creating an image of the internal structure of an object. MRI offers numerous advantages over other types of medical imaging, such as adjustable parameters, clear imaging, and harmlessness to the human body. Consequently, it has gained widespread application in the medical field.

[0003] However, during the acquisition of MRI images, due to the influence of the image acquisition equipment, problems such as low contrast, low signal-to-noise ratio and low light intensity are prone to occur, resulting in noise in the MRI images. The brightness value is affected by noise, etc., which often leads to image segmentation methods that rely on brightness for specific organ areas (such as brain tissue) to ultimately produce erroneous segmentation results. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a brain region segmentation method based on nuclear magnetic resonance images that can improve the accuracy of brain tissue segmentation.

[0005] To solve the above technical problems, the present invention provides a brain region segmentation method based on magnetic resonance imaging, comprising:

[0006] Perform brain tissue pre-segmentation on the MRI image according to the standard brain mask to obtain a roughly separated brain tissue image;

[0007] Segmenting and screening the brain tissue points according to the brightness values ​​of the tissues in the brain tissue image, and searching and screening the obtained brain tissue points using a breadth-first search algorithm to obtain a brain single-connected region image;

[0008] The convolution kernel is used to convolve the image of the single connected area of ​​the brain, and the breadth-first search algorithm is used to traverse the convolution-processed image to identify whether there are multiple independent connected regions in the image. If so, the connected regions are screened according to their area, and the connected regions whose area exceeds the preset screening threshold are deleted to obtain the brain region image.

[0009] A further technical solution is as follows: the brain region segmentation method based on nuclear magnetic resonance images further includes:

[0010] Taking the intersection of pixel points at various positions of a plurality of brain region images of magnetic resonance images adjacent to the current magnetic resonance image to obtain an image intersection pixel point set;

[0011] The position pixel points of the brain region image of the current magnetic resonance image are screened and supplemented according to the image intersection pixel point set to obtain a final brain region image.

[0012] A further technical solution is that the position pixel points of the brain region image of the current MRI image are screened and supplemented according to the image intersection pixel point set, specifically including:

[0013] traversing and detecting whether each pixel point at each position of the brain region image of the current MRI image is within the image intersection pixel point set, and if not, deleting the pixel point at that position;

[0014] Traverse and detect whether there are position pixels in the image intersection pixel set that do not exist in the brain region image of the current nuclear magnetic resonance image. If so, add the image intersection pixel points to the brain region image of the current nuclear magnetic resonance image.

[0015] A further technical solution is: performing brain tissue pre-segmentation processing on the MRI image according to the standard brain mask to obtain a roughly separated brain tissue image, specifically comprising:

[0016] Sagittal sections were obtained by sectioning along the long axis of the brain in MRI images;

[0017] The sagittal plane was processed according to the standard brain mask to obtain a brain mask image;

[0018] The brain mask image is traversed using the Graham scanning algorithm to calculate the convex hull of the brain mask;

[0019] The coordinate points of the convex hull are screened according to a preset vertical coordinate threshold value, so as to delete the coordinate points whose vertical coordinates are smaller than the preset vertical coordinate threshold value, thereby obtaining a roughly separated brain tissue image.

[0020] A further technical solution is: using the Graham scanning algorithm to traverse the brain mask image to calculate the convex hull of the brain mask, specifically including:

[0021] Use the Graham scanning algorithm to traverse the brain mask image and find the coordinate point with the smallest vertical coordinate in the brain mask;

[0022] Sort the remaining coordinate points in the brain mask according to the polar angle size relative to the coordinate point with the smallest vertical coordinate;

[0023] Add the coordinate points to a stack in the order of sorting. If each coordinate point added to the stack is in the clockwise / counterclockwise direction of the coordinate point at the top of the stack, then keep the coordinate point; otherwise, pop the coordinate point from the stack.

[0024] The convex hull of the brain mask is constructed based on all coordinate points in the stack.

[0025] A further technical solution is: performing segmentation and screening according to the brightness value of each tissue in the brain tissue image to obtain brain tissue points specifically includes:

[0026] The brightness value of each tissue image pixel in the brain tissue image is calculated, and the brightness value of each tissue image pixel is compared with a preset brightness threshold, and tissues with brightness values ​​higher than the preset brightness threshold are screened out to form brain tissue points.

[0027] A further technical solution is: the preset brightness threshold is half of the difference between the brightest pixel value and the darkest pixel value in the image.

[0028] A further technical solution is: the convolution kernel is used to perform convolution processing on the image of the single-connected area of ​​the brain, specifically including:

[0029] A 3*3 convolution kernel is used to perform convolution processing on the brain single connected area image to filter out the noise pixels in the brain single connected area image.

[0030] The beneficial technical effect of the present invention is that: compared with the prior art, the present invention performs brain tissue pre-segmentation processing on the nuclear magnetic resonance image according to the standard brain mask to extract the brain tissue, and at the same time, it can also separate it from the non-brain tissue that has brightness overlap with the brain area, which is beneficial to the segmentation inside the brain area, and then the brain tissue image obtained after the pre-segmentation processing is screened and segmented based on the brightness value, and combined with the breadth-first search algorithm, a single-connected region containing brain tissue can be screened out to obtain a brain single-connected region image, thereby roughly separating the brain area from the non-brain area and retaining most of the brain tissue, and the convolution kernel is used to perform convolution processing on the brain single-connected region image to filter out the noise pixels in the brain single-connected region image, and then the breadth-first search algorithm is used to filter out the noise pixels in the brain single-connected region image. The algorithm traverses and searches the image after convolution processing to identify whether there are multiple independent connected areas in the image. If so, it screens according to the area of ​​each connected area to delete the connected areas with an area greater than a preset area threshold, thereby eliminating a large number of non-brain tissues connected to the brain tissue through noise to obtain a brain area image. It can be seen that the brain area segmentation method based on magnetic resonance imaging of the present invention not only performs brain tissue pre-segmentation processing on the magnetic resonance imaging, deletes some non-brain tissues whose brightness overlaps with the brain tissue, but also filters the noise pixels of the brain single connected area image processed based on the brightness and width-first search algorithm to eliminate the influence of noise, thereby more accurately segmenting the brain tissue and obtaining a more accurate brain area image. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 It is a flowchart of a specific embodiment of the brain region segmentation method based on nuclear magnetic resonance images of the present invention.

[0032] Figure 2 It is a sub-flow diagram of a specific embodiment of the brain region segmentation method based on nuclear magnetic resonance images of the present invention.

[0033] Figure 3 It is an image schematic diagram of the processing process of the brain region segmentation method based on nuclear magnetic resonance images using the present invention. DETAILED DESCRIPTION

[0034] In order to enable those skilled in the art to more clearly understand the objectives, technical solutions and advantages of the present invention, the present invention is further described below with reference to the accompanying drawings and embodiments.

[0035] Reference Figure 1 , Figure 1 The figure is a flow chart of a specific embodiment of a method for brain region segmentation based on MRI images of the present invention. In the embodiment shown in the figure, the method for brain region segmentation based on MRI images includes:

[0036] S100 , performing brain tissue pre-segmentation processing on the MRI image according to a standard brain mask to obtain a roughly separated brain tissue image.

[0037] The MRI image may be a MRI image of the head, in which the brain tissue is mainly located in the middle, while other interfering non-brain tissues are mainly distributed around the brain tissue. In the present invention, a rough separation is first performed on the area around the brain tissue to remove some non-brain tissues with overlapping brightness, that is, the MRI image is pre-segmented to extract the brain tissue. At the same time, the brain tissue and some non-brain tissues with overlapping brightness with the brain area can be separated.

[0038] like Figure 2 As shown, the brain tissue pre-segmentation processing is performed on the MRI image according to the standard brain mask to obtain a roughly separated brain tissue image, which specifically includes the following steps:

[0039] S110. Cut along the long axis of the brain in the MRI image to obtain a sagittal plane.

[0040] In the present invention, the sagittal plane is selected as the reference for pre-segmentation. This plane divides the brain into left and right hemispheres along the long axis of the brain.

[0041] S120 . Process the sagittal plane according to a standard brain mask to obtain a brain mask image.

[0042] In this step, a standard brain mask (i.e., a brain region image mask) is used to preliminarily remove most of the non-brain tissue in the sagittal plane to obtain a brain mask image, i.e., a preliminary brain image.

[0043] S130. Use the Graham scanning algorithm to traverse the brain mask image to calculate the convex hull of the brain mask.

[0044] Specifically, this step includes: using the Graham scanning algorithm to traverse the brain mask image to find the coordinate point with the smallest ordinate in the brain mask; sorting the remaining coordinate points in the brain mask according to the polar angle size relative to the coordinate point with the smallest ordinate; adding the coordinate points to a stack in the sorted order, if each coordinate point currently added to the stack is in the clockwise / counterclockwise direction of the coordinate point at the top of the stack, then retain the coordinate point, otherwise, pop the coordinate point out of the stack; constructing the convex hull of the brain mask based on all the coordinate points in the stack.

[0045] That is, in this step, the brain mask image is traversed by the Graham scanning algorithm to find the coordinate point with the smallest ordinate among all the coordinate points in the brain mask image as the reference point. If there are multiple coordinate points with the smallest ordinate, the coordinate point with the smallest abscissa is selected from these coordinate points as the reference point. Then, the remaining coordinate points are sorted according to their polar angle relative to the reference coordinate point. If there are multiple coordinate points with the same polar angle relative to the reference coordinate point, these coordinate points are sorted according to the distance from the coordinate point to the reference coordinate point. Finally, by maintaining a stack, the coordinate points that meet the requirements are gradually constructed to obtain the convex hull. That is, the coordinate points are added to a stack in the order of sorting. If each coordinate point currently added to the stack is in the clockwise / counterclockwise direction of the coordinate point at the top of the stack (that is, each coordinate point added to the stack during construction is in the clockwise or counterclockwise direction of the previous coordinate point retained in the stack), then the coordinate point is retained. Otherwise, the coordinate point is popped out of the stack. Finally, the convex hull of the brain mask is constructed based on all the coordinate points in the stack.

[0046] S140 , screening the coordinate points of the convex hull according to a preset vertical coordinate threshold, so as to delete the coordinate points whose vertical coordinates are smaller than the preset vertical coordinate threshold, and obtain a roughly separated brain tissue image.

[0047] In this step, a preset vertical coordinate threshold is set to remove the coordinate points of the convex hull that are below the preset vertical coordinate threshold, thereby removing the non-brain tissue below and obtaining a roughly separated brain tissue image.

[0048] S200 , performing segmentation and screening according to the brightness values ​​of each tissue in the brain tissue image to obtain brain tissue points, and using a breadth-first search algorithm to search and screen the obtained brain tissue points to obtain a brain single-connected region image.

[0049] Specifically, in this step, the brightness value of each tissue image pixel in the brain tissue image is calculated, and the brightness value of each tissue image pixel is compared with a preset brightness threshold, and tissues with brightness values ​​higher than the preset brightness threshold are screened out to form brain tissue points, wherein the preset brightness threshold is half of the difference between the brightest pixel value and the darkest pixel value in the image; then the brain tissue points are traversed and searched using a breadth-first search algorithm to find the largest single-connected region in the brain tissue points (i.e., the brain single-connected region). Because real brain tissue is a relatively large single-connected region in the nuclear magnetic resonance image, and some skull noise, although also within the preset brightness threshold, is in another single-connected region that is not connected to the brain tissue, the width information of the breadth-first search can be used to screen out a largest single-connected region from the brain tissue points, thereby separating the brain area from the non-brain area and obtaining a brain single-connected region image.

[0050] S300, using a convolution kernel to perform convolution processing on the image of a single connected region of the brain, and using a breadth-first search algorithm to traverse and search the convolution-processed image to identify whether there are multiple independent connected regions in the image. If so, screening is performed based on the area of ​​each connected region to delete the connected regions whose area exceeds a preset screening threshold to obtain a brain region image.

[0051] In this step, a 3*3 convolution kernel is used to perform convolution processing on the brain single connected area image to filter out the noise pixels in the brain single connected area image.

[0052] It is understandable that if the filtered brain singly connected region image becomes multiple independent connected regions, it indicates that noise pixels exist in the brain singly connected region image, and the multiple independent connected regions were connected by noise before the convolution process. In this embodiment, the connection through noise is set as fine connection, and a breadth-first search algorithm is used to traverse the brain singly connected region image after the convolution process to determine whether there are multiple independent connected regions, thereby determining whether the brain singly connected region image before the convolution process has fine connections caused by the presence of noise. Moreover, after identifying the existence of multiple independent connected regions, in order to prevent connections between some small local real brain tissue from being mistakenly identified as fine connections, that is, to avoid eliminating small local brain tissue, a preset screening threshold is set to filter those connected regions identified as having fine connections. Connected regions exceeding this preset screening threshold are deemed not to belong to local brain tissue and are eliminated, thereby achieving the effect of removing relatively large non-brain noise while retaining local brain tissue. Preferably, in this embodiment, the preset screening threshold can be 3×3.

[0053] S400 , obtaining the intersection of pixel points at various locations of a plurality of brain region images of nuclear magnetic resonance images adjacent to the current nuclear magnetic resonance image to obtain an image intersection pixel point set.

[0054] In the present invention, in order to avoid the problem of erroneous addition of non-brain tissue parts and erroneous deletion of brain tissue that may still occur after the above steps, the current MRI image is further processed in combination with the information of adjacent image sequences in the MRI image to correct the erroneous processing of the brain area image of a single MRI image, making the segmentation process more accurate.

[0055] In this step, the adjacent multiple MRI images are also processed through steps S100-S300 to obtain a brain region image. The brain region images of the adjacent multiple MRI images are then used for intersection processing, deleting non-interesting portions and completing brain details. Specifically, an image registration operation is first performed on the brain region images of the adjacent MRI images. When taking the intersection, a majority rule is preferably employed to count the number of pixels at each position in each image. If more than half of the pixels at each position are attributable to the intersection, the pixel is considered to be included in the brain region. That is, the pixel at this position is considered to be the intersection of the brain region images of the multiple adjacent MRI images and is added to the image intersection pixel set.

[0056] S500 , screening and supplementing positional pixel points of the brain region image of the current magnetic resonance image according to the image intersection pixel point set to obtain a final brain region image.

[0057] Specifically, this step includes: traversing and detecting whether each position pixel point of the brain region image of the current magnetic resonance image is within the image intersection pixel point set, and if not, deleting the position pixel point; traversing and detecting whether there is a position pixel point in the image intersection pixel point set that does not exist in the brain region image of the current magnetic resonance image, and if so, adding the image intersection pixel point to the brain region image of the current magnetic resonance image.

[0058] In this step, the brain region images of the adjacent MRI images are used as a reference for screening the brain region image of the current MRI image. If the position pixel point of the brain region image of the current MRI image is not within the image intersection pixel set, that is, it is not included in most of the adjacent brain region images, it will be discarded. If the position pixel point is included in most of the adjacent brain region images but not in the current brain region image, the pixel point will be added to the current brain region image. The position pixel points are deleted and supplemented through traversal detection to obtain the final brain region image.

[0059] Reference Figure 3 , Figure 3 This is a schematic diagram of the processing process of the brain region segmentation method based on MRI images of the present invention. The arrows in the figure indicate the flow of image segmentation processing through the steps of the present method. The origin image on the far left is the initial MRI image of the head. Following the arrows, the origin_c image is the coarsely separated brain tissue image obtained after processing in step S100, the opt_threshold image is the brain singly connected region image obtained after processing in step S200, the opt_fineconnection image is the brain region image obtained after processing in step S300, and the opt_seq image is the final brain region image obtained after processing in steps S400 and S500.

[0060] In summary, the brain region segmentation method based on nuclear magnetic resonance images of the present invention not only performs brain tissue pre-segmentation processing on the nuclear magnetic resonance images, deletes some non-brain tissues that have brightness overlap with the brain tissue, but also proposes the concept of fine connection, filters the noise pixels of the brain single connected region image after processing based on the brightness and width-first search algorithm, eliminates the influence of the connection between non-brain tissue and brain tissue due to noise, and eliminates a large amount of non-brain tissue that is finely connected to the brain tissue, which can more accurately segment the brain tissue and obtain a more accurate brain region image. In addition, the image sequence information of the image is also utilized, and the brain region image of the current nuclear magnetic resonance image is corrected by the brain region image of the adjacent nuclear magnetic resonance image to obtain a more accurate brain region image.

[0061] It should be further explained that, in the above embodiments, for the sake of simplicity of description, they are all expressed as a series of action combinations, but those skilled in the art should know that the present invention is not limited to the described order of actions, because according to the present invention, certain steps can be performed in other orders or simultaneously.

[0062] The above description is merely a preferred embodiment of the present invention and does not constitute any form of limitation to the present invention. Those skilled in the art may make various equivalent changes and improvements based on the above embodiment. Any equivalent changes or modifications made within the scope of the claims shall fall within the scope of protection of the present invention.

Claims

1. A brain region segmentation method based on magnetic resonance imaging, characterized in that: The brain region segmentation method based on nuclear magnetic resonance images includes: Perform brain tissue pre-segmentation on the MRI image according to the standard brain mask to obtain a roughly separated brain tissue image; Segmenting and screening the brain tissue points according to the brightness values ​​of the tissues in the brain tissue image, and searching and screening the obtained brain tissue points using a breadth-first search algorithm to obtain a brain single-connected region image; The convolution kernel is used to convolve the image of the single connected area of ​​the brain, and the breadth-first search algorithm is used to traverse the convolution-processed image to identify whether there are multiple independent connected regions in the image. If so, the connected regions are screened according to their area, and the connected regions whose area exceeds the preset screening threshold are deleted to obtain the brain region image.

2. The brain region segmentation method based on nuclear magnetic resonance images according to claim 1, characterized in that: The brain region segmentation method based on nuclear magnetic resonance images also includes: Taking the intersection of pixel points at various positions of a plurality of brain region images of magnetic resonance images adjacent to the current magnetic resonance image to obtain an image intersection pixel point set; The position pixel points of the brain region image of the current magnetic resonance image are screened and supplemented according to the image intersection pixel point set to obtain a final brain region image.

3. The brain region segmentation method based on nuclear magnetic resonance images according to claim 2, characterized in that: The filtering and supplementing of position pixels of the brain region image of the current magnetic resonance image according to the image intersection pixel set specifically includes: traversing and detecting whether each pixel point at each position of the brain region image of the current MRI image is within the image intersection pixel point set, and if not, deleting the pixel point at that position; Traverse and detect whether there are position pixels in the image intersection pixel set that do not exist in the brain region image of the current nuclear magnetic resonance image. If so, add the image intersection pixel points to the brain region image of the current nuclear magnetic resonance image.

4. The brain region segmentation method based on MRI according to claim 1, wherein: The brain tissue pre-segmentation processing is performed on the MRI image according to the standard brain mask to obtain a roughly separated brain tissue image, specifically including: Sagittal sections were obtained by sectioning along the long axis of the brain in MRI images; The sagittal plane was processed according to the standard brain mask to obtain a brain mask image; The brain mask image is traversed using the Graham scanning algorithm to calculate the convex hull of the brain mask; The coordinate points of the convex hull are screened according to a preset vertical coordinate threshold value, so as to delete the coordinate points whose vertical coordinates are smaller than the preset vertical coordinate threshold value, thereby obtaining a roughly separated brain tissue image.

5. The brain region segmentation method based on MRI images according to claim 4, characterized in that: The Graham scanning algorithm is used to traverse the brain mask image to calculate the convex hull of the brain mask, specifically including: Use the Graham scanning algorithm to traverse the brain mask image and find the coordinate point with the smallest vertical coordinate in the brain mask; Sort the remaining coordinate points in the brain mask according to the polar angle size relative to the coordinate point with the smallest vertical coordinate; Add the coordinate points to a stack in the order of sorting. If each coordinate point added to the stack is in the clockwise / counterclockwise direction of the coordinate point at the top of the stack, then keep the coordinate point; otherwise, pop the coordinate point from the stack. The convex hull of the brain mask is constructed based on all coordinate points in the stack.

6. The brain region segmentation method based on MRI images according to claim 1, wherein: The segmentation and screening according to the brightness value of each tissue in the brain tissue image to obtain brain tissue points specifically includes: The brightness value of each tissue image pixel in the brain tissue image is calculated, and the brightness value of each tissue image pixel is compared with a preset brightness threshold, and tissues with brightness values ​​higher than the preset brightness threshold are screened out to form brain tissue points.

7. The brain region segmentation method based on MRI images according to claim 6, characterized in that: The preset brightness threshold is half of the difference between the brightest pixel value and the darkest pixel value in the image.

8. The brain region segmentation method based on MRI images according to claim 1, wherein: The convolution kernel is used to perform convolution processing on the image of the single-connected area of ​​the brain, specifically including: A 3*3 convolution kernel is used to perform convolution processing on the brain single connected area image to filter out the noise pixels in the brain single connected area image.

Images