Brain tumor diagnosis system based on magnetic resonance images and hybrid numerical segmentation algorithm

Abstract

This invention relates to a brain tumor diagnostic system based on MRI images and a hybrid numerical segmentation algorithm, belonging to the field of medical image processing and computer-aided diagnosis technology. It addresses the problems of existing brain tumor diagnostic systems, such as missegmentation, missed segmentation, or loss of detail, leading to reduced diagnostic accuracy and efficiency. The brain tumor diagnostic system based on MRI images and a hybrid numerical segmentation algorithm includes: an image acquisition module for acquiring MRI images of a patient's brain; an image processing module for processing the MRI images using a hybrid numerical segmentation algorithm to segment the brain tumor region and obtain image segmentation results; and a diagnostic output module for generating and outputting diagnostic auxiliary information based on the image segmentation results. This invention supports arbitrary domain segmentation, significantly reduces the number of iterations, and possesses unconditional stability, further improving the diagnostic accuracy and efficiency of the brain tumor diagnostic system.

Description

Technical Field

[0001] This invention relates to the field of medical image processing and computer-aided diagnosis technology, and in particular to a brain tumor diagnostic system based on magnetic resonance imaging and a hybrid numerical segmentation algorithm. Background Technology

[0002] Medical imaging technology is a crucial foundation of modern clinical diagnosis and treatment. Among these technologies, magnetic resonance imaging (MRI) plays an irreplaceable role in the diagnosis and treatment of brain tumors due to its advantages such as high resolution, high soft tissue contrast, and no ionizing radiation. MRI enables multi-parameter and multi-modal imaging of intracranial tissues, accurately revealing the morphology and structure of lesions and their relationship with surrounding tissues. It is particularly suitable for the precise assessment and preoperative planning of complex lesion areas. Furthermore, MRI technology is less affected by individual patient differences, imaging parameters, and magnetic field environment, exhibiting strong stability and adaptability, and providing high-quality image data for clinical practice. Therefore, it has significant clinical value and broad application prospects in key medical fields such as neurosurgery, oncology, and image-guided therapy. For example, it can be used for grading gliomas, precise surgical navigation guided by multimodal image fusion, and radiotherapy planning.

[0003] However, compared with ordinary natural images, brain MRI tumor images often have characteristics such as low gray-level contrast, blurred boundaries, large differences in lesion morphology, and significant differences between images. They also have significant features such as obvious noise interference, severe intensity inhomogeneity, and tissue heterogeneity. As a result, traditional brain MRI tumor image segmentation techniques such as threshold segmentation, region segmentation, edge detection, cluster analysis, graph theory methods, energy functional segmentation, and texture feature-based segmentation methods are prone to missegmentation and omissions in brain tumor diagnostic systems. Although these traditional image segmentation methods perform reasonably well in simple scenarios, they are sensitive to image quality and edge information. When processing complex brain MRI tumor images, they are prone to missegmentation or loss of details. The accuracy and efficiency of image segmentation need to be improved, which poses a serious challenge to the high-precision and high-efficiency diagnosis of brain tumor diagnostic systems. Summary of the Invention

[0004] The purpose of this invention is to address the problems of missegmentation, missed segmentation, or loss of detail in existing brain tumor diagnostic systems, which lead to reduced diagnostic accuracy and efficiency. This invention proposes a brain tumor diagnostic system based on MRI images and a hybrid numerical segmentation algorithm. The image processing module in this system uses a hybrid numerical segmentation algorithm based on the Allen-Cahn equation and fitting terms to process brain MRI images. This image processing module constructs an image segmentation model based on the Allen-Cahn equation and fitting terms, and uses an operator splitting method for numerical solution. By introducing an alternating direction implicit scheme, image segmentation in any domain can be achieved. Compared with traditional image segmentation methods, this not only enables efficient image segmentation in any domain but also significantly improves computational efficiency and segmentation accuracy, thereby enhancing the diagnostic accuracy and efficiency of the brain tumor diagnostic system.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0006] A brain tumor diagnostic system based on MRI images and hybrid numerical segmentation algorithms includes:

[0007] Image acquisition module, used to acquire MRI images of the patient's brain;

[0008] The image processing module is used to process the brain MRI image using a hybrid numerical segmentation algorithm to segment the brain tumor region and obtain the image segmentation result. The hybrid numerical segmentation algorithm includes the following steps:

[0009] Step 1: Construct an image segmentation model based on the Allen-Cahn equation and fitting terms. This model minimizes the piecewise constant Mumford-Shah energy function. The gradient descent flow equation of this model is:

[0010] (1);

[0011] in, Phase field function Regarding time The partial derivatives; The potential energy function of dihydrazine The derivative of ; Phase field function The Laplace operator; The brain MRI image; The average gray value of the target area; This represents the average grayscale value of the background area. The gradient energy coefficient is related to the interface energy. The parameter is non-negative.

[0012] Step 2: Discretize the gradient descent flow equation in two-dimensional space. superior:

[0013] set up To achieve a uniform grid step size, Denotes the set of center points of the unit, where It is the time step. It is the end time. This is the total number of time steps; Represents the time layer, and ; and The grid is in direction and The number of nodes in each direction is a positive integer; the operator splitting numerical algorithm is used for the gradient descent flow equation:

[0014] (2);

[0015] in, ,and , Defined in the following operator splitting scheme:

[0016] (3);

[0017] (4);

[0018] (5);

[0019] in, , ;

[0020] Step 3: Solve the analytical solutions of equations (3) and (5) using the method of separation of variables, as follows:

[0021] (6);

[0022] (7);

[0023] in, Given the initial conditions, The solution to be sought is the solution obtained from the given conditions. Seeking ;

[0024] Step 4: Solve equation (4) using the alternating direction implicit scheme to obtain the analytical solution. The formula is:

[0025] (9);

[0026] Step 5: Extract the segmentation boundaries of the brain tumor region based on the numerical solutions obtained in Steps 3 and 4 to obtain the image segmentation results;

[0027] The diagnostic output module is used to generate and output diagnostic auxiliary information based on the image segmentation results.

[0028] The present invention has the following beneficial effects:

[0029] This invention proposes a brain tumor diagnostic system based on MRI images and a hybrid numerical segmentation algorithm, specifically comprising an image acquisition module, an image processing module, and a diagnostic output module. Specifically, the image processing module, when processing brain MRI images, combines the Allen-Cahn equation and operator splitting method, employing an alternating direction implicit scheme for efficient equation solving. This supports segmentation of arbitrary domains (including prime pixel images), significantly reduces the number of iterations, and exhibits unconditional stability. It optimizes the limitations of traditional multigrid methods in selecting domain size and overcomes the shortcomings of previous graph correlation segmentation methods in prime-number region selection, improving the accuracy and efficiency of image segmentation, thereby enhancing the diagnostic accuracy and efficiency of the brain tumor diagnostic system. Attached Figure Description

[0030] Figure 1 This is a schematic diagram of the brain tumor diagnostic system according to an embodiment of the present invention;

[0031] Figure 2 A flowchart of a hybrid numerical segmentation algorithm;

[0032] Figure 3 A schematic diagram illustrating the overlay of evolutionary outlines onto the original image at different time steps;

[0033] Figures 4-7 The segmentation evolution process at different time steps includes the initial phase field, intermediate transition state and final stable state;

[0034] Figure 8 The images shown are the initial images and segmentation results of traditional image segmentation methods and the hybrid numerical segmentation algorithm in this invention on fingerprint, blood vessel, brain MRI and texture images.

[0035] Figure 9 A comparison chart showing the image segmentation results of traditional image segmentation methods and the hybrid numerical segmentation algorithm of this invention;

[0036] Figure 10 This is a comparison chart of the normalized total discrete energy evolution curves of traditional image segmentation methods and the hybrid numerical segmentation algorithm in this invention. Detailed Implementation

[0037] The technical solution of the present invention will now be described in detail with reference to the accompanying drawings and preferred embodiments.

[0038] This invention proposes a brain tumor diagnostic system based on magnetic resonance imaging and a hybrid numerical segmentation algorithm, such as... Figure 1 As shown, the system specifically includes an image acquisition module 100, an image processing module 200, and a diagnostic output module 300.

[0039] The image acquisition module 100 is used to acquire brain MRI images of patients from the hospital image archiving and communication system or directly from the MRI equipment, and transmit the acquired brain MRI images to the image processing module 200 for further analysis and processing.

[0040] After receiving a brain MRI image, the image processing module 200 inputs the brain MRI image into a hybrid numerical segmentation algorithm. The algorithm calculates the numerical solution from the brain MRI image and segments the brain tumor region based on the numerical solution, ultimately obtaining the image segmentation result. This result is then output to the diagnostic output module 300. The principle of the image processing module 200 is to first use a mathematical model (Partial Differential Equation, PDE) to describe the evolution and boundary formation of different regions in the image, and then use finite difference and time discretization to approximate the solution to obtain a numerically approximate solution. Then through Contour lines are used to extract segmentation boundaries, thereby achieving image segmentation.

[0041] Specifically, such as Figure 2 As shown, the hybrid numerical segmentation algorithm used by the image processing module 200 specifically includes the following steps:

[0042] Step 1 (S100): Construct an image segmentation model based on the Allen-Cahn equation and fitting terms. This model combines the Allen-Cahn equation and fitting terms, successfully solving the problem of minimizing the piecewise constant Mumford-Shah functional in image segmentation. Therefore, it can be used to minimize the piecewise constant Mumford-Shah energy function. The gradient descent flow equation of the image segmentation model is:

[0043] (1);

[0044] in, Phase field function Regarding time The partial derivatives; The potential energy function of dihydrazine The derivative of ; Phase field function The Laplace operator, phase field function Used to represent the two-phase structure of regions in an image; The image is a brain MRI image, i.e., the initially given image; For the target area ( This refers to the average gray value of the tumor region; Background area ( The average gray value; The gradient energy coefficient is related to the interface energy. It is a non-negative parameter.

[0045] In the above equation, These are the Allen-Cahn equations, used to approximate the uniform curvature motion at the interface (i.e., the interface separating the -1 and 1 stage solutions). Therefore, according to the signal term... The symbol indicates that the interface will shrink or expand, thus forming a segmented curve.

[0046] Step 2 (S200): Discretize formula (1) in two-dimensional space superior:

[0047] set up To achieve a uniform grid step size, Denotes the set of center points of the unit, where It is the time step. It is the end time. This is the total number of time steps. In addition, Represents the time layer, and . and The grid is in direction and The number of nodes in each direction is a positive integer. Formula (1) is solved using the operator splitting numerical algorithm:

[0048] (2);

[0049] in, ,and , Defined in the following operator splitting scheme:

[0050] (3);

[0051] (4);

[0052] (5);

[0053] in, , .

[0054] Step 3 (S300): Solve the analytical solutions of equations (3) and (5) using the method of separation of variables.

[0055] make for It is a primitive, continuous physical quantity. (In the partial differential equation model of image segmentation, an auxiliary function is usually defined) (or written as) This is called the phase field function, which acts as a "virtual physical quantity" to represent different regions in an image. This "physical quantity" is not essentially the grayscale of the image itself, but rather an artificially constructed function evolved through partial differential equations, whose zero level set (i.e., ...) The contour lines (the contour lines) are the image segmentation boundaries we are looking for. The image segmentation problem can be viewed as an interface evolution problem, mathematically similar to the interface motion between two phases of a material in physics. At time... and spatial location The approximate value at, where This is the time step. Therefore, at a given time... Solution of time The equation can be solved analytically using the method of separation of variables, with a time step of [missing information]. The solution is:

[0056] (6);

[0057] (7);

[0058] in, Given the initial conditions; The solution to be sought is the solution obtained from the given conditions. Seeking That is, the numerical result obtained after all calculation steps or iteration steps are completed.

[0059] Step 4 (S400): Solve equation (4) using the alternating direction implicit scheme, i.e., equation (8) is a discrete solution of equation (4).

[0060] In this step, equation (4) is solved using the alternating direction implicit scheme of Dirichlet boundary conditions. The alternating direction implicit method includes the following discretization steps:

[0061] (8);

[0062] in, It is the phase field function along Second-order central difference operator in direction, It is the phase field function along Second-order central difference operator in direction, For time step Numerical solution at time, Phase field function The Time layer.

[0063] Finally, the calculation results of equation (8) are used. And using the method of separation of variables, the analytical solution of equation (4) is:

[0064] (9);

[0065] so, This means: in the iteration to the... During the step, the first in the image The degree to which a pixel belongs to the foreground or background. It is not the grayscale value of the original image, but an "auxiliary variable" or "physical quantity" that evolves during the iteration process. Its distribution can determine the segmented regions of the image.

[0066] Step 5 (S500): Based on the numerical solutions obtained in Steps 3 and 4, extract the segmentation boundaries of the brain tumor region to obtain the image segmentation results, and finally complete the image segmentation.

[0067] Traditional active contour models have some ability to segment heterogeneous regions in brain MRI images, but they still have significant shortcomings in terms of variational regularization, adaptability to complex regions, and solution efficiency, especially the difficulty of handling arbitrary segmentation regions using traditional multigrid methods. The hybrid numerical segmentation algorithm used in the image processing module 200 of this invention combines the Allen-Cahn equation and fitting terms, successfully solving the problem of minimizing the correlation between the piecewise constant Mumford-Shah functional in image segmentation. This enables rapid solution of discrete equation systems, significantly improving computational efficiency. It can also perform image segmentation over arbitrary domains.

[0068] The diagnostic output module 300 is used to generate and output diagnostic auxiliary information based on the image segmentation results obtained by the image processing module 200. The diagnostic auxiliary information includes at least one of the following: a visualized image with the tumor region annotated, the location information of the tumor, and the size information of the tumor.

[0069] The brain tumor diagnostic system proposed in this invention, based on MRI images and a hybrid numerical segmentation algorithm, specifically includes an image acquisition module, an image processing module, and a diagnostic output module. Specifically, the image processing module, when processing brain MRI images, combines the Allen-Cahn equation and operator splitting method, employing an alternating direction implicit scheme for efficient equation solving. This supports segmentation of arbitrary domains (including prime pixel images), significantly reduces the number of iterations, and exhibits unconditional stability. It optimizes the limitations of traditional multigrid methods in selecting domain size and overcomes the shortcomings of previous graph correlation segmentation methods in prime selection region partitioning, improving the accuracy and efficiency of image segmentation, thereby enhancing the diagnostic accuracy and efficiency of the brain tumor diagnostic system.

[0070] Furthermore, to verify the convergence of equation (8), this invention chooses to make the analytical solution a function. The initial values, boundary conditions, and forcing functions are defined. In numerical calculations, the same grid spacing is used in each direction. ,Right now And the numerical solution was calculated. Norm error and maximum norm error:

[0071] (10);

[0072] in, for Norm error, For an exact solution, For numerical solutions, For the maximum norm error, For time step, This is the final time layer.

[0073] The numerical solutions obtained through mesh subdivision are shown in Table 1, where the coarsest mesh is... Set the time step to... And by observing the form of error , , and They are The proportionality constants of norm error and maximum norm error were calculated. The experimental convergence order is defined as:

[0074] (11);

[0075] (12).

[0076] Finally, it was observed that the numerical solution obtained using the ADI difference scheme has second-order accuracy.

[0077] Table 1. ADI Differential Scheme at Time Step Upper maximum norm sum convergence results of norm

[0078]

[0079] The hybrid numerical segmentation algorithm used in the image processing module will be explained below with specific examples.

[0080] Consider in two-dimensional space A composite image above:

[0081] (13);

[0082] in, , Represents the approximate number of grid points. The image is in Figure 3 The first line shows the values. White areas are close to 1, and gray areas are close to 0. The initial phase field is defined. for:

[0083] (14);

[0084] The parameters used in the experiment included: 64×64 grid, interface parameters. , Tolerance tol = 0.2 .

[0085] Figure 3 The first line shows the evolutionary outline superimposed on the original image; the second line shows the evolutionary outline superimposed on the original image according to the present invention. Figure 3 In the diagram, (a), (b), (c), and (d) correspond to the time steps, respectively. And the positive value of the fitted term indicates It will increase; a negative value indicates... It will decrease until the segmentation curve reaches the boundary of the target object, that is, it is positive inside the disk and negative outside the disk.

[0086] Figures 4-7 The numerical solutions are shown at different time steps (step=0, 1, 3, 10). It gradually evolves from the initial state, approximating the target image. The evolutionary process.

[0087] from Figures 4-7 As can be seen from this, when step=0, the numerical solution... The distribution is consistent with the initial phase field, and remains a symmetrical distribution. The shape has a large deviation; when step=1 and step=3, with iteration, The distribution began to gradually shift towards As you move closer, you can observe a transition effect starting to appear in the boundary change area; when step=10... The distribution is almost the same as The coincidence indicates that the phase field evolution is gradually completed and the numerical solution is close to stability.

[0088] in addition, Figure 3 (b) second line and Figure 3 The segmentation accuracy in the first row of (c) is approximate, indicating that the hybrid numerical segmentation algorithm can improve the time efficiency of segmentation evolution to reach a steady state.

[0089] To evaluate the performance of the brain tumor diagnostic system, this embodiment compares the hybrid numerical segmentation algorithm with traditional image segmentation methods on four test problems: fingerprint images, vascular images, brain MRI images, and texture images. Figure 8 The images shown are the initial images and segmentation results of traditional image segmentation methods and the hybrid numerical segmentation algorithm of this invention. Figure 8 (a), (b), (c), and (d) in the table correspond to fingerprint images, vascular images, brain MRI images, and texture images, respectively. The tests were conducted on a 2.20GHz Intel Core i7-14650HX CPU and 16.00 GB of RAM. Other experimental parameters are shown in Table 2.

[0090] Table 2 Experimental parameters

[0091]

[0092] from Figure 8 It can be seen from this that Figure 8 In (a), the third line shows a clearer segmentation than the second line. Furthermore, Figure 8 The segmentation results of the third row in (b), (c), and (d) are almost identical to those of the second row. Furthermore, Table 2 shows that for the data in Table 2... Figure 8 In (c) of the table, the number of iterations is the same for both methods, but for the methods in Table 2... Figure 8 As shown in (a), (b), and (d) of this invention, the hybrid numerical segmentation algorithm requires fewer iterations than traditional image segmentation methods. Therefore, this further demonstrates that the hybrid numerical segmentation algorithm of this invention improves the time efficiency for reaching a stable segmentation evolution state.

[0093] In recent research, multigrid methods have been considered to overcome the problem of slow convergence and accelerate computation. However, multigrid methods have limitations in selecting the number of pixels in the input image. This invention overcomes these constraints by introducing an implicit method with alternating directions.

[0094] Figure 9 The images show a comparison of the image segmentation results of the hybrid numerical segmentation algorithm and the traditional image segmentation method in this invention. (a), (b), and (c) represent the original image, and the segmentation results for 233×239 and 257×263 pixels, respectively, using the hybrid numerical segmentation algorithm. 233×239 and 257×263 are prime numbers that are difficult to segment using the multigrid method. Figure 9 In the table, (d) represents the image segmentation result using a traditional image segmentation method with a 256×384 pixel multigrid. Other parameters are the same as in Table 2. Figure 5 As can be seen from this, the hybrid numerical segmentation algorithm in this invention achieves image segmentation over arbitrary domains.

[0095] Figure 10 This demonstrates the normalized total discrete energy obtained through numerical solutions using traditional image segmentation methods and the hybrid numerical segmentation algorithm of this invention. The evolution. From Figure 10 It is evident that both energies exhibit non-increasing behavior. In particular, the normalized total discrete energy of the hybrid numerical segmentation algorithm in this invention is slightly lower than that of traditional image segmentation methods.

[0096] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0097] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this invention patent should be determined by the appended claims.

Claims

1. A brain tumor diagnosis system based on a magnetic resonance image and a hybrid numerical segmentation algorithm, characterized in that, include: Image acquisition module, used to acquire MRI images of the patient's brain; The image processing module is used to process the brain MRI image using a hybrid numerical segmentation algorithm to segment the brain tumor region and obtain the image segmentation result. The hybrid numerical segmentation algorithm includes the following steps: Step 1: Construct an image segmentation model based on the Allen-Cahn equation and fitting terms. This model minimizes the piecewise constant Mumford-Shah energy function. The gradient descent flow equation of this model is: （1）； in, Phase field function Regarding time The partial derivatives; The potential energy function of dihydrazine The derivative of ; Phase field function The Laplace operator; The brain MRI image; The average gray value of the target area; This represents the average grayscale value of the background area. The gradient energy coefficient is related to the interface energy. The parameter is non-negative. Step 2: Discretize the gradient descent flow equation in two dimensions Up: set up To achieve a uniform grid step size, Denotes the set of center points of the unit, where It is the time step. It is the end time. This is the total number of time steps; Represents the time layer, and ; and The grid is in direction and The number of nodes in each direction is a positive integer; the operator splitting numerical algorithm is used for the gradient descent flow equation: （2）； wherein and , defined in the following operator splitting scheme: （3）； （4）； （5）； wherein , ; Step 3: Solve the analytical solutions of equations (3) and (5) using the method of separation of variables, as follows: （6）； （7）； in, Given the initial conditions, The solution to be sought is the solution obtained from the given conditions. Seek ; Step 4: The analytical solution is obtained by solving equation (4) using the alternating direction implicit scheme The formula is: （9）； Step 5: Extract the segmentation boundaries of the brain tumor region based on the numerical solutions obtained in Steps 3 and 4 to obtain the image segmentation results; The diagnostic output module is used to generate and output diagnostic auxiliary information based on the image segmentation results.

2. The brain tumor diagnostic system according to claim 1, characterized by, The diagnostic auxiliary information includes at least one of the following: a visual image with the tumor region marked, the location information of the tumor, and the size information of the tumor.

3. The brain tumor diagnostic system according to claim 1, characterized by, 。

Images