Brain tumor image segmentation method based on adaptive threshold

Abstract

The invention provides a brain tumor image segmentation method based on an adaptive threshold, and the method comprises the steps: collecting a brain tumor nuclear magnetic resonance image, and analyzing the segmentation problem characteristics of the brain tumor nuclear magnetic resonance image; analyzing the brain tumor nuclear magnetic resonance image optimization segmentation model, and selecting an objective function reflecting a real segmentation condition; constructing a multi-mutation operator optimized numerical spiking neural P system evolved along the gradient direction; a numerical spiking neural P system is optimized based on an objective function and a multi-mutation operator evolved along a gradient direction. Respectively constructing a brain tumor nuclear magnetic resonance image multi-threshold segmentation model of the adaptive threshold optimization numerical spiking neural P system and a brain tumor nuclear magnetic resonance multi-modal image segmentation model of the adaptive threshold optimization numerical spiking neural P system; and carrying out image segmentation by using the two models. The segmentation result is superior to that in the prior art.

Description

Technical Field

[0001] The present invention relates to the technical field of image segmentation, and in particular to a method for segmenting brain tumor images based on an adaptive threshold. Background Art

[0002] Brain tumors are one of the most common diseases of the brain and nervous system. Among the incidence rates of human tumors in China, the average incidence rate of brain tumors is 1.2 per 10,000 people, and the mortality rate after onset exceeds 81%. It is one of the tumors second only to tumors in parts such as the stomach, breast, and esophagus in terms of incidence rate. Clinically, in order to quantitatively evaluate the generation mechanism and tumor characteristics (causes, pathological features, biological behaviors, tumor tissue types, tumor area sizes, number of tumor regions, and tumor area shapes) of brain tumors, and reasonably formulate radiotherapy and treatment plans and pre-determine surgical plans, doctors need to use intelligent segmentation algorithms to discover more case information from magnetic resonance imaging (MRI) of brain tumors. Therefore, accurate segmentation results of brain tumor magnetic resonance images become a key step in understanding and treating brain tumors, and lay a foundation for determining the subsequent surgical resection range and formulating radiotherapy plans. Summary of the Invention

[0003] The present invention provides a method for segmenting brain tumor images based on an adaptive threshold. First, analyze the imaging mechanism of brain tumor magnetic resonance images and the core of the problem of segmenting brain tumor magnetic resonance images. Secondly, analyze the application of different threshold segmentation models in the optimized segmentation of medical images, and select a suitable objective function. Then, design an adaptive threshold optimization numerical pulse neural membrane system. Finally, based on the adaptive threshold optimization numerical pulse neural membrane system, realize the segmentation of brain tumor magnetic resonance images.

[0004] To achieve the above object, the present invention adopts the following technical solutions:

[0005] A method for segmenting brain tumor images based on an adaptive threshold, comprising:

[0006] S1. Collect magnetic resonance images of brain tumors with different case backgrounds, and analyze the characteristics of the segmentation problem of magnetic resonance images of brain tumors;

[0007] S2. Based on the characteristics of the segmentation problem, analyze the optimized segmentation model of magnetic resonance images of brain tumors, and select an objective function that reflects the actual segmentation situation;

[0008] S3. Construct a multi-mutation operator optimized numerical spiking neural P system evolving along the gradient direction. The framework of the multi-mutation operator optimized numerical spiking neural P system evolving along the gradient direction consists of multiple parallel frameworks for expanding the self-optimized numerical spiking neural P system and two director modules. Each expanded self-optimized numerical spiking neural P system is used to generate a candidate population. The two director modules are a probability adjustment module and a population update module respectively, both of which are used to adjust the mutation probability to achieve crossover and selection between individuals.

[0009] S4. Based on the objective function selected in S2 and the multi-mutation operator optimized numerical spiking neural P system evolving along the gradient direction constructed in S3, construct a multi-threshold segmentation model for brain tumor magnetic resonance images of the self-adaptive threshold optimized numerical spiking neural P system.

[0010] S5. Based on the objective function selected in S2 and the multi-mutation operator optimized numerical spiking neural P system evolving along the gradient direction constructed in S3, construct a multi-modal image segmentation model for brain tumor magnetic resonance of the self-adaptive threshold optimized numerical spiking neural P system.

[0011] S6. Use the multi-threshold segmentation model for brain tumor magnetic resonance images and the multi-modal image segmentation model for brain tumor magnetic resonance to perform image segmentation.

[0012] In this specification, the process of the multi-threshold segmentation model for brain tumor magnetic resonance images of the self-adaptive threshold optimized numerical spiking neural P system is as follows:

[0013] Input the brain tumor magnetic resonance image and convert it into 2D data.

[0014] Use bilateral filtering to smooth the brain tumor magnetic resonance image of the 2D data through the non-linear combination average operator of image neighboring values while preserving the image edges.

[0015] Based on the smoothed image, use the Otsu threshold segmentation model as the objective function, and perform brain tumor segmentation through the multi-mutation operator optimized numerical spiking neural P system evolving along the gradient direction, and calculate the Otsu threshold based on the initial threshold. If the Otsu threshold meets the stop condition, output multiple segmentation regions, and use the connectivity algorithm to select the largest segmentation region. Otherwise, use the multi-mutation operator optimized numerical spiking neural P system evolving along the gradient direction to continue searching and calculate the Otsu threshold until the stop condition is met.

[0016] After selecting the largest segmentation region, mark this region and assign a new gray value, assign the gray value of 0 to the unmarked regions, then the bright regions in the image are tumor regions, and the remaining regions are background regions.

[0017] In this specification, the process of the brain tumor MRI multi-modal image segmentation model of the adaptive threshold optimization numerical pulsed neural membrane system is as follows:

[0018] Perform a complete segmentation of the brain tumor by analyzing MRI images of different modalities of the brain tumor to obtain the Flair modality image and the T1 ce modality image;

[0019] Convert the Flair modality image and the T1 ce modality image into 2D data;

[0020] Use bilateral filtering to smooth the Flair modality image and the T1 ce modality image of the 2D data by the non-linear combination average operator of the image neighboring values while preserving the image edges;

[0021] Use the Otsu threshold segmentation model as the objective function, and use AONSNPS to iteratively search for the optimal threshold combination for the smoothed Flair modality image until the stopping condition is met, and perform connectivity processing to filter out the largest segmented area region in the brain tumor segmentation image;

[0022] After finding the largest segmented area region, mark this region and assign a new gray value, assign a gray value of 0 to the unmarked region, then the bright region in the image is the edema region, and the remaining regions are the background regions; obtain the Flair modality image segmentation result;

[0023] Map the Flair modality image segmentation result into the smoothed T1 ce modality image so that the region to be segmented of T1 ce is in the same position as the Flair modality image segmentation result region to avoid the influence of non-tumor regions on the segmentation of the target region;

[0024] Use the Otsu threshold segmentation model as the objective function, and use AONSNPS to iteratively search for the optimal threshold combination for the mapped T1 ce modality image until the stopping condition is met, mark the segmented region and assign a new gray value, assign a gray value of 0 to the unmarked region, then the brightest region in the image is the enhancement region, the second brightest region is the gangrene region, and the remaining regions are the background regions;

[0025] Fuse the Flair modality image segmentation region and the T1 ce modality image segmentation region, and the fusion is as follows:

[0026]

[0027] Among them, I(i,j) is the gray value of the pixel at the i-th row and j-th column after fusion, I(i,j) F is the gray value of the pixel at the i-th row and j-th column of the Flair modality segmentation image, I(i,j) Tis the gray value of the pixel at the i-th row and j-th column of the T1ce modality segmentation image.

[0028] In this specification, the formalization Π of an extended self-optimizing numerical pulse membrane system of degree m, where m is greater than or equal to 1, is as follows:

[0029] Π = (σ1,…,σ m , syn, in, out);

[0030] Where: σ1,…,σ m represent m neurons in the extended self-optimizing numerical pulse membrane system, where any one of σ1,…,σ m is formally represented as: σ i = (Vr i , Pr i , Vr i (0), 1 ≤ i ≤ m); where, is a set of finite numerical variable sets in the i-th neuron, H represents the number of variables, and in a population, H represents the population size;

[0031] is a set of initial numerical value sets of the numerical variable Vr i in the i-th neuron;

[0032] represents a set of finite production function sets in the i-th neuron, and are formally represented as follows:

[0033] where l represents the number of mutation production functions in the i-th neuron; z represents the z-th mutation production function in the i-th neuron; is the k-th variable in the i-th neuron;

[0034] is a set of mutation probabilities in the i-th neuron. If is larger, then the probability that the z-th production function is executed is greater;

[0035] syn = {(σ i , σ j )||1 ≤ i ≤ m AND 1 ≤ j ≤ AND 1 ≠ j};

[0036] in = {1, 2,…, m} is a set of input neurons; out = {1, 2,…, m} is a set of output neurons; σ1,…,σ m are both input neurons and output neurons.

[0037] In this specification, there are three mutation operators in the extended self-optimizing numerical pulse membrane system, namely rand / 2, current-to-rand / 1, and gradient-to-best / 2. The form of rand / 2 is as follows:

[0038]

[0039] The form of current-to-rand / 2 is as shown below:

[0040]

[0041]

[0042] Among them, F random is a random number between [0, 1];

[0043] The form of gradient-to-best / 2 is as shown below:

[0044]

[0045] Among them, G momentum is the momentum gradient, and β is a random number between [0, 1] controlled by the change of the momentum gradient.

[0046] In this specification, at each time step, the updated numerical variable has a final value equal to the existing value plus the other values of other connected presynaptic neurons.

[0047] In this specification, brain tumor magnetic resonance images include T1 images, T1 ce images, T2 images, and Flair images.

[0048] In this specification, for the problem of segmenting brain tumor magnetic resonance images with a size of 240×240, global distribution noise, and blurred tumor boundaries due to brightness differences, the Otsu threshold segmentation model is used as the objective function of the optimized segmentation model for brain tumor magnetic resonance images.

[0049] In this specification, the optimized segmentation model for brain tumor magnetic resonance images includes the maximum entropy model, the minimum error model, and the Otsu threshold segmentation model.

[0050] In summary, the present invention has at least the following beneficial effects:

[0051] The present invention first analyzes the imaging mechanism of nuclear magnetic resonance (NMR) images and the core of the problem of NMR image segmentation of brain tumors. Secondly, it analyzes the application of different threshold segmentation models in the optimized segmentation of medical images, selects a suitable objective function, then designs an adaptive threshold-optimized numerical spiking neural P system, and finally realizes the segmentation of brain tumor NMR images based on the adaptive threshold-optimized numerical spiking neural P system, and the segmentation result is better than the prior art. Description of the Drawings

[0052] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0053] Figure 1 It is a schematic flow chart of the method for segmenting brain tumor images based on an adaptive threshold involved in the present invention.

[0054] Figure 2 It is a schematic diagram of an optimized numerical spiking membrane system with multiple mutation operators evolving along the gradient direction involved in the present invention.

[0055] Figure 3 It is a schematic diagram of a multi-threshold segmentation model of brain tumor NMR images of an adaptive threshold-optimized numerical spiking neural P system involved in the present invention.

[0056] Figure 4 It is a schematic diagram of a segmentation model of brain tumor NMR multi-modal images of an adaptive threshold-optimized numerical spiking neural P system involved in the present invention.

[0057] Figure 5 It is a schematic diagram of the original Flair modality image involved in the present invention.

[0058] Figure 6 It is a schematic diagram of the original T1ce modality image involved in the present invention.

[0059] Figure 7 It is a schematic diagram of the standard segmentation map involved in the present invention.

[0060] Figure 8 It is a schematic diagram of bilateral filtering of Flair modality images involved in the present invention.

[0061] Figure 9 It is a schematic diagram of the segmentation result of Flair modality involved in the present invention.

[0062] Figure 10Schematic diagram of the T1ce modal image based on the Flair segmentation image mapping involved in the present invention.

[0063] Figure 11 Schematic diagram of the T1ce modal segmentation result involved in the present invention.

[0064] Figure 12 Schematic diagram of the multi-modal segmentation result of the adaptive threshold optimization numerical pulse neural membrane system involved in the present invention. Detailed implementation manners

[0065] In the following, only some exemplary embodiments are simply described. As those skilled in the art can recognize, the described embodiments can be modified in various different ways without departing from the spirit or scope of the embodiments of the present invention. Therefore, the drawings and the description are considered to be exemplary in nature rather than restrictive.

[0066] The following disclosure provides many different implementation manners or examples for implementing different structures of the embodiments of the present invention. To simplify the disclosure of the embodiments of the present invention, the components and settings of specific examples are described below. Of course, they are only examples and are not intended to limit the embodiments of the present invention. In addition, the embodiments of the present invention may repeat reference numerals and / or reference letters in different examples. This repetition is for the purpose of simplification and clarity, and does not itself indicate the relationship between the various implementation manners and / or settings discussed.

[0067] The embodiments of the present invention will be described in detail below with reference to the drawings.

[0068] As Figure 1 shown, this embodiment provides a brain tumor image segmentation method based on an adaptive threshold, including:

[0069] S1. Collect magnetic resonance images of brain tumors with different case backgrounds, and analyze the characteristics of the segmentation problems of the magnetic resonance images of brain tumors;

[0070] S2. Based on the characteristics of the segmentation problems, analyze the optimized segmentation model of the magnetic resonance images of brain tumors, and select an objective function that reflects the real segmentation situation;

[0071] S3. Construct a multi-mutation operator optimized numerical pulse neural membrane system that evolves along the gradient direction. The framework of the multi-mutation operator optimized numerical pulse neural membrane system that evolves along the gradient direction is composed of multiple parallel frameworks of extended self-optimized numerical pulse membrane systems and two director modules; each extended self-optimized numerical pulse membrane system is used to generate candidate populations; the two director modules are respectively a probability adjustment module and a population update module, both of which are used to adjust the mutation probability to achieve crossover and selection between individuals;

[0072] S4. Based on the objective function selected in S2 and the multi-mutation operator optimized numerical pulse neural membrane system evolving along the gradient direction constructed in S3, construct a multi-threshold segmentation model for brain tumor magnetic resonance images of the adaptive threshold optimized numerical pulse neural membrane system;

[0073] S5. Based on the objective function selected in S2 and the multi-mutation operator optimized numerical pulse neural membrane system evolving along the gradient direction constructed in S3, construct a multi-modal image segmentation model for brain tumor magnetic resonance of the adaptive threshold optimized numerical pulse neural membrane system;

[0074] S6. Use the multi-threshold segmentation model for brain tumor magnetic resonance images and the multi-modal image segmentation model for brain tumor magnetic resonance to perform image segmentation.

[0075] In some embodiments, as Figure 3 shown, the process of the multi-threshold segmentation model for brain tumor magnetic resonance images of the adaptive threshold optimized numerical pulse neural membrane system is as follows:

[0076] Input the brain tumor magnetic resonance image and convert it into 2D data;

[0077] Use bilateral filtering to smooth the brain tumor magnetic resonance image of the 2D data by the non-linear combination average operator of image neighboring values while preserving the image edges;

[0078] Based on the smoothed image, use the Otsu threshold segmentation model as the objective function, and perform brain tumor segmentation by the multi-mutation operator optimized numerical pulse neural membrane system evolving along the gradient direction, calculate the Otsu threshold based on the initial threshold; if the Otsu threshold meets the stop condition, output the multi-segmentation regions, and use the connectivity algorithm to select the largest segmentation region; otherwise, use the multi-mutation operator optimized numerical pulse neural membrane system evolving along the gradient direction to continue searching and calculating the Otsu threshold until the stop condition is met;

[0079] After selecting the largest segmentation region, label this region and assign a new gray value, assign the gray value of 0 to the unlabeled regions, then the bright regions in the image are the tumor regions, and the remaining regions are the background regions.

[0080] In some embodiments, as Figure 4 shown, the process of the multi-modal image segmentation model for brain tumor magnetic resonance of the adaptive threshold optimized numerical pulse neural membrane system is as follows:

[0081] Perform complete segmentation of the brain tumor by analyzing different modalities of brain tumor magnetic resonance images to obtain the Flair modality image and the T1ce modality image;

[0082] Convert the Flair modality image and the T1ce modality image into 2D data;

[0083] Use bilateral filtering to smooth the Flair modality image and the T1ce modality image of 2D data by non-linearly combining the neighboring values of the image through an averaging operator while preserving the image edges.

[0084] Use the Otsu threshold segmentation model as the objective function, and use AONSNPS to iteratively search for the optimal threshold combination for the smoothed Flair modality image until the stopping condition is met, and perform connectivity processing to filter out the largest segmented area region in the brain tumor segmentation image.

[0085] After finding the largest segmented area region, mark this region and assign a new gray value, and assign a gray value of 0 to the unmarked regions. Then, the bright regions in the image are the edema regions, and the remaining regions are the background regions; obtain the segmentation result of the Flair modality image.

[0086] Map the segmentation result of the Flair modality image into the smoothed T1ce modality image so that the region to be segmented in T1ce is in the same position as the segmentation result region of the Flair modality image to avoid the influence of non-tumor regions on the segmentation of the target region.

[0087] Use the Otsu threshold segmentation model as the objective function, and use AONSNPS to iteratively search for the optimal threshold combination for the mapped T1ce modality image until the stopping condition is met. Mark the segmented regions and assign new gray values, and assign a gray value of 0 to the unmarked regions. Then, the brightest region in the image is the enhancement region, the second brightest region is the gangrene region, and the remaining regions are the background regions.

[0088] Fuse the segmentation regions of the Flair modality image and the T1ce modality image. The fusion is as follows:

[0089]

[0090] Where, I(i,j) is the gray value of the pixel at the i-th row and j-th column after fusion, I(i,j) F is the gray value of the pixel at the i-th row and j-th column of the Flair modality segmentation image, I(i,j) T is the gray value of the pixel at the i-th row and j-th column of the T1ce modality segmentation image.

[0091] In some embodiments, as Figure 2 shown, the formalization Π of an extended autonomous optimization numerical pulse membrane system of degree m, where m is greater than or equal to 1, is as follows:

[0092] Π = (σ1,…,σ m , syn, in, out);

[0093] Where: σ1,…,σm represent m neurons in the extended self-optimizing numerical pulse membrane system, where σ1, …, σ m any one neuron is represented in the form of: σ i =(Vr i , Pr i , Vr i (0), 1 ≤ i ≤ m); where, is a set of finite numerical variable sets in the i-th neuron, H represents the number of variables, and in a population, H represents the population size;

[0094] is a set of initial numerical value sets of the numerical variable Vr i in the i-th neuron;

[0095] represents a set of finite production function sets in the i-th neuron, and are represented in the following form:

[0096] where l represents the number of mutation production functions in the i-th neuron; z represents the z-th mutation production function in the i-th neuron; is the k-th variable in the i-th neuron;

[0097] is a set of mutation probabilities in the i-th neuron. If is larger, the probability that the z-th production function is executed is greater;

[0098] syn = {(σ i , σ j ) || 1 ≤ i ≤ m AND 1 ≤ j ≤ AND 1 ≠ j};

[0099] in = {1, 2, …, m} is a set of input neurons; out = {1, 2, …, m} is a set of output neurons; σ1, …, σ m are both input neurons and output neurons at the same time.

[0100] In some embodiments, as Figure 2 shown, there are three mutation operators in the extended self-optimizing numerical pulse membrane system, namely rand / 2, current-to-rand / 1, and gradient-to-best / 2. The form of rand / 2 is as follows:

[0101]

[0102] The form of current-to-rand / 2 is shown as follows:

[0103]

[0104] Among them, F random is a random number between [0, 1];

[0105] The form of gradient-to-best / 2 is as follows:

[0106]

[0107] Among them, G momentum is the momentum gradient, and β is a random number between [0, 1] controlled by the change of the momentum gradient.

[0108] The technical concept of the present invention is as follows:

[0109] As Figure 1 shown, the method for segmenting magnetic resonance images of brain tumors based on an adaptive threshold of the present invention includes:

[0110] (1) Collect magnetic resonance images of brain tumors with different case backgrounds, analyze the characteristics of the problem of segmenting magnetic resonance images of brain tumors, and elaborate on the importance of this problem in the medical diagnosis process.

[0111] (2) Analyze the optimized segmentation model of magnetic resonance images of brain tumors, and select an objective function that can reflect the true segmentation situation.

[0112] (3) On the basis of optimizing the numerical pulsed neural membrane system, design a multi-mutation operator that evolves along the gradient direction to optimize the numerical pulsed neural membrane system.

[0113] (4) For the problem of segmenting magnetic resonance images of brain tumors, design a multi-threshold segmentation model of magnetic resonance images of brain tumors based on an adaptive threshold to optimize the numerical pulsed neural membrane system.

[0114] (5) Construct a multi-modal image segmentation model of magnetic resonance images of brain tumors based on an adaptive threshold to optimize the numerical pulsed neural membrane system.

[0115] (6) Combine the collected examples of magnetic resonance images of brain tumors, and use the adaptive threshold to optimize the numerical pulsed neural membrane system (i.e., the multi-threshold segmentation model of magnetic resonance images of brain tumors and the multi-modal image segmentation model of magnetic resonance images of brain tumors) to perform multi-modal image segmentation of magnetic resonance images of brain tumors.

[0116] The multi-threshold segmentation model of magnetic resonance images of brain tumors is that the multi-modal image segmentation model of magnetic resonance images of brain tumors can adaptively calculate the number of thresholds required for segmenting images in the face of different images, rather than manually testing the number of thresholds.

[0117] The objective function of S2 and the multi-mutation operator optimized numerical pulse neural membrane system evolving along the gradient direction of S3 constitute the multi-threshold optimized numerical pulse neural membrane system for brain tumor magnetic resonance images of S4 - the multi-threshold image segmentation model. However, different numbers of thresholds need to be set for different brain tumor magnetic resonance images. Therefore, based on the S4 system structure, a multi-threshold pulse optimized numerical pulse neural membrane system with adaptive threshold optimization for brain tumor magnetic resonance images - the adaptive multi-threshold image segmentation model is designed to automatically determine the number of segmentation thresholds required for different brain tumor magnetic resonance images. Based on the multi-threshold optimized numerical pulse neural membrane system with adaptive threshold optimization, a multi-modal optimized numerical pulse neural membrane system for brain tumor magnetic resonance images - the multi-modal image segmentation model is designed. Finally, corresponding to the brain tumor image segmentation method with adaptive thresholds, it is to use the multi-modal image segmentation model - the adaptive threshold multi-modal optimized numerical pulse neural membrane system to achieve the segmentation of brain tumor magnetic resonance images.

[0118] Data collection and problem analysis:

[0119] The dataset is sourced from public medical imaging databases. The public data is selected from the datasets of the internationally authoritative Brain Tumor Segmentation Challenge (BraTs2019) and the Cancer Imaging Archive (TCIA), covering multi-modal magnetic resonance images, including: T1 images, T1 ce (T1-weighted contrast-enhanced) images, T2 images, and Flair images, and providing expert-annotated tumor region labels (including necrotic cores, edema regions, and enhanced tumor regions). Different modal images in the dataset have different characteristics. In T1 images, white matter in the brain shows relatively high signal intensity and appears brighter overall; while gray matter shows relatively low signal intensity and appears darker overall; cerebrospinal fluid appears black in the image. This imaging mode can clearly reveal various anatomical structures and healthy tissues on cross-sections. T1 ce images, through the contrast agent injected before magnetic resonance scanning, can clearly show tissues with sufficient blood supply, which are usually the areas where tumors are located; different lesions show different brightness levels in the image. T2 images clearly depict areas containing free water, especially emphasizing the edema of the surrounding tissues caused by mass effect, resulting in an increase in the overall image brightness. In Flair images, cerebrospinal fluid shows relatively low signal intensity, making the overall image appear darker; while real lesion areas and lesions containing bound water show relatively high signal intensity, making the overall image appear brighter.

[0120] The ultimate goal of brain tumor imaging analysis is to extract important patient-specific clinical information and its diagnostic features. The information in multimodal image data can be used for disease detection, location guidance, and monitoring of intervention measures, ultimately laying the foundation for the clinical diagnosis, staging, and treatment of diseases. The main goal of medical image segmentation is to divide an image into mutually exclusive regions such that each region is spatially continuous and the pixels within the region are homogeneous with respect to a predefined criterion. This definition is a major limitation of most segmentation methods, especially when defining "abnormal tissue types" because the tumors to be segmented are anatomical structures that are usually non-rigid, have complex shapes, vary greatly in size and location, and differ from patient to patient, making it difficult to formulate effective segmentation rules. In brain tumor research, it may be easy to detect the presence of abnormal tissue most of the time. However, accurate and reproducible segmentation and characterization are not straightforward. The contrast agent uptake and image acquisition time after contrast agent injection may vary, which can significantly change the appearance of the tumor, and there is controversy over whether and how to process the non-imaginable components of the tumor by segmentation algorithms. Due to poor spatial resolution, low contrast, unclear boundaries, inhomogeneity, partial volume effects, noise, variability in the shape of objects in the retrieved data, acquisition of some other artifacts, and the lack of an anatomical structure model that can fully capture the possible deformations in each structure. At the same time, for the problem of brain tumor magnetic resonance image segmentation, there are usually difficulties such as image data noise interference, low contrast of image boundary information, complex brain tumor morphology, and high computational complexity.

[0121] Image optimization segmentation model:

[0122] Medical image segmentation occupies a core position in the field of computer vision. The key to solving the problem of brain tumor magnetic resonance image segmentation by optimization methods is to transform the brain tumor magnetic resonance image threshold segmentation problem into a continuous real-number optimization problem. And the objective function in the optimization segmentation method mainly reflects the real segmentation situation. Therefore, the selection of the objective function is the key to the brain tumor optimization segmentation method. Due to its simple algorithm, stable performance, and solid theoretical support, the threshold segmentation model has been widely used in medical image optimization segmentation. Threshold segmentation techniques are mainly divided into two categories: single-threshold segmentation and multi-threshold segmentation. Among them, single-threshold segmentation is a special form of multi-threshold segmentation. The basic idea of multi-threshold segmentation is to simplify a grayscale image into an image with multiple colors, which is achieved by selecting different numbers of grayscale thresholds. Each pixel point in the image is divided into different intervals according to the comparison result of its grayscale value with the threshold. Currently, the threshold segmentation model (brain tumor magnetic resonance image optimization segmentation model) mainly includes: the maximum entropy model, the minimum error model, and the Otsu threshold segmentation model.

[0123] The Otsu threshold segmentation model seeks to find the optimal value of the global threshold to separate the object from the background in the image. In Otsu's method, the histogram is assumed to be bimodal, and an optimal threshold needs to be found to maximize the between-class variance of the image. The goal of the Otsu threshold segmentation model is to find a set of optimal thresholds to maximize the between-class variance. The maximum entropy segmentation model, being a statistical method, has better performance for classification problems in the presence of uncertainty. The minimum error method is characterized by strong robustness and insensitivity to outliers, but it is not applicable to non-linear data or cases with large noise in the data. The Otsu threshold method effectively segments by taking advantage of the obvious separation of the gray level distributions in different regions of the histogram. Therefore, in the problem of segmenting brain tumor magnetic resonance images with a size of 240×240 and having obvious brightness tumor regions with global distributed noise, the Otsu threshold segmentation model is used as the objective function of the optimized segmentation method.

[0124] Multi-mutation operator optimized numerical spiking neural P systems evolving along the gradient direction:

[0125] Based on optimizing the numerical spiking neural P system, the present invention introduces adaptive multi-mutation operator iteration, and proposes corresponding mutation operator selection probability strategies and population update strategies to balance exploration and exploitation capabilities.

[0126] The framework of the multi-mutation operator optimized numerical spiking neural P system evolving along the gradient direction consists of multiple parallel frameworks for expanding and autonomously optimizing the numerical spiking neural P system and two director modules. Each expanded and autonomously optimized numerical spiking neural P system is used to generate candidate populations. The two director modules (probability adjustment module and population update module) are used to adjust the mutation probability and achieve crossover and selection among individuals. The formal definition of an optimized numerical spiking neural P system of degree m (m≥1) is as follows:

[0127] Π=(σ1,…,σ m ,syn,in,out);

[0128] Where: σ1,…,σ m represent m neurons in the optimized numerical spiking neural P system,

[0129] Where any one of σ1,…,σ m can be represented as: σ i =(Vr i ,Pr i ,Vr i (0), 1≤i≤m).

[0130] Where: is a finite set of numerical variables in the \(i\)-th neuron. \(H\) represents the number of variables (in a population, \(H\) represents the population size);

[0131] is a finite set of numerical variables \(Vr\) in the \(i\)-th neuron i of the initial numerical set;

[0132] represents a finite set of production functions in the \(i\)-th neuron. and can be represented in the following form:

[0133] (the \(z\)-th production function); where \(l\) represents the number of mutant production functions in the \(i\)-th neuron; \(z\) represents the \(z\)-th mutant production function in the \(i\)-th neuron; is the \(k\)-th variable in the \(i\)-th neuron;

[0134] is a set of mutation probabilities in the \(i\)-th neuron. If is larger, then the probability that the \(z\)-th production function is executed is greater.

[0135] syn = \(\{(σ\) i , \(σ\) j ) || 1 ≤ \(i\) ≤ \(m\) AND 1 ≤ \(j\) ≤ AND \(i\neq j\}\);

[0136] in = \(\{1, 2, …, m\}\) is a set of input neurons; out = \(\{1, 2, …, m\}\) is a set of output neurons; \(σ_1, …, σ\) m are both input neurons and output neurons at the same time.

[0137] The mutant generation function can be expressed by any mutation operator. Generally speaking, the form of the mutation operator is a polynomial function. There are three mutation operators in the extended autonomous optimization numerical impulse membrane system (rand / 2, current-to-rand / 1, and gradient-to-best / 2), where rand / 2 uses two differences to enhance the global convergence ability. The form of rand / 2 is as follows:

[0138]

[0139] current-to-rand / 2 has two differences and a random scale factor, and replaces the binary crossover operation with a rotation-invariant linear recombination operator, which is more suitable for solving rotation problems. The form of current-to-rand / 2 is as follows:

[0140]

[0141] Among them, F rando is a random number between [0, 1].

[0142] gradient-to-best / 2 utilizes the gradient information of the best individuals. While introducing the momentum term to smooth the gradient direction, it combines the differential evolution strategy to increase diversity. The momentum gradient can maintain the balance between the early exploration ability and the late convergence ability. Through the random perturbation of the exponentialized differential factor, it compensates for the possible single early search direction caused by the momentum term and reduces unnecessary perturbations in the late stage, improving the convergence efficiency. The form of gradient-to-best / 2 is as follows:

[0143]

[0144] Among them, G momentum is the momentum gradient, and β is a random number between [0, 1] controlled by the change of the momentum gradient.

[0145] In the multi-mutation operator optimized numerical impulse membrane system evolving along the gradient direction, the execution of the mutation generation function is determined by the mutation probability, that is, if is larger, the probability of executing the z-th generation function is larger. At each time step, the updated numerical variable The final value of is equal to the existing value plus the other values of other connected presynaptic neurons. Compared with the numerical impulse membrane system, the realization of the generation function is uncertain. However, the mutation generation function in the extended self-optimizing numerical impulse membrane system is executed by the mutation probability (when the probabilities of multiple generation functions in the neuron σ i are not different, one of the generation functions is executed). In addition, in the extended self-optimizing numerical impulse membrane system, the generation functions executed in all neurons are parallel.

[0146] Multiple parallel extended self-optimizing numerical impulse membrane systems can generate multiple candidate populations according to the initial values and generation functions. The candidate populations are input into the probability adjustment module and the population update module after mutation. Based on the population convergence and diversity in the probability adjustment module, the mutation probability in each mutation generation function in the extended self-optimizing numerical impulse membrane system is updated. The number of individuals, the crossover between two individuals, and the mutation between individuals are completed in the population update module.

[0147] Adaptive threshold optimization of multi-threshold segmentation of brain tumor magnetic resonance images in a numerical impulse membrane system

[0148] The accuracy of the segmentation results of brain tumor magnetic resonance images is crucial for ensuring accurate diagnosis, formulating effective treatment plans, and evaluating treatment effects. To accurately segment brain tumor magnetic resonance images, the present invention proposes a multi-threshold segmentation method for brain tumor images based on an adaptive threshold optimization numerical pulse neural membrane system.

[0149] The multi-threshold segmentation method for brain tumor images based on an adaptive threshold optimization numerical pulse neural membrane system realizes optimal threshold selection by iteratively calculating candidate segmentation regions under 2-5 thresholds and combining morphological boundary expansion and gray-scale contrast maximization criteria: First, a multi-level Otsu algorithm is used to generate tumor candidate masks under different numbers of thresholds; then, a circular structure with a radius of 5 pixels is used to dynamically dilate the boundary of each mask to construct a transition region between tumors and normal tissues; the average absolute gray-scale difference between the core region and the transition region is calculated respectively, and the best number of thresholds can be determined by maximizing the difference. This method combines the Otsu between-class variance and the boundary gray-scale jump characteristics, overcomes the problem of missed segmentation of tumors with weak boundaries in traditional methods; uses a circular structure element with adaptive radius to improve the adaptability to tumors of different sizes; accelerates the solution of multi-threshold Otsu by pre-computing the gray-scale histogram, reducing the time complexity.

[0150] The multi-threshold segmentation process of brain tumor magnetic resonance images based on an adaptive threshold optimization numerical pulse neural membrane system is as follows:

[0151] Input magnetic resonance image. Input the brain tumor magnetic resonance image and convert it into 2D data.

[0152] Preprocessing. Use bilateral filtering to smooth the image by non-linearly combining the neighboring values of the image through an averaging operator while preserving the image edges.

[0153] Otsu fitness function. Use Otsu as the fitness function. Otsu is a mathematical model for threshold segmentation, and its core idea is to use different thresholds to calculate the maximum between-class variance.

[0154] Use the self-optimizing numerical pulse neural membrane system for brain tumor segmentation. Given a set of initial thresholds, calculate the Otsu value. If the Otsu value (as the number of iterations increases, a better threshold combination will calculate a larger Otsu value) meets the stopping condition, output the multi-segmentation regions. At this time, use the connectivity algorithm to select the largest segmentation region. Otherwise, use the self-optimizing numerical pulse neural membrane system to continue searching and calculating Otsu until the stopping condition is met.

[0155] Connectivity algorithm processing. The connectivity algorithm is used to filter out the region with the largest segmentation area in the brain tumor segmentation image.

[0156] Output the segmented regions. After finding the largest tumor region through the connectivity algorithm, mark this region and assign a new gray value to it, and assign a gray value of 0 to the unmarked regions. Then, the bright regions in the image are the tumor regions, and the remaining regions are the background regions.

[0157] Adaptive Threshold Optimization of Numerical Pulse Neural P Systems for Multi-modal MRI Segmentation of Brain Tumors:

[0158] The brain has a complex structure, and tumors are often closely adjacent to important functional regions. A single-modal MRI image of a brain tumor can only reflect the pathological tissue state of a specific region of the tumor, while using multi-modal images for complete segmentation of the tumor region can more accurately determine the location of the tumor, which helps neurosurgeons formulate surgical plans and minimize damage to normal brain tissue. Usually, during the clinical treatment process, it is necessary to perform complete segmentation of the brain tumor by analyzing multi-modal MRI images of the brain tumor to obtain the necrotic area, enhanced area, and edema area. T1 ce and Flair modal images can relatively clearly present the necrotic area, enhanced area, and edema area of the brain tumor. Therefore, in this invention, Flair and T1 ce modal images are selected for multi-modal segmentation. Based on the segmentation of the Flair image, the T1 ce modal image mapped by Flair is segmented for the necrotic area and the enhanced area, which can ensure that it will not be affected by other regions on the region of interest. After the segmentation is completed, the different modal images are fused according to the segmented regions to obtain a complete MRI segmentation image of the brain tumor.

[0159] The process of multi-modal MRI segmentation of brain tumors by adaptive threshold optimization of numerical pulse neural P systems is as follows:

[0160] Input the MRI image. Input the brain tumor MRI image and convert it into 2D data.

[0161] Preprocessing. Use bilateral filtering to smooth the image by non-linearly combining the neighboring values of the image through an averaging operator while preserving the image edges.

[0162] Optimized segmentation of the Flair modal image. Use Otsu to establish an objective function, and use AONSNPS to iteratively search for the optimal threshold combination for the Flair modal image until the stopping condition is met.

[0163] Connectivity processing. Perform connectivity processing on the optimized segmentation result to filter out the largest segmented area region in the brain tumor segmentation image.

[0164] Output the segmented region of the Flair modality image. After finding the largest tumor region through the connectivity algorithm, label this region and assign a new gray value. Assign a gray value of 0 to the unlabeled regions. Then, the bright region in the image is the edema region, and the remaining regions are the background regions.

[0165] Map the Flair modality image. Map the segmentation result of the Flair modality image into the T1 ce modality image so that the region to be segmented in T1 ce has the same position as the segmentation result region of the Flair modality image, in order to avoid the influence of non-tumor regions on the segmentation of the target region.

[0166] Optimize the segmentation of the T1 ce modality image. Use Otsu to establish the objective function and use AONSNPS to iteratively search for the optimal threshold combination for the T1 ce modality image until the stopping condition is met.

[0167] Output the segmented region of the T1 ce modality image. Label the segmented region and assign a new gray value. Assign a gray value of 0 to the unlabeled regions. Then, the brightest region in the image is the enhanced region, the second brightest region is the gangrenous region, and the remaining regions are the background regions.

[0168] Image fusion. Fuse the segmented region of the Flair modality image and the segmented region of the T1 ce modality image. The fusion is as follows.

[0169]

[0170] Among them, I(i,j) is the gray value of the pixel at the i-th row and j-th column after fusion, I(i,j) F is the gray value of the pixel at the i-th row and j-th column of the Flair modality segmented image, I(i,j) T is the gray value of the pixel at the i-th row and j-th column of the T1ce modality segmented image.

[0171] Image segmentation result:

[0172] In the present invention, 12 Flair modality images and T1ce modality images in the BraTs2019 dataset are selected for multi-modal segmentation and fusion. The slice sources of the 12 images are shown in Table 1 below. Figure 5 is the original Flair modality image (from top left to bottom right are MRI 1 to MRI 12), Figure 6 is the original T1ce modality image (from top left to bottom right are MRI 1 to MRI 12), Figure 7 is the standard segmentation image (from top left to bottom right are MRI 1 to MRI 12), Figure 8For the bilateral filtering of Flair modality images (from top left to bottom right are MRI 1 to MRI 12), the figure shows the Flair modality segmentation results (from top left to bottom right are MRI 1 to MRI 12). Figure 10 For the T1ce modality images mapped based on the Flair segmentation images (from top left to bottom right are MRI 1 to MRI 12). Figure 11 For the T1ce modality segmentation results (from top left to bottom right are MRI 1 to MRI 12). Figure 12 For the multi-modal segmentation results of the adaptive threshold optimization numerical pulse neural membrane system (from top left to bottom right are MRI 1 to MRI 12).

[0173] Table 1 Selection Table of Brain Tumor Images

[0174] Figure number Slice image MRI1 BraTS19_CBICA_AUW_1_98 MRI2 BraTS19_CBICA_AYC_1_71 MRI3 BraTS19_CBICA_BBG_1_79 MRI4 BraTS19_CBICA_BCL_1_56 MRI5 BraTS19_CBICA_BDK_1_70 MRI6 BraTS19_CBICA_BGE_1_109 MRI7 BraTS19_CBICA_BGG_1_112 MRI8 BraTS19_CBICA_BGN_1_76 MRI9 BraTS19_CBICA_BGO_1_75 MRI10 BraTS19_CBICA_BGX_1_77 MRI11 BraTS19_CBICA_BGX_1_88 MRI12 BraTS19_CBICA_BGX_1_96

[0175] Based on the above preprocessing, segmentation, and fusion processes, the multi-modal segmentation of brain tumor magnetic resonance images is completed. By comparing the multi-modal segmentation fusion map of brain tumors with the standard segmentation image, it can be seen that after the numerical optimization pulse neural membrane system completes the segmentation of the Flair modality of brain tumors, and the T1ce modality is segmented on the mapped area of the T1ce modality image, the enhanced area and the gangrene area can be clearly segmented.

[0176] The above-described embodiments are used to illustrate the present invention, not to limit the present invention. Therefore, changes in the example numerical values or replacement of equivalent elements should still fall within the scope of the present invention.

[0177] From the above detailed description, those of ordinary skill in the art can clearly understand that the present invention can indeed achieve the foregoing objectives and actually meet the requirements of the patent law.

[0178] Although the preferred embodiments of the present invention have been described, those skilled in the art can make additional changes and modifications to these embodiments once they know the basic creative concepts. Therefore, the appended claims are intended to be interpreted to include the preferred embodiments as well as all changes and modifications falling within the scope of the present invention. The above is only the preferred embodiment of the present invention and is not used to limit the present invention. It should be noted that any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

[0179] It should be noted that the above description of the process is only for illustration and explanation and does not limit the scope of application of this specification. Those skilled in the art can make various corrections and changes to the process under the guidance of this specification. However, these corrections and changes are still within the scope of this specification.

[0180] The basic concepts have been described above. Obviously, for those of ordinary skill in the art after reading this application, the above invention disclosure is only for example and does not constitute a limitation to this application. Although not explicitly stated here, those of ordinary skill in the art may make various modifications, improvements, and corrections to this application. Such modifications, improvements, and corrections are suggested in this application, so such modifications, improvements, and corrections still fall within the spirit and scope of the exemplary embodiments of this application.

[0181] Meanwhile, this application uses specific terms to describe the embodiments of this application. For example, "one embodiment", "an embodiment", and / or "some embodiments" mean a certain feature, structure, or characteristic related to at least one embodiment of this application. Therefore, it should be emphasized and noted that the "an embodiment" or "one embodiment" or "an alternative embodiment" mentioned two or more times at different positions in this specification is not necessarily referring to the same embodiment. In addition, certain features, structures, or characteristics in one or more embodiments of this application can be appropriately combined.

[0182] In addition, those of ordinary skill in the art can understand that various aspects of this application can be illustrated and described by several patentable types or situations, including any new and useful process, machine, product, or composition of matter, or any new and useful improvement thereof. Therefore, various aspects of this application can be implemented entirely by hardware, entirely by software (including firmware, resident software, microcode, etc.), or by a combination of hardware and software. The above hardware or software can all be referred to as "unit", "module", or "system". In addition, various aspects of this application can take the form of a computer program product embodied in one or more computer-readable media, in which computer-readable program code is included.

[0183] The computer program code required for the operations of various parts of this application can be written in any one or more programming languages, including object-oriented programming languages such as Java, Scala, Smalltalk, Eiffel, JADE, Emerald, C++, C#, VB.NET, Python, etc., conventional procedural programming languages such as C programming language, VisualBasic, Fortran2103, Perl, COBOL2102, PHP, ABAP, dynamic programming languages such as Python, Ruby and Groovy, or other programming languages. This program code can run entirely on the user's computer, or run on the user's computer as an independent software package, or run partially on the user's computer and partially on a remote computer, or run entirely on a remote computer or server. In the latter case, the remote computer can be connected to the user's computer through any network form, such as a local area network (LAN) or a wide area network (WAN), or connected to an external computer (for example, through the Internet), or in a cloud computing environment, or used as a service such as software as a service (SaaS).

[0184] In addition, unless clearly stated in the claims, the order of the processing elements and sequences, the use of numerical letters, or the use of other names in this application are not used to limit the order of the processes and methods of this application. Although some currently considered useful embodiments of the invention are discussed through various examples in the above disclosure, it should be understood that such details only serve the purpose of illustration, and the appended claims are not limited to the disclosed embodiments. On the contrary, the claims are intended to cover all modifications and equivalent combinations that conform to the essence and scope of the embodiments of this application. For example, although the implementation of the various components described above can be embodied in a hardware device, it can also be implemented as a pure software solution, for example, installed on an existing server or mobile device.

[0185] Similarly, it should be noted that, in order to simplify the description of this application disclosure and thus help the understanding of one or more embodiments of the invention, in the previous description of the embodiments of this application, sometimes multiple features are merged into one embodiment, drawing or description thereof. However, this method of this application should not be construed as reflecting the intention that the claimed subject matter requires more features than those clearly recited in each claim. On the contrary, the subject matter of the invention should have fewer features than the above single embodiment.

Claims

1. A brain tumor image segmentation method based on adaptive threshold, characterized in that: include: S1. Collect brain tumor MRI images of different case backgrounds and analyze the characteristics of brain tumor MRI image segmentation problems; S2. Based on the characteristics of the segmentation problem, analyze the optimized segmentation model of brain tumor MRI images and select the objective function that reflects the actual segmentation situation; S3. Construct a multi-variation operator optimized numerical pulse neural membrane system evolving along the gradient direction, wherein the framework of the multi-variation operator optimized numerical pulse neural membrane system evolving along the gradient direction is composed of a plurality of parallel extended autonomous optimized numerical pulse membrane system frameworks and two guide modules; each extended autonomous optimized numerical pulse membrane system is used to generate a candidate population; the two guide modules are respectively a probability adjustment module and a population update module, both of which are used to adjust the mutation probability to achieve crossover and selection between individuals; S4. Based on the objective function selected in S2 and the multi-variation operator optimized numerical pulse neural membrane system evolved along the gradient direction constructed in S3, a multi-threshold segmentation model of brain tumor MRI images of the adaptive threshold optimized numerical pulse neural membrane system is constructed; S5. Based on the objective function selected in S2 and the multi-variation operator optimized numerical pulse neural membrane system evolved along the gradient direction constructed in S3, a brain tumor MRI multimodal image segmentation model of the adaptive threshold optimized numerical pulse neural membrane system was constructed; S6. Perform image segmentation using the brain tumor MRI image multi-threshold segmentation model and the brain tumor MRI multimodal image segmentation model.

2. The brain tumor image segmentation method based on adaptive threshold according to claim 1, characterized in that: The process of adaptive threshold optimization of the multi-threshold segmentation model of brain tumor MRI images of the numerical pulsed neurolemma system is as follows: Input brain tumor MRI images and convert them into 2D data; Bilateral filtering is used to smooth the 2D brain tumor MRI images by using the nonlinear combination average operator of image neighboring values ​​while preserving the image edges. Based on the smoothed image, the Otsu threshold segmentation model is used as the objective function, and the brain tumor segmentation is performed by optimizing the numerical pulse neural membrane system through the multi-variation operator evolving along the gradient direction, and the Otsu threshold is calculated based on the initial threshold. If the Otsu threshold meets the stopping condition, multiple segmented regions are output, and the largest segmented region is selected using the connectivity algorithm; Otherwise, the multi-variation operator optimized numerical pulse neural membrane system that evolves along the gradient direction is used to continue searching and calculating the Otsu threshold until the stopping condition is met; after selecting the largest segmented area, the area is marked and assigned a new grayscale value, and the grayscale value of the unmarked area is assigned 0. The bright area in the image is the tumor area, and the rest of the area is the background area.

3. The brain tumor image segmentation method based on adaptive threshold according to claim 1, characterized in that: The process of adaptive threshold optimization of the numerical pulse neurolemma system brain tumor MRI multimodal image segmentation model is as follows: By analyzing brain tumor MRI images of different modalities, the brain tumor is completely segmented to obtain Flair modality images and T1 ce modality images; Convert Flair modality images and T1 ce modality images into 2D data; The Flair modality image and T1 ce modality image of 2D data were smoothed by using bilateral filtering through the nonlinear combination average operator of image neighboring values ​​while preserving the image edge. The Otsu threshold segmentation model is used as the objective function, and AONSNPS is used to iteratively search for the optimal threshold combination on the smoothed Flair modality image until the stop condition is met, and connectivity processing is performed to filter out the largest segmentation area in the brain tumor segmentation image; After finding the area with the largest segmentation area, mark the area and assign a new gray value. The gray value of the unmarked area is assigned 0. The bright area in the image is the edema area, and the rest is the background area. Get the Flair modality image segmentation result; The segmentation result of the Flair modality image is mapped to the smoothed T1 ce modality image, so that the T1 ce area to be segmented is in the same position as the area of ​​the Flair modality image segmentation result, so as to avoid the influence of the non-tumor area on the segmentation of the target area; The Otsu threshold segmentation model is used as the objective function, and the AONSNPS is used to iteratively search for the optimal threshold combination for the mapped T1 ce modality image until the stop condition is met. The segmented area is marked and assigned a new gray value, and the gray value of the unmarked area is assigned to 0. The brightest area in the image is the enhanced area, the second brightest area is the gangrene area, and the rest of the area is the background area. The segmentation region of the Flair modality image is fused with the segmentation region of the T1 ce modality image, and the fusion is as follows: Among them, I(i,j) is the gray value of the pixel in the i-th row and j-th column after fusion, I(i,j) F is the gray value of the pixel in the i-th row and j-th column of the Flair modality segmentation image, I(i,j) T is the gray value of the pixel in the i-th row and j-th column of the T1ce modality segmentation image.

4. The brain tumor image segmentation method based on adaptive threshold according to claim 3, characterized in that: The formalization Π of an extended autonomous optimization numerical pulse membrane system with degree m, m greater than or equal to 1, is as follows: ∏=(σ1,…,σ m ,syn,in,out); Where: σ1,…,σ m represents the m neurons in the extended autonomous optimization numerical pulse membrane system, where σ1,…,σ m The form of any neuron is expressed as: i =(Vr i ,Pr i ,Vr i (0),1≤i≤m); where is a finite set of numerical variables in the i-th neuron, H represents the number of variables, and in a population, H represents the population size; is a finite set of numerical variables Vr in the ith neuron i The initial value set of ; represents a finite set of generating functions in the i-th neuron, and The form is as follows: 1≤z≤l,1≤k≤H; where l represents the number of mutation generating functions in the i-th neuron; z represents the z-th mutation generating function in the i-th neuron; is the kth variable in the ith neuron; is a set of mutation probabilities in the ith neuron, if The larger it is, the greater the probability that the zth generation function will be executed; syn={(σ i ,σ j )||1≤i≤mAND1≤j≤AND1≠j}; in={1,2,…,m} is a set of input neurons; out={1,2,…,m} is a set of output neurons; σ1,…,σ m It is both an input neuron and an output neuron.

5. The brain tumor image segmentation method based on adaptive threshold according to claim 4, characterized in that: There are three mutation operators in the extended autonomous optimization numerical pulse membrane system: rand / 2, current-to-rand / 1 and gradient-to-best / 2. The form of rand / 2 is as follows: The form of current-to-rand / 2 is as follows: Among them, F random is a random number between [0,1]; The form of gradient-to-best / 2 is as follows: Among them G momentum is the momentum gradient, and β is a random number between [0,1] controlled by the change of momentum gradient.

6. The brain tumor image segmentation method based on adaptive threshold according to claim 5, characterized in that: At each time step, the numerical variables are updated The final value of is equal to the existing value plus the additional values ​​of other connected presynaptic neurons.

7. The brain tumor image segmentation method based on adaptive threshold according to claim 1, characterized in that: Brain tumor MRI images include T1 images, T1 ce images, T2 images and Flair images.

8. The brain tumor image segmentation method based on adaptive threshold according to claim 1, characterized in that: For the brain tumor MRI image segmentation problem with a size of 240×240, global distribution noise and blurred tumor boundaries due to brightness differences, the Otsu threshold segmentation model is used as the objective function of the brain tumor MRI image optimization segmentation model.

9. The brain tumor image segmentation method based on adaptive threshold according to claim 1, characterized in that: The optimized segmentation models of brain tumor MRI images include the maximum entropy model, the minimum error model and the Otsu threshold segmentation model.

Images