Brain magnetic resonance image segmentation method, device, equipment and medium
By combining the subspace model and deep convolutional neural network, and utilizing the strength and structural prior information of brain tissue, the problems of insufficient accuracy and stability of brain segmentation in existing technologies are solved, and high-precision segmentation is achieved under multi-center and large-scale data.
Patent Information
- Application Number
- CN202410293887.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-03-14
- Publication Date
- 2025-09-16
AI Technical Summary
Existing brain segmentation methods have shortcomings in terms of accurate segmentation and stability of multi-center and large-scale data. In particular, deep learning networks have strong capabilities in capturing local features but insufficient capabilities in capturing long-distance global features. Prior information is difficult to capture the heterogeneity of the spatial distribution of group brains and ignores intensity information, resulting in limited segmentation accuracy and stability.
A subspace model is used to extract spatial prior information of brain tissue, combined with Bayesian segmentation and position-related networks, and a fusion network is used to achieve accurate segmentation of brain magnetic resonance images. Intensity prior and structure prior information are used to guide the network to learn the probability distribution of brain tissue, and the attention mechanism is combined to improve the segmentation performance.
The accuracy and stability of brain tissue segmentation are improved, especially in areas where intensity information is difficult to segment, and the robustness to interference signals and the generalization performance of the network are enhanced.
Smart Images
Figure CN120655656A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of digital image technology, and in particular to a brain magnetic resonance image segmentation method, device, equipment and medium based on a subspace model and a deep convolutional neural network. Background Art
[0002] Brain segmentation is a crucial step in research on normal brain development, aging, and brain diseases, including lesion detection and surgical planning. The task of brain segmentation is to classify each voxel in a brain image into distinct anatomically or functionally meaningful categories, such as gray matter, white matter, and cerebrospinal fluid, based on properties such as texture, intensity, and shape.
[0003] Existing brain segmentation software, including Statistical Parametric Mapping (SPM) and Freesurfer, achieves voxel-by-voxel segmentation by estimating the intensity likelihood function of brain tissue and incorporating priors based on probabilistic brain maps. These methods typically achieve relatively stable segmentation results, but due to their simplistic model assumptions, they struggle to capture the complex spatial and intensity distributions of brain tissue, resulting in limited accuracy and difficulty in applications requiring precise segmentation.
[0004] Deep learning-based approaches have significantly improved segmentation performance by leveraging deep convolutional neural networks (CNNs) to learn the joint spatial-intensity probability distribution of brain images, replacing hand-crafted models. However, due to the inherent properties of convolutional layers, CNNs are highly capable of capturing local features but struggle to capture long-range global features. Common solutions to address this issue include U-Nets, which leverage multi-scale information and skip connections; attention networks, which utilize various attention mechanisms, such as spatial attention, scale attention, and channel attention, to focus on learning important features; and neural networks that combine Transformer modules with CNNs, such as TransUNet and UTNet. These networks significantly improve the segmentation performance of existing networks. However, the introduction of additional modules requires more anatomical data labeled by neuroscientists for network training, which is often limited. Consequently, existing methods often perform poorly when the test data is acquired using different acquisition sequences or equipment than the training data, limiting the model's generalization performance and robustness to interference from noise, artifacts, and other interfering signals. Consequently, existing networks often struggle to achieve stable and accurate segmentation for multi-center, large-scale data.
[0005] To address the issue of network generalization performance, existing technologies typically introduce spatial constraints by introducing prior information into the network to improve its segmentation performance for images of varying contrast and interference. Commonly used prior information includes multi-atlas priors and probabilistic atlas priors, but the former requires multiple manually labeled images as support and has difficulty capturing the heterogeneity of brain spatial distribution across a population. While the latter can model the spatial distribution of brain tissue in large populations, its construction process involves averaging a series of brain segmentation results, ignoring intensity information and high-order statistical information, making it difficult to complete segmentation in more challenging areas. Summary of the Invention
[0006] The purpose of the present invention is to provide a brain magnetic resonance image segmentation method, device, equipment and medium based on a subspace model and a deep convolutional neural network, and to achieve accurate segmentation of heterogeneous brain images by integrating a subspace-based spatial-intensity prior model and a segmentation network model based on an attention mechanism.
[0007] The purpose of the present invention can be achieved by the following technical solutions:
[0008] A brain magnetic resonance imaging segmentation method based on a subspace model and a deep convolutional neural network comprises the following steps:
[0009] S1, extracting brain tissue spatial prior information from public brain magnetic resonance imaging datasets based on subspace model;
[0010] S2, iteratively segmenting the brain MRI image based on the spatial prior information of brain tissue;
[0011] S3, constructing a position-dependent network, based on the subspace prior obtained in step S1 and the segmentation result obtained in step S2, using them as prior information to guide the position-dependent network to learn the probability distribution of brain tissue;
[0012] S4, constructing a fusion network, taking the subspace prior obtained in step S1, the segmentation result obtained in step S2, and the segmentation result of the position-related network obtained in step S3 as input, and fusing them to output the final segmentation result.
[0013] The brain tissue spatial prior information includes intensity prior information and structure prior information, wherein the intensity prior information is represented by an intensity likelihood function.
[0014] Said S1 comprises the following steps:
[0015] S11, obtain a public dataset of brain magnetic resonance images, register them to the standard space, and perform initial segmentation on the images:
[0016] S12, performing singular value decomposition on the segmentation results of each tissue, and determining the rank of the subspace through the singular value decay curve to construct the subspace;
[0017] S13, subspace projection: Project the initial segmentation result onto the constructed subspace using the least squares method to obtain the corresponding spatial prior information.
[0018] The initial segmentation adopts a maximum a posteriori-expectation maximization model-based approach.
[0019] The S2 comprises the following steps:
[0020] S21, image pre-segmentation of the brain magnetic resonance image to be segmented based on the Bayesian segmentation method;
[0021] S22, projecting the pre-segmentation result of step S21 onto the subspace obtained in step S1 to obtain the corresponding subspace prior;
[0022] S23, the intensity likelihood function of each spatial position learned in the public dataset is combined with the subspace prior obtained in step S22 to obtain the corresponding posterior segmentation by maximum a posteriori estimation;
[0023] S24, based on the posterior segmentation obtained in step S23, the intensity likelihood of the image to be segmented is calculated and projected onto the subspace to update the subspace prior, and step S23 is executed again to obtain the posterior segmentation result, thereby completing the segmentation.
[0024] The position-dependent network includes two inputs, one is the original brain magnetic resonance image to be segmented, and the other is an uncertainty map extracted from the subspace tissue prior, wherein the uncertainty map is obtained by calculating the entropy of each voxel based on the subspace prior result.
[0025] The position-dependent network includes two parallel U-Net structures. Each decoding module and encoding module of the two U-Nets are combined through an attention gate to enable the network to pay attention to areas with higher uncertainty and simultaneously learn the intensity distribution of brain tissue at different spatial locations.
[0026] A brain magnetic resonance image segmentation device based on a subspace model and a deep convolutional neural network, comprising:
[0027] Subspace prior construction module: extracts brain tissue spatial prior information from public brain magnetic resonance imaging datasets based on the subspace model;
[0028] Subspace prior-assisted segmentation module: Iteratively segment the brain MRI image based on the spatial prior information of brain tissue;
[0029] Position-dependent segmentation module: This module constructs a position-dependent network. Based on the subspace prior obtained by the subspace prior construction module and the segmentation results obtained by the subspace prior-assisted segmentation module, these subspace priors are used as prior information to guide the position-dependent network to learn the probability distribution of brain tissue.
[0030] Fusion module: Constructs a fusion network, takes the subspace prior obtained by the subspace prior construction module, the segmentation result obtained by the subspace prior assisted segmentation module, and the segmentation result of the position-related network obtained by the position-related segmentation module as input, and fuses them to output the final segmentation result.
[0031] An electronic device includes a memory and a processor, wherein the memory stores a computer program, and the processor implements the method described above when executing the program.
[0032] A computer-readable storage medium stores a computer program, which implements the method described above when executed by a processor.
[0033] Compared with the prior art, the present invention has the following beneficial effects:
[0034] 1. The present invention uses a subspace model to capture high-order statistical information of the brain tissue structure of the human population. Compared with traditional brain probability maps, it introduces additional information in the high-order subspace basis, making the prior information boundary clearer and the structural features richer. At the same time, it allows it to adaptively change according to the characteristics of the test data, making it closer to the actual structure of the test data, which facilitates subsequent segmentation.
[0035] 2. Unlike traditional deep learning-based methods, the present invention introduces subspace priors and Bayesian segmentation results into the network. On the one hand, it serves as prior information to guide the network to learn the probability distribution of brain tissue, greatly reducing the complexity of the required learning features. It also relies on the attention network to learn the probability distribution of brain tissue by combining the complementary information in the subspace priors and Bayesian segmentation results. Compared with existing deep learning methods, the present invention has higher accuracy in brain segmentation results for different populations, especially in areas where intensity information is difficult to use for accurate segmentation. On the other hand, since these prior information are relatively stable, the stability of the network to various interference signals is greatly improved, making it more stable than neural networks that do not use priors. It can also be used as a constraint to improve the robustness of the network to interference signals. BRIEF DESCRIPTION OF THE DRAWINGS
[0036] Figure 1 is a flow chart of the method of the present invention;
[0037] Figure 2 A diagram of a location-related network structure in one embodiment;
[0038] Figure 3 This is a diagram showing the segmentation results of a multi-age group dataset in one embodiment;
[0039] Figure 4 4 is a segmentation result diagram of a simulated image with an interference signal in an embodiment. DETAILED DESCRIPTION
[0040] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.
[0041] This embodiment provides a brain magnetic resonance image segmentation method based on a subspace model and a deep convolutional neural network, the purpose of which is to segment a brain structure magnetic resonance image X into a tissue label map C. Figure 1 As shown, the method includes the following steps:
[0042] S1, extracting brain tissue spatial prior information from brain magnetic resonance imaging public datasets based on subspace model,
[0043] In this embodiment, the brain tissue spatial prior information includes intensity prior information and structure prior information, wherein the intensity prior information is represented by an intensity likelihood function.
[0044] Specifically, S1 includes the following steps:
[0045] S11, obtain a large number of publicly available brain magnetic resonance imaging datasets, align them to the standard space, and use the maximum a posteriori-expectation maximization model-based method to perform initial image segmentation.
[0046] The maximum a posteriori-expectation maximization model-based method is a common method used by those skilled in the art and will not be described in detail in this embodiment.
[0047] S12, performing singular value decomposition on the segmentation result of each tissue, and determining the rank of the subspace through the singular value decay curve to construct the subspace.
[0048] S13, subspace projection: Project the initial segmentation result onto the constructed subspace using the least squares method to obtain the corresponding spatial prior information.
[0049] S2, based on the spatial prior information of brain tissue, iteratively segment the brain magnetic resonance image to be segmented.
[0050] Specifically, S2 includes the following steps:
[0051] S21, performing image pre-segmentation on the brain magnetic resonance image to be segmented based on the Bayesian segmentation method.
[0052] S22, projecting the pre-segmentation result of step S21 onto the subspace obtained in step S1 to obtain a corresponding subspace prior.
[0053] S23, the intensity likelihood function of each spatial position learned in the public dataset is combined with the subspace prior obtained in step S22 to obtain the corresponding posterior segmentation by maximum a posteriori estimation;
[0054] S24, based on the posterior segmentation obtained in step S23, the intensity likelihood of the image to be segmented is calculated and projected onto the subspace to update the subspace prior, and step S23 is executed again to obtain the posterior segmentation result, thereby completing the segmentation.
[0055] S3, constructing a position-dependent network, based on the subspace prior obtained in step S1 and the segmentation result obtained in step S2, using them as prior information to guide the position-dependent network to learn the probability distribution of brain tissue,
[0056] The position-dependent network includes two inputs: one is the original brain magnetic resonance image to be segmented, and the other is the uncertainty map extracted from the subspace tissue prior. The uncertainty map is obtained by calculating the entropy of each voxel based on the subspace prior result.
[0057] like Figure 2 As shown, the position-dependent network consists of two parallel U-Net structures. Each decoding module and encoding module of the two U-Nets are combined through an attention gate to focus on areas with high uncertainty and simultaneously learn the intensity distribution of brain tissue at different spatial locations. The two inputs are fed into the two U-Net structures respectively, and the two outputs are convolved to obtain the segmentation results.
[0058] S4, constructing a fusion network, taking the subspace prior obtained in step S1, the segmentation result obtained in step S2, and the segmentation result of the position-related network obtained in step S3 as input, and fusing them to output the final segmentation result.
[0059] The fusion network effectively combines the information of tissue subspace model, Bayesian classifier and location correlation network to further improve the segmentation performance.
[0060] Figure 3 The segmentation results of multi-age data sets are shown. By comparing this application with other existing technologies, it can be found that in the three data sets of infants, adults, and the elderly, the segmentation results of other methods have relatively obvious errors, while the segmentation results of this technology are relatively accurate, verifying the accuracy improvement of this technology compared to the most advanced deep learning algorithms, as well as its application potential in people of different age groups.
[0061] Figure 4The segmentation results of an image with simulated B1 field inhomogeneity are shown. It can be seen that the simulated interference signal causes obvious errors in the segmentation results of other methods, but the results of this technique are still relatively accurate, demonstrating the stability of the segmentation algorithm of the present invention.
[0062] The above is an introduction to a method embodiment. The following further illustrates the solution of the present invention through an apparatus embodiment.
[0063] This embodiment further provides a brain magnetic resonance image segmentation device based on a subspace model and a deep convolutional neural network, comprising:
[0064] Subspace prior construction module: extracts brain tissue spatial prior information from public brain magnetic resonance imaging datasets based on the subspace model;
[0065] Subspace prior-assisted segmentation module: Iteratively segment the brain MRI image based on the spatial prior information of brain tissue;
[0066] Position-dependent segmentation module: This module constructs a position-dependent network. Based on the subspace prior obtained by the subspace prior construction module and the segmentation results obtained by the subspace prior-assisted segmentation module, these subspace priors are used as prior information to guide the position-dependent network to learn the probability distribution of brain tissue.
[0067] Fusion module: Constructs a fusion network, takes the subspace prior obtained by the subspace prior construction module, the segmentation result obtained by the subspace prior assisted segmentation module, and the segmentation result of the position-related network obtained by the position-related segmentation module as input, and fuses them to output the final segmentation result.
[0068] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the described module can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.
[0069] In one embodiment, the electronic device includes a computing unit that can perform various appropriate actions and processes based on a computer program stored in a read-only memory (ROM) or a computer program loaded from a storage unit into a random access memory (RAM). Various programs and data required for device operation can also be stored in the RAM. The computing unit, ROM, and RAM are connected to each other via a bus. An input / output (I / O) interface is also connected to the bus.
[0070] Many components in an electronic device are connected to the I / O interface, including: input units, such as a keyboard and mouse; output units, such as various types of displays and speakers; storage units, such as magnetic disks and optical disks; and communication units, such as network cards, modems, and wireless communication transceivers. The communication unit allows the device to exchange information / data with other devices via computer networks such as the Internet and / or various telecommunication networks.
[0071] The computing unit can be a variety of general and / or special processing components with processing and computing capabilities. Some examples of computing units include, but are not limited to, central processing units (CPUs), graphics processing units (GPUs), various dedicated artificial intelligence (AI) computing chips, various computing units that run machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The computing unit performs the various methods and processes described above, such as image processing methods and / or model training methods. For example, in some embodiments, the image processing methods and / or model training methods can be implemented as computer software programs that are tangibly contained in machine-readable media, such as storage units. In some embodiments, part or all of the computer program can be loaded and / or installed on the device via ROM and / or a communication unit. When the computer program is loaded into RAM and executed by the computing unit, one or more steps of the image processing methods and / or model training methods described above can be performed. Alternatively, in other embodiments, the computing unit can be configured to perform image processing methods and / or model training methods in any other appropriate manner (e.g., by means of firmware).
[0072] The program code for implementing the method of the present disclosure can be written in any combination of one or more programming languages. These program codes can be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device so that when the program code is executed by the processor or controller, the functions / operations specified in the flow chart and / or block diagram are implemented. The program code can be executed entirely on the machine, partially on the machine, as a stand-alone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.
[0073] In the context of the present disclosure, a machine-readable medium can be a tangible medium that can contain or store a program for use by or in conjunction with an instruction execution system, device or equipment. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium can include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.
[0074] It should be understood that the various forms of the processes shown above can be used to reorder, add, or delete steps. For example, the steps described in this disclosure can be performed in parallel, sequentially, or in a different order, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved. This is not limited herein.
[0075] The above describes in detail the preferred embodiments of the present invention. It should be understood that those skilled in the art can make numerous modifications and variations based on the concepts of the present invention without inventive effort. Therefore, any technical solutions that can be derived by those skilled in the art through logical analysis, reasoning, or limited experimentation based on the concepts of the present invention and the prior art should be within the scope of protection defined by the claims.
Claims
1. A brain magnetic resonance imaging segmentation method based on a subspace model and a deep convolutional neural network, characterized in that: The following steps are involved: S1, extracting brain tissue spatial prior information from public brain magnetic resonance imaging datasets based on subspace model; S2, iteratively segmenting the brain MRI image based on the spatial prior information of brain tissue; S3, constructing a position-dependent network, based on the subspace prior obtained in step S1 and the segmentation result obtained in step S2, using them as prior information to guide the position-dependent network to learn the probability distribution of brain tissue; S4, constructing a fusion network, taking the subspace prior obtained in step S1, the segmentation result obtained in step S2, and the segmentation result of the position-related network obtained in step S3 as input, and fusing them to output the final segmentation result.
2. The brain magnetic resonance image segmentation method based on subspace model and deep convolutional neural network according to claim 1, characterized in that: The brain tissue spatial prior information includes intensity prior information and structure prior information, wherein the intensity prior information is represented by an intensity likelihood function.
3. The brain magnetic resonance image segmentation method based on subspace model and deep convolutional neural network according to claim 1, characterized in that: Said S1 comprises the following steps: S11, obtain a public dataset of brain magnetic resonance images, register them to a standard space, and perform initial segmentation on the images; S12, performing singular value decomposition on the segmentation results of each tissue, and determining the rank of the subspace through the singular value decay curve to construct the subspace; S13, subspace projection: Project the initial segmentation result onto the constructed subspace using the least squares method to obtain the corresponding spatial prior information.
4. The brain magnetic resonance image segmentation method based on subspace model and deep convolutional neural network according to claim 2, characterized in that: The initial segmentation adopts a maximum a posteriori-expectation maximization model-based approach.
5. The brain magnetic resonance image segmentation method based on subspace model and deep convolutional neural network according to claim 1, characterized in that: The S2 comprises the following steps: S21, image pre-segmentation of the brain magnetic resonance image to be segmented based on the Bayesian segmentation method; S22, projecting the pre-segmentation result of step S21 onto the subspace obtained in step S1 to obtain the corresponding subspace prior; S23, the intensity likelihood function of each spatial position learned in the public dataset is combined with the subspace prior obtained in step S22 to obtain the corresponding posterior segmentation by maximum a posteriori estimation; S24, based on the posterior segmentation obtained in step S23, the intensity likelihood of the image to be segmented is calculated and projected onto the subspace to update the subspace prior, and step S23 is executed again to obtain the posterior segmentation result, thereby completing the segmentation.
6. The brain magnetic resonance image segmentation method based on subspace model and deep convolutional neural network according to claim 1, characterized in that: The position-dependent network includes two inputs, one is the original brain magnetic resonance image to be segmented, and the other is an uncertainty map extracted from the subspace tissue prior, wherein the uncertainty map is obtained by calculating the entropy of each voxel based on the subspace prior result.
7. The brain magnetic resonance image segmentation method based on subspace model and deep convolutional neural network according to claim 1, characterized in that: The position-dependent network includes two parallel U-Net structures. Each decoding module and encoding module of the two U-Nets are combined through an attention gate to enable the network to pay attention to areas with higher uncertainty and simultaneously learn the intensity distribution of brain tissue at different spatial locations.
8. A brain magnetic resonance image segmentation device based on a subspace model and a deep convolutional neural network, characterized in that: include: Subspace prior construction module: extracts brain tissue spatial prior information from public brain magnetic resonance imaging datasets based on the subspace model; Subspace prior-assisted segmentation module: Iteratively segment the brain MRI image based on the spatial prior information of brain tissue; Position-dependent segmentation module: This module constructs a position-dependent network. Based on the subspace prior obtained by the subspace prior construction module and the segmentation results obtained by the subspace prior-assisted segmentation module, these subspace priors are used as prior information to guide the position-dependent network to learn the probability distribution of brain tissue. Fusion module: Constructs a fusion network, takes the subspace prior obtained by the subspace prior construction module, the segmentation result obtained by the subspace prior assisted segmentation module, and the segmentation result of the position-related network obtained by the position-related segmentation module as input, and fuses them to output the final segmentation result.
9. An electronic device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the program, the method according to any one of claims 1 to 7 is implemented.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.