Organ segmentation method, device, computer equipment and storage medium

Abstract

The present application relates to an organ segmentation method, device, computer equipment and storage medium. Based on the mutual information between a template image and an image to be processed determined by any modality, the template image and the image to be processed are registered to obtain a registration result, and according to the registration result, the target organ mask in the template image is processed to obtain the target organ region in the image to be processed, and then the target organ region in the image to be processed is corrected for non-uniform fields, and the tissue in the target organ is segmented to obtain a three-dimensional mask image of the target organ in the image to be processed. The method can achieve segmentation of the image to be processed for any modality input in various scenarios, thereby improving the versatility of magnetic resonance images of various modal structures.

Description

Technical Field

[0001] The present application relates to the field of image processing technology, and in particular to an organ segmentation method, apparatus, computer equipment and storage medium. Background Art

[0002] Computer-aided diagnosis and treatment is an important development direction of modern medicine. Segmentation and three-dimensional reconstruction of organs and tissues to be analyzed in medical images are both important and difficult technologies in computer-aided treatment.

[0003] Taking the segmentation and three-dimensional reconstruction of the cerebral cortex and sulci as an example, with the assistance of three-dimensional display technologies such as VR (virtual reality), the real situation of the patient's cortical surface can be presented most intuitively, which is helpful for diagnosing diseases and planning surgical operations based on medical images. Magnetic resonance images are common medical images. Magnetic resonance corresponds to different modalities under different sequences. The grayscale values ​​of various tissues in magnetic resonance images obtained under different modalities are different. Common structural magnetic resonance images (Structural magnetic resonance imaging, SMRI) include T1w and T2w. In these two images, the cerebral cortex has obvious structural characteristics.

[0004] However, the prior art cannot adapt to different application scenarios and data inputs under multiple modalities when segmenting and reconstructing structural magnetic resonance images, and lacks versatility. Summary of the invention

[0005] Based on this, it is necessary to provide an organ segmentation method, device, computer equipment and storage medium that can adapt to different application scenarios and data inputs under multiple modalities and improve versatility in response to the above technical problems.

[0006] In a first aspect, an embodiment of the present application provides a method for segmenting an organ, the method comprising:

[0007] Based on the mutual information between the template image and the image to be processed, the template image and the image to be processed are registered to obtain a registration result; the template image is determined according to any modality;

[0008] According to the registration result, the target organ mask in the template image is processed to obtain the target organ region in the image to be processed;

[0009] The target organ region of the image to be processed is subjected to non-uniform field correction, and the tissue in the target organ is segmented to obtain a three-dimensional mask image of the target organ in the image to be processed.

[0010] In one embodiment, the above-mentioned registering the template image and the image to be processed to obtain the registration result includes:

[0011] Performing rigid registration on the template image and the image to be processed to obtain an affine matrix, and performing elastic registration on the template image and the image to be processed to obtain an elastic deformation field;

[0012] According to the registration result, the target organ mask in the template image is processed to obtain the target organ region in the image to be processed, including:

[0013] An affine matrix and an elastic deformation field are applied to the target organ mask in the template image to obtain the target organ region in the image to be processed.

[0014] In one embodiment, the above-mentioned non-uniform field correction is performed on the target organ region of the image to be processed, and the tissue in the target organ is segmented to obtain a three-dimensional mask image of the target organ in the image to be processed, including:

[0015] Based on the grayscale difference between tissues, the target organ region of the image to be processed is corrected for non-uniform field, and the tissue in the target organ is roughly segmented to obtain a primary segmentation image of the target organ in the image to be processed;

[0016] The tissue in the initial segmentation image is finely segmented to obtain a three-dimensional mask image of the target organ in the image to be processed.

[0017] In one embodiment, the above-mentioned non-uniform field correction is performed on the target organ region of the image to be processed based on the grayscale difference between tissues, and the tissue in the target organ is roughly segmented to obtain the initial segmentation image of the target organ in the image to be processed, including:

[0018] Downsampling the target organ region of the image to be processed by a preset multiple to obtain the downsampled target organ region of the image to be processed;

[0019] Construct an objective function including a category grayscale vector, a membership matrix and an inhomogeneous field, and obtain the category grayscale vector, the membership matrix and the inhomogeneous field corresponding to the minimum value of the objective function; the category grayscale vector represents the central grayscale value of each category of tissue, and the membership degree represents the probability of each pixel point in the corresponding category of tissue;

[0020] The target organ region of the downsampled image to be processed is corrected for the uneven field according to the uneven field, and the tissue in the target organ region is roughly segmented according to the membership matrix to obtain a primary segmentation image of the target organ in the downsampled image to be processed.

[0021] In one embodiment, the above construction includes the category grayscale vector, the membership matrix and the objective function of the inhomogeneous field, including:

[0022] Obtaining a model expression of a target organ's real image and a non-uniform field model expression; the model expression of the target organ's real image includes a category grayscale vector and a membership matrix;

[0023] An image model expression including a multiplicative inhomogeneous field is established according to a model expression of a real image of a target organ and a model expression of an inhomogeneous field;

[0024] The objective function is constructed based on the image model expression including the multiplicative inhomogeneous field.

[0025] In one embodiment, before performing fine segmentation on the initial segmented image of the target organ in the image to be processed, the method further comprises:

[0026] According to the category grayscale vector, determining the number of pixels of the tissue with the smallest grayscale value and the number of pixels of the tissue with the largest grayscale value in the target organ of the image to be processed;

[0027] comparing the number of pixels of the tissue with the smallest gray value with the number of pixels of the designated tissue in the target organ, and comparing the number of pixels of the tissue with the largest gray value with the number of pixels of the designated tissue in the target organ;

[0028] If the number of pixels of the tissue with the smallest grayscale value belongs to the range of the number of pixels of the specified tissue in the target organ, it is determined that the tissue in the initial segmentation image needs to be finely segmented into the first value category number; if the number of pixels of the tissue with the largest grayscale value belongs to the range of the number of pixels of the specified tissue in the target organ, it is determined that the tissue in the initial segmentation image needs to be finely segmented into the second value category number.

[0029] In one embodiment, the fine segmentation of the tissue in the initial segmentation image comprises:

[0030] Upsampling the initial segmented image by a preset multiple to obtain an upsampled initial segmented image;

[0031] If the tissue in the initially segmented image needs to be finely segmented into a first value of the number of categories, the center of each category in the category gray vector is used as the initial category center for fine segmentation, and the membership matrix after upsampling the membership matrix by a preset multiple is used as the initial membership matrix for fine segmentation, and iterative clustering is performed on the histogram of the upsampled initially segmented image to perform fine segmentation on the tissue in the upsampled initially segmented image;

[0032] If the tissue in the initial segmentation image needs to be finely segmented into a second value number of categories, then the second value number of grayscale values ​​are taken at an average interval from the category grayscale vector as the initial category center for fine segmentation, and the second value number of memberships are taken at an average interval from the range of the membership matrix as the initial membership matrix for fine segmentation. Iterative clustering is performed on the histogram of the upsampled initial segmentation image to perform fine segmentation on the tissue in the upsampled initial segmentation image.

[0033] In one embodiment, the above-mentioned fine segmentation of the tissue in the primary segmented image to obtain a three-dimensional mask image of the target organ in the image to be processed includes:

[0034] After fine segmentation of the tissue in the initial segmentation image, the number of pixels of each category of tissue in the fine segmentation image is obtained;

[0035] Determine the tissue categories to be retained according to the number of tissue categories and the number of pixels of each tissue category in the image after fine segmentation;

[0036] The retained tissue categories are merged into one category, and the merged tissue category is morphologically repaired to obtain a three-dimensional mask image of the target organ in the image to be processed.

[0037] In a second aspect, an embodiment of the present application provides an organ segmentation device, the device comprising:

[0038] A registration module is used to register the template image and the image to be processed based on the mutual information between the template image and the image to be processed to obtain a registration result; the template image is determined according to any modality;

[0039] A processing module, used for processing the target organ mask in the template image according to the registration result to obtain the target organ region in the image to be processed;

[0040] The correction and segmentation module is used to perform non-uniform field correction on the target organ area of ​​the image to be processed and complete the segmentation of the tissue in the target organ to obtain a three-dimensional mask image of the target organ in the image to be processed.

[0041] In a third aspect, an embodiment of the present application provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of any method provided in the embodiment of the first aspect are implemented.

[0042] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of any method provided in the embodiment of the first aspect above are implemented.

[0043] The embodiments of the present application provide an organ segmentation method, apparatus, computer device and storage medium, which align the template image and the image to be processed based on the mutual information between the template image and the image to be processed determined by any modality to obtain a registration result, and based on the registration result, process the target organ mask in the template image to obtain the target organ region in the image to be processed, then perform non-uniform field correction on the target organ region of the image to be processed, and complete segmentation of the tissue in the target organ to obtain a three-dimensional mask image of the target organ in the image to be processed. In this method, the template image and the image to be processed are registered based on the mutual information between the template image and the image to be processed. By taking the mutual information as a similarity measurement index, better registration can be achieved, and the change information (registration result) between the image to be processed and the template image can be obtained more accurately. Moreover, the template image is obtained by using a structural magnetic resonance image under any modality. No matter which modality the template image or the image to be processed is, the change information between the image to be processed and the template image can be accurately obtained. In this way, any modality input in various scenarios can realize the segmentation of the image to be processed, thereby improving the versatility of structural magnetic resonance images of various modalities. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 FIG. 1 is an application environment diagram of an organ segmentation method in one embodiment;

[0045] Figure 2 is a schematic flow chart of an organ segmentation method in one embodiment;

[0046] Figure 3 is a schematic flow chart of an organ segmentation method in another embodiment;

[0047] Figure 4 is a schematic flow chart of an organ segmentation method in another embodiment;

[0048] Figure 5 is a schematic flow chart of an organ segmentation method in another embodiment;

[0049] Figure 6 A schematic diagram of a brain tissue image under a magnetic resonance image T1w mode in one embodiment;

[0050] Figure 7 A schematic diagram of a brain tissue image under a magnetic resonance image T2w mode in one embodiment;

[0051] Figure 8 is a schematic flow chart of an organ segmentation method in another embodiment;

[0052] Fig. 9 is a schematic flow chart of an organ segmentation method in another embodiment;

[0053] Fig.10 A schematic diagram of post-processing execution logic in one embodiment;

[0054] Fig.11 is a flow chart of a method for segmenting an organ in another embodiment;

[0055] Fig.12 FIG. 4 is a structural block diagram of an organ segmentation device in one embodiment. DETAILED DESCRIPTION

[0056] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0057] The organ segmentation method provided in this application can be applied to Figure 1 In the application environment shown in FIG. 1 , the organ segmentation method is applied in Figure 1 In the computer device shown, the computer device may be a server, and its internal structure diagram can be seen in Figure 1 . The computer device includes a processor, a memory and a network interface connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store organ segmentation-related data. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for segmenting an organ is implemented.

[0058] The embodiments of the present application provide an organ segmentation method, apparatus, computer equipment and storage medium, which can adapt to different application scenarios and data inputs in multiple modes and improve versatility. The technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems will be described in detail below through embodiments and in combination with the accompanying drawings. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. It should be noted that the organ segmentation method provided by the present application, Figure 2-Figure 11 The execution subject is a computer device. Figure 2-Figure 11 The execution subject may also be an organ segmentation device, which may be implemented as part or all of a computer device through software, hardware, or a combination of software and hardware.

[0059] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments.

[0060] In one embodiment, Figure 2 As shown, a method for segmenting an organ is provided, which involves a specific process in which a computer device registers the template image and the image to be processed based on the mutual information between the template image and the image to be processed determined by any modality, obtains a registration result, and processes the target organ mask in the template image according to the registration result to obtain the target organ region in the image to be processed, then performs non-uniform field correction on the target organ region in the image to be processed, and completes segmentation of the tissue in the target organ to obtain a three-dimensional mask image of the target organ in the image to be processed, and the method is applied to Figure 1 The computer device in the example is used to illustrate, including the following steps:

[0061] S101, registering the template image and the image to be processed based on the mutual information between the template image and the image to be processed to obtain a registration result; the template image is determined according to any modality.

[0062] The template image refers to a standard image of the tissue segmentation result in the target organ, wherein the template image can be implemented using a standard series of templates. Taking the target organ as the brain as an example, when the cerebral cortex is segmented, the MNI152 series templates can be used to make the template image (MNI is the abbreviation of the Montreal Neurosciences Institute in Canada, and MNI152 uses the weighted average of 152 3D T1 data of 152 healthy people), or the ICBM152 series templates can be used to make the template image (ICBM is the International Brain Imaging Association). When making the template image, the average template of the structural magnetic resonance image under any modality can be used as the template image, for example, the magnetic resonance T1W modality or T2W modality is used, and this embodiment does not limit this.

[0063] The image to be processed refers to a structural magnetic resonance image in which the target organ present therein needs to be segmented, for example, a structural magnetic resonance image of the cerebral cortex to be processed, etc., which is not limited in this embodiment. The mutual information between the template image and the image to be processed represents the correlation between the template image and the image to be processed, the amount of information contained in each other, etc. For example, in image registration, the mutual information between two images reflects the degree of mutual inclusion of information between them through their entropy and joint entropy. When the similarity between the two images is higher or the overlap is larger, the correlation is greater, and the joint entropy is smaller, that is, the mutual information is greater.

[0064] Since the template image and the image to be processed are registered based on the mutual information between the template image and the image to be processed, the change information (registration result) between the image to be processed and the template image can be accurately obtained. Moreover, since the template image is obtained by adopting the average template of the structural magnetic resonance image under any modality, that is, there is no restriction on the modality of the template image, in practical applications, no matter which modality the template image or the image to be processed is, the change information between the image to be processed and the template image can be accurately obtained. In this way, when the image to be processed is segmented, any modality can be realized, thereby improving the versatility of structural magnetic resonance images of various modalities.

[0065] Taking the cerebral cortex as an example, when segmenting the cerebral cortex, a preprocessing step must be performed first: scalping. Since scalping is a preprocessing step, the accuracy of local details during scalping has little effect on subsequent tissue segmentation. However, the overall result of scalping should not deviate too much. The mutual information of the template image and the image to be processed is used for registration, so that the template image of any modality and the image to be processed of any modality can be well registered, achieving a good scalping effect, making the scalping effect more robust, and thus making the results of subsequent tissue segmentation more accurate.

[0066] S102, processing the target organ mask in the template image according to the registration result to obtain the target organ region in the image to be processed.

[0067] After obtaining the registration result between the template image and the image to be processed, the target organ mask in the template image is processed according to the registration result. It can be understood here that although the template image can be obtained by using any modal structure magnetic resonance image, the target organ must exist in the template image, and the template image is a standard image containing the target organ tissue segmentation result, so the target organ mask in the template image itself is also a standard mask of the target organ. The registration result includes the deformation information of the corresponding position between the image to be processed and the template image. For example, if rigid registration and elastic registration are performed, the affine matrix and elastic deformation field are obtained respectively.

[0068] Specifically, the target organ mask in the template image is processed by the registration result, that is, the target organ mask in the template image is processed according to the change information in the registration result, and the target organ region in the image to be processed is obtained. For example, an affine matrix and an elastic deformation field are applied to the target organ mask in the template image to obtain the target organ mask in the image to be processed, and the target organ region is obtained.

[0069] S103, performing non-uniform field correction on the target organ region of the image to be processed, and completing segmentation of the tissue in the target organ to obtain a three-dimensional mask image of the target organ in the image to be processed.

[0070] After obtaining the target organ region in the image to be processed, it is necessary to perform certain processing (such as non-uniform field correction) on the target organ region in the image to be processed. After the processing, it is necessary to further segment the tissue in the target organ region of the image to be processed, and finally obtain a three-dimensional mask image of the target organ in the image to be processed. Among them, a pre-trained correction network model can be used to perform non-uniform field correction on the target organ region of the image to be processed, or an MRI non-uniform field correction algorithm can be used, such as the N4ITK algorithm integrated in the ITK tool, etc. This embodiment does not limit the method of non-uniform field correction.

[0071] In practical applications, when the target organ region of the processed image is subjected to inhomogeneous field correction and tissue segmentation in the target organ, tissue segmentation can be performed after the inhomogeneous field correction is completed, or inhomogeneous field correction and tissue segmentation can be combined together, for example, MRI inhomogeneous field and tissue segmentation are jointly modeled into the model to be realized. Taking the cerebral cortex as an example, the medical image segmentation technology based on deep learning can be used to segment the tissue in the brain using the convolutional neural network of multi-scale pathways. Alternatively, the brain tissue segmentation tool F AST (FMRIB's Automated Segmentation Tool) in FSL software (DTI image processing software under Linux system) can be used to segment the gray matter, white matter and cerebrospinal fluid in the brain tissue, and the spatial relationship between brain tissue pixels is modeled using a hidden Markov model, and the MRI inhomogeneous field correction task is combined into the framework, and finally the EM algorithm (Expectation-Maximization algorithm, maximum expectation algorithm) is used to solve the model parameters to achieve tissue segmentation.

[0072] The organ segmentation method provided in this embodiment is based on the mutual information between the template image and the image to be processed determined by any modality, and the template image and the image to be processed are registered to obtain the registration result, and according to the registration result, the target organ mask in the template image is processed to obtain the target organ region in the image to be processed, and then the target organ region of the image to be processed is corrected for non-uniform field, and the tissue in the target organ is segmented to obtain the three-dimensional mask image of the target organ in the image to be processed. In this method, the template image and the image to be processed are registered based on the mutual information between the template image and the image to be processed, and the mutual information is used as a similarity measurement index, so that better registration can be achieved, and the change information (registration result) between the image to be processed and the template image can be obtained more accurately, and the template image is obtained by using the structural magnetic resonance image under any modality. No matter which modality the template image or the image to be processed is, the change information between the image to be processed and the template image can be accurately obtained, so that any modality input in various scenarios can achieve segmentation of the image to be processed, thereby improving the versatility of various modality structural magnetic resonance images.

[0073] The process of registering a template image and an image to be processed is illustrated by an embodiment. Optionally, in one embodiment, the process of registering the template image and the image to be processed to obtain a registration result includes: rigidly registering the template image and the image to be processed to obtain an affine matrix, and elastically registering the template image and the image to be processed to obtain an elastic deformation field; accordingly, processing the target organ mask in the template image according to the registration result to obtain the target organ region in the image to be processed includes: applying the affine matrix and the elastic deformation field to the target organ mask in the template image to obtain the target organ region in the image to be processed.

[0074] Two types of registration are performed on the template image and the image to be processed: rigid registration and elastic registration. After rigid registration, an affine matrix is ​​obtained, and after elastic registration, an elastic deformation field is obtained. The affine matrix can reflect the change relationship between the vector space of the template image and the vector space of the image to be processed, and the elastic deformation field can reflect the deformation relationship between the template image and the image to be processed.

[0075] Based on this, when processing the target organ mask in the template image, an affine matrix can be applied to the target organ mask in the template image. In this way, the target organ mask in the template image is converted to the vector space where the image to be processed is located by applying a radial matrix; then the elastic deformation field is applied to the target organ mask after the affine matrix is ​​applied, and the deformed target organ mask with the affine matrix applied is obtained, which is the target organ area in the image to be processed.

[0076] The organ segmentation method provided in this embodiment performs rigid registration and elastic registration on the template image and the image to be processed, and then applies the radiation matrix and elastic deformation field obtained by the registration to the target organ mask in the template image, so as to obtain the target organ region in the image to be processed. In this way, through the multi-dimensional change relationship between the template image and the image to be processed, the target organ mask in the template image is processed to obtain the target organ region in the image to be processed, so that subsequent processing can be limited to the target organ region range, thereby ensuring the stability and accuracy of subsequent processing.

[0077] In the above S103, in the process of performing non-uniform field correction on the target organ region of the image to be processed and completing segmentation of the tissue in the target organ to obtain the three-dimensional mask image of the target organ in the image to be processed, the segmentation process can be divided into two processes: coarse segmentation and fine segmentation. For details, see Figure 3 As shown, in one embodiment, S103 includes the following steps:

[0078] S201, based on the grayscale difference between tissues, performing non-uniform field correction on the target organ region of the image to be processed, and roughly segmenting the tissue in the target organ to obtain a primary segmentation image of the target organ in the image to be processed.

[0079] In this embodiment, the non-uniform field correction and the rough segmentation are combined and performed, which can be achieved only by the grayscale difference between the tissues to be segmented in the target organ region. In this way, any modality of the magnetic resonance image can meet this requirement.

[0080] Among them, the target organ region of the image to be processed needs to be subjected to non-uniform field correction and rough tissue segmentation at the same time, and in order to ensure the accuracy of the final non-uniform field correction result and the rough segmentation result, it can be performed iteratively, for example, after the target organ region is subjected to non-uniform field correction and rough segmentation once, the target organ region is subjected to non-uniform field correction and rough segmentation again. Among them, the non-uniform field correction can be performed using a preset correction model, which is not limited in this embodiment. Among them, the rough segmentation can be performed according to the grayscale value of each tissue, that is, the area composed of pixels with the same or similar grayscale values ​​is one tissue, and the area composed of another group of pixels with the same or similar grayscale values ​​is another tissue.

[0081] Usually, there are a large number of pixel-by-pixel operations during the process of non-uniform field correction and coarse segmentation, which results in the algorithm efficiency of the whole process being related to the size of the image to be processed. An image to be processed that is too large will cause the algorithm to run very long. Therefore, in order to improve the segmentation efficiency of the target organ, optionally, in one embodiment, the image to be processed can be downsampled, and then non-uniform field correction and coarse tissue segmentation can be performed on the downsampled image. It should be noted that although downsampling will reduce the resolution of the entire image, the grayscale value difference between each pixel in the image still exists, and the combination of non-uniform field correction and coarse segmentation only requires the grayscale difference between the tissues to be segmented in the target organ area. Therefore, performing non-uniform field correction and coarse tissue segmentation on the downsampled image to be processed will not affect the accuracy of the final correction result and the segmentation result.

[0082] S202, performing fine segmentation on the tissue in the initially segmented image to obtain a three-dimensional mask image of the target organ in the image to be processed.

[0083] After obtaining the initial segmentation results of each tissue in the target organ region of the image to be processed after the above-mentioned rough segmentation, the tissue in the initial segmentation result is finely segmented, for example, the grayscale value of each pixel value is further judged according to the grayscale value of each tissue to accurately determine which tissue each pixel belongs to, so as to perform fine segmentation on each tissue in the target organ region. Optionally, if the image to be processed is downsampled during the rough segmentation, then when performing the fine segmentation, the image to be processed is upsampled back to the original resolution, so that the tissue to which each pixel belongs is determined in the image to be processed that is upsampled to the original resolution, which further improves the accuracy of the fine segmentation of each tissue in the target organ region.

[0084] In the organ segmentation method provided in the present embodiment, the inhomogeneous field and the rough segmentation of the image to be processed are simultaneously performed in the image to be processed after downsampling, and then the rough segmentation result is finely segmented in the image to be processed that is upsampled to the original resolution. In this way, the inhomogeneous field and the rough segmentation are simultaneously performed in the image to be processed after downsampling, which saves computing resources and improves the efficiency of organ segmentation in the image to be processed.

[0085] like Figure 4 As shown, an embodiment is provided, which relates to a specific process of simultaneously performing non-uniform field and coarse segmentation in the downsampled image to be processed. The above S201 process includes:

[0086] S301, downsampling the target organ region of the image to be processed by a preset multiple to obtain the downsampled target organ region of the image to be processed.

[0087] The target organ region of the image to be processed is downsampled, and the preset downsampling multiple, for example, three times, four times, etc., can be determined according to the actual situation and is not limited to this. Among them, the downsampling method includes but is not limited to the neighboring sampling method, the quadratic interpolation method, the bicubic convolution method, etc. After the target organ region of the image to be processed is downsampled by a preset multiple, the target organ region of the image to be processed after the downsampling is obtained. The target organ region of the image to be processed after the downsampling has fewer data samples, so the operation time can be reduced when performing non-uniform field correction and coarse segmentation.

[0088] S302, construct an objective function including a category grayscale vector, a membership matrix and an inhomogeneous field, and obtain the category grayscale vector, the membership matrix and the inhomogeneous field corresponding to the minimum value of the objective function; the category grayscale vector represents the central grayscale value of each category of tissue, and the membership represents the probability of each pixel point in the corresponding category of tissue.

[0089] After obtaining the target organ region of the downsampled image to be processed, the size of the inhomogeneous field must be determined before performing the inhomogeneous field correction, and then the inhomogeneous field correction is performed on the target organ region of the downsampled image to be processed according to the determined inhomogeneous field.

[0090] Similarly, before roughly segmenting the tissue in the target organ region, it is necessary to first determine to which category of tissue each pixel in the target organ region belongs. For example, there are four categories of tissue in the cerebral cortex: gray matter, white matter, cerebrospinal fluid, and interstitial space. The grayscale values ​​of these four categories of tissue are definitely different. Therefore, as long as it is determined to which category of tissue each pixel in the target organ region belongs (for example, determined by membership and category grayscale vector), the area where each tissue is located can be determined, thereby completing the rough segmentation of each tissue.

[0091] The size, membership and category grayscale vector of the inhomogeneous field can be determined by a pre-established objective function including the inhomogeneous field, membership and category grayscale vector. The category grayscale vector represents the central grayscale value of each category of tissue, and the membership represents the probability of each pixel in the corresponding category of tissue. After constructing the objective function including the category grayscale vector, membership matrix and inhomogeneous field, the minimum value of the objective function is obtained, and the category grayscale vector, membership matrix and inhomogeneous field corresponding to the minimum value of the objective function are used as the size, membership and category grayscale vector of the inhomogeneous field required in the end.

[0092] Optionally, in one embodiment, Figure 5 As shown, constructing the objective function including the category grayscale vector, the membership matrix and the inhomogeneous field includes the following steps:

[0093] S401, obtaining a model expression of a target organ's real image and a non-uniform field model expression; the model expression of the target organ's real image includes a category grayscale vector and a membership matrix.

[0094] The real image of the target organ represents an image that is not affected by any noise and uneven field images. The grayscale value of each type of tissue in the real image should be the same. Taking the brain as an example, the real image of the brain is represented by T, which contains gray matter, white matter, cerebrospinal fluid and other tissues with different grayscales. The grayscale value of each type of tissue is the same. If the tissues in the brain can be classified into N categories, then the model expression of T can be defined as the formula: In this formula, Ci is the grayscale value of the i-th category in N categories, that is, the category grayscale vector; Mi(x,y,z) is the membership of the pixel at the (x,y,z) position to the i-th category, and x, y, z represent the coordinates of the pixel in the three-dimensional voxel in the x-axis, y-axis and z-axis directions.

[0095] Among them, any MRI inhomogeneous field can be simulated, and the model expression of the inhomogeneous field B(x, y, z) is as follows: In this formula, bi is the i-th base field, ai is the weight coefficient of the i-th base field, and M is the number of base fields.

[0096] S402: establishing an image model expression including a multiplicative inhomogeneous field according to a model expression of a real image of a target organ and a model expression of an inhomogeneous field.

[0097] S403, constructing an objective function according to an image model expression including a multiplicative inhomogeneous field.

[0098] After the model expression of the target organ's real image and the inhomogeneous field model expression are established, an image model expression including a multiplicative inhomogeneous field is established according to the model expression of the target organ's real image and the inhomogeneous field model expression.

[0099] Among them, the model expression of the image I(x, y, z) including the multiplicative inhomogeneous field is as follows: I(x, y, z)=B(x, y, z)*T(x, y, z)+n(x, y, z), wherein I in the formula is the grayscale matrix of the image superimposed with the multiplicative inhomogeneous field and additive random noise, B is the inhomogeneous field, T is the grayscale matrix of the real image of the target organ, and n is the noise.

[0100] Furthermore, an objective function may be constructed based on the above-mentioned image model expression including the multiplicative inhomogeneous field, the model expression of the target organ's real image, and the inhomogeneous field model expression. The objective function is as follows:

[0101] The minimum value of the objective function is obtained by optimizing the objective function to obtain the category grayscale vector C, the membership matrix M and the weight coefficient ai of the basis field (B can be further obtained according to ai) that minimize the objective function. For example, the process of optimizing the objective function includes: first initializing Ci and Mi, then taking the partial derivative of ai through F, taking the extreme value of ai, updating B according to the extreme value of ai, and taking the partial derivative of C through F, taking the extreme value of C, and obtaining the updated C, and taking the partial derivative of M through F, taking the extreme value of M, and obtaining the updated M, repeating the process of iteratively updating B, C, and M until the preset convergence conditions are met, and obtaining the category grayscale vector C, the membership matrix M and the inhomogeneous field B corresponding to the minimum of the objective function.

[0102] S303, performing non-uniform field correction on the target organ region of the downsampled image to be processed according to the non-uniform field, and performing coarse segmentation on the tissue in the target organ region according to the membership matrix, to obtain a primary segmentation image of the target organ in the downsampled image to be processed.

[0103] After the iterative convergence of the objective function, the final membership matrix M, category grayscale vector C and non-uniform field B are obtained. Then, the non-uniform field correction is performed on the target organ region of the downsampled image to be processed according to the non-uniform field B. For example, the brain mask of the downsampled image to be processed is divided by B to obtain the brain region of the downsampled image to be processed after the non-uniform field correction. The tissues in the target organ region are roughly segmented according to the membership matrix M. Of course, the tissues in the target organ region can also be roughly segmented according to the category grayscale vector C. The membership matrix M or the category grayscale vector C can be used to determine which tissue category each pixel in the target organ region of the image to be processed belongs to. The segmentation of each pixel can be completed according to the tissue category to which each pixel belongs, thereby completing the rough segmentation of each tissue in the target organ in the image to be processed.

[0104] The organ segmentation method provided in the present embodiment establishes an objective function including a category grayscale vector, a membership matrix and an inhomogeneous field, and performs inhomogeneous field correction and coarse segmentation of tissues in the target organ region of the processed image according to the category grayscale vector, the membership matrix and the inhomogeneous field corresponding to the minimum value of the objective function, so that the process of inhomogeneous field correction and coarse segmentation is more accurate and efficient; and the inhomogeneous field correction and coarse segmentation are performed in the downsampled image, and the number of pixel data points in the downsampled image is greatly reduced, thereby reducing the amount of calculation and further improving the efficiency of the process of inhomogeneous field correction and coarse segmentation.

[0105] After the tissue in the target organ region in the image to be processed is roughly segmented, the next step is to continue fine segmentation to improve the final tissue segmentation accuracy. Optionally, before fine segmentation of the rough segmentation result of the target organ in the processed image, the number of fine segmentation categories needs to be adjusted according to the specific modality of the image to be processed, that is, it is necessary to determine how many categories the tissue in the target organ region should be segmented into during fine segmentation.

[0106] Taking the brain as an example, between the cerebral cortex and the skull, there are cerebrospinal fluid, meningeal structures, and the gap between the cerebrospinal fluid and the meninges (hereinafter referred to as the brain gap). During the scalp removal process (the above-mentioned registration process), it is not only impossible to remove the cerebrospinal fluid, but it is even difficult to perfectly remove the meninges and the brain gap. The grayscale value of the meningeal structure is very similar to the gray matter. It is necessary to rely on the accurate classification of the cerebrospinal fluid and the brain gap as isolation to use the morphological post-processing method to cleanly remove the meningeal structure. There are four types of tissues in the brain: gray matter, white matter, cerebrospinal fluid, and brain gap. However, the grayscale values ​​of each tissue are different in the magnetic resonance images of the brain in different modes. For example, in the T1w mode, the cerebrospinal fluid and the skull gap are very close low grayscale values ​​(please refer to Figure 6 ), even if the scalp removal effect is not good, there will be no adhesion between the meninges and the gray matter. During fine segmentation, only three categories (gray matter, white matter, and cerebrospinal fluid) can be performed; however, in the T2w mode, the grayscale performance of the cerebrospinal fluid and the interstitial space is completely different (please refer to Figure 7 ), there is only a thin layer of cerebrospinal fluid. Only by doing four-part classification during fine segmentation can we avoid adhesion between the meninges or brain space and gray matter, so as to achieve complete removal of the extracortical structure.

[0107] Based on this, in one embodiment, Figure 8 As shown, this embodiment relates to a specific process of determining the number of tissue categories that need fine segmentation before fine segmentation of the initial segmentation image of the target organ in the image to be processed. This embodiment includes the following steps:

[0108] S501, determining the number of pixels of the tissue with the smallest grayscale value and the number of pixels of the tissue with the largest grayscale value in the target organ of the image to be processed according to the category grayscale vector.

[0109] In the previous rough segmentation process, since the modality of the image to be processed is unclear, for example, it is unclear whether it is T1w modality or T2w modality, four categories are uniformly selected: gray matter, white matter, cerebrospinal fluid and intercerebral space (the brain is also used as an example for explanation in this embodiment). For different tissues in the target organ of the image to be processed, the area occupied by each tissue in the target organ is fixed. For example, in the magnetic resonance image of the brain in the T1w modality, the grayscale of cerebrospinal fluid and intercerebral space is lower than that of gray matter and white matter, and their number of pixels is also lower than that of gray matter and white matter; in the magnetic resonance image of the brain in the T2w modality, the grayscale of cerebrospinal fluid is higher than that of all other tissue categories, but the number of pixels is definitely less than that of gray matter and white matter. Therefore, the number of pixels occupied by different grayscale tissues in the rough classification results can be used to determine the modality of the data and select the number of categories required for fine segmentation.

[0110] Through the category grayscale vector determined in the rough segmentation process, the grayscale value of each category of tissue can be counted. Then, based on the above category grayscale vector, the category of tissue with the smallest grayscale value in the target organ of the image to be processed is determined, and the number of pixels of this category of tissue is obtained to obtain the number of pixels of the tissue with the smallest grayscale value; similarly, based on the above category grayscale vector, the category of tissue with the largest grayscale value in the target organ of the image to be processed is determined, and the number of pixels of this category of tissue is obtained to obtain the number of pixels of the tissue with the largest grayscale value.

[0111] S502, comparing the number of pixels of the tissue with the smallest gray value with the number of pixels of the designated tissue in the target organ, and the number of pixels of the tissue with the largest gray value with the number of pixels of the designated tissue in the target organ.

[0112] After determining the number of pixels of a type of tissue with the smallest grayscale value and the number of pixels of a type of tissue with the largest grayscale value, the number of pixels thereof is respectively compared with the number of pixels of a designated tissue in the target organ.

[0113] The designated tissue in the target organ refers to the tissue with a relatively small number of pixels in the standard magnetic resonance images of the target organ in different modalities. For example, in the magnetic resonance image of the brain in the T1w modality, the grayscale value of the cerebrospinal fluid is the lowest, and the number of pixels of the cerebrospinal fluid is also less than that of the gray matter and the white matter; while in the magnetic resonance image of the brain in the T2w modality, the grayscale value of the cerebrospinal fluid is the highest, but the number of pixels of the cerebrospinal fluid is indeed the least, that is, in different modalities, although the grayscale value of the cerebrospinal fluid in T1w is the lowest and the grayscale value in T2w is the highest, it is a type of tissue with a small number of pixels in both modalities.

[0114] Therefore, the number of pixels of the tissue with the smallest grayscale value is compared with the number of pixels of the specified tissue in the target organ to determine whether the type of tissue with the smallest grayscale value belongs to any tissue in the less-numbered category. If so, it can be determined that the image to be processed belongs to the T1w modality; however, if the number of pixels of the tissue with the largest grayscale value is compared with the number of pixels of the specified tissue in the target organ, and the type of tissue with the largest grayscale value belongs to any tissue in the less-numbered category, then the image to be processed belongs to the T2w modality. It should be noted that in the brain, only one of the tissue with the smallest grayscale value or the tissue with the largest grayscale value belongs to the less-numbered category in the target organ, and both do not belong to the less-numbered category of tissue at the same time. However, if it is in other organs, it can be determined according to actual conditions, and this embodiment does not limit this.

[0115] Generally, the number of pixels of each tissue in the target organ of the standard to-be-processed images under different modalities is known in advance. Taking the brain as an example, the number of pixels of gray matter, white matter, cerebrospinal fluid and interstitial tissue in the brain under various modalities can be directly determined. Moreover, based on this, the number of pixels of these four categories can be divided into two categories in advance: the category with a smaller number and the category with a larger number. For example, the number of pixels of these four categories of tissues is in the following order from large to small: gray matter, white matter, cerebrospinal fluid, and interstitial space. Then gray matter and white matter belong to the category with a larger number, and cerebrospinal fluid and interstitial space belong to the category with a smaller number. The number of pixels of the tissue with a smaller number can be determined as the number of pixels of the specified tissue in the target organ.

[0116] S503, if the number of pixels of the tissue with the smallest grayscale value belongs to the range of the number of pixels of the specified tissue in the target organ, it is determined that the tissue in the initial segmentation image needs to be finely segmented into the first value category number; if the number of pixels of the tissue with the largest grayscale value belongs to the range of the number of pixels of the specified tissue in the target organ, it is determined that the tissue in the initial segmentation image needs to be finely segmented into the second value category number.

[0117] The number of pixels of the tissue with the smallest gray value is compared with the number of pixels of the specified tissue in the target organ. If the type of tissue with the smallest gray value belongs to any of the pixel numbers of the specified tissue (the type with fewer pixels), it is determined that the image to be processed belongs to the T1w mode. In this case, it is determined that the tissue in the initial segmentation image needs to be finely segmented into the first value category number, for example, finely segmented into 3 categories (gray matter, white matter, cerebrospinal fluid).

[0118] However, if after comparing the number of pixels of the tissue with the largest gray value with the number of pixels of the designated tissue in the target organ, the type of tissue with the largest gray value belongs to any of the tissues with the number of pixels of the designated tissue (the type with a smaller number), then the image to be processed belongs to the T2w modality. In this case, it is determined that the tissues in the initial segmentation image need to be finely segmented into the second value category number, for example, finely segmented into 4 categories (gray matter, white matter, cerebrospinal fluid, and brain space).

[0119] In the organ segmentation method provided in the present embodiment, the modality of the image to be processed is determined based on the grayscale value of each tissue in the initial segmentation result obtained by coarse segmentation, that is, the corresponding number of pixels, and the number of categories that need to be finely segmented for each tissue in the target organ in the image to be processed is determined based on the modality. The accuracy of the subsequent fine segmentation results is guaranteed by judging the modality and the number of fine segmentations.

[0120] An embodiment is provided, which relates to a specific process of finely segmenting tissues in a primary segmented image. The embodiment includes the following steps: upsampling the primary segmented image by a preset multiple to obtain an upsampled primary segmented image; if the tissues in the primary segmented image need to be finely segmented into a first value number of categories, the centers of each category in the category grayscale vector are used as initial category centers for fine segmentation, and the membership matrix after upsampling the membership matrix by a preset multiple is used as the initial membership matrix for fine segmentation, and iterative clustering is performed on the histogram of the upsampled primary segmented image to finely segment the tissues in the upsampled primary segmented image; if the tissues in the primary segmented image need to be finely segmented into a second value number of categories, the grayscale values ​​of the second value number are taken at an average interval from the category grayscale vector as the initial category centers for fine segmentation, the memberships of the second value number are taken at an average interval within the range of the membership matrix as the initial membership matrix for fine segmentation, and iterative clustering is performed on the histogram of the upsampled primary segmented image to finely segment the tissues in the upsampled primary segmented image.

[0121] Since the non-uniform field correction and rough segmentation are performed in the target organ area of ​​the downsampled image to be processed, the initial segmented image obtained is also a downsampled image. In order to ensure the accuracy of the fine segmentation result, the initial segmented image needs to be upsampled to the original resolution of the image to be processed. Specifically, the preset multiple is downsampled before, and the preset multiple is also upsampled here. The downsampling and upsampling multiples are the same, so that the initial segmented image can be upsampled to the original resolution of the image to be processed.

[0122] During the specific fine segmentation, the fine segmentation is performed in combination with the number of tissue categories determined above, that is, the fine segmentation process is different when the number of tissue categories is different.

[0123] Specifically, if the number of finely segmented tissues is the number of first-value categories, the center of each category in the category grayscale vector is used as the initial category center for fine segmentation, and the membership matrix after upsampling the membership matrix by a preset multiple is used as the initial membership matrix for fine segmentation. Iterative clustering is performed on the histogram of the upsampled initial segmentation image to perform fine segmentation on the tissues in the upsampled initial segmentation image.

[0124] For example, the number of categories in the first value is four. Since the number of categories is the same as the default four categories in coarse segmentation, the center of each category in the category grayscale vector in the coarse segmentation can be directly used as the initial category center for fine segmentation, and the membership matrix after upsampling the membership matrix by a preset multiple can be used as the initial membership matrix for fine segmentation. Then, the new category center and membership matrix are iteratively updated until the iteration is terminated to obtain the final category center and membership matrix, and the four types of tissues are finely segmented according to the final category center and membership matrix.

[0125] If the number of finely segmented tissues is the number of second-value categories, then the grayscale values ​​of the second value are taken at average intervals from the category grayscale vector as the initial category centers of the fine segmentation, and the memberships of the second value are taken at average intervals from the range of the membership matrix as the initial membership matrix of the fine segmentation. Iterative clustering is performed on the histogram of the upsampled initial segmentation image to perform fine segmentation on the tissues in the upsampled initial segmentation image.

[0126] For example, the number of categories in the second value is three. Since the number of categories is different from the default four categories during coarse segmentation, the category center and the membership matrix need to be randomly initialized again, that is, three nodes are taken on average from the category grayscale vector between the highest grayscale value and the lowest grayscale value. These three nodes are the initial category centers of the three categories. Three nodes are taken at an average interval from the range of the membership matrix as the initial membership matrix for fine segmentation. Then, the iterative update is started to determine the new category center and membership matrix until the iteration is terminated to obtain the final category center and membership matrix. The three types of tissues are finely segmented according to the final category center and membership matrix.

[0127] Among them, whether it is three or four categories, the fuzzy mean clustering algorithm of the histogram is used in the final iteration of the category center and the membership matrix. The algorithm's computational efficiency is greatly improved because it is iterated on the histogram instead of the original three-dimensional voxel space. Of course, since the histogram clustering algorithm cannot combine spatial related information, in order to avoid affecting the accuracy of the results, combined with certain spatial information, to achieve the effect of edge preservation and random scattered noise removal, the final membership function obtained by clustering can be median filtered, which can greatly improve the final segmentation effect.

[0128] After fine segmentation of the tissue in the initial segmentation image, some post-processing is required to improve the quality of the final 3D mask image, such as Fig. 9 As shown, in one embodiment, the above S202 includes the following steps:

[0129] S601, after fine segmentation of the tissue in the initially segmented image, the number of pixels of each type of tissue in the finely segmented image is obtained.

[0130] Taking the brain as an example, after fine segmentation of various tissues in the brain, post-processing can remove other tissues except gray matter and white matter, and then merge the gray and white matter to form the final brain segmentation result. After post-processing, the cerebrospinal fluid, meninges and brain spaces can be cleanly removed, and the sulci and convolutions of the cerebral cortex can be fully presented.

[0131] In post-processing, different post-processing methods need to be performed according to the number of finely segmented tissue categories and the number of pixels of each tissue. Therefore, the number of finely segmented tissue categories and the number of pixels of each category of tissue need to be determined first. The number of finely segmented tissue categories can be directly obtained in the above process, and the number of pixels of each category of tissue can also be determined based on the fine segmentation results.

[0132] S602: Determine the tissue categories to be retained according to the number of tissue categories and the number of pixels of each tissue category in the finely segmented image.

[0133] The tissue categories to be retained are determined based on the number of tissue categories and the number of pixels of each category of tissue in the image after fine segmentation.

[0134] Specifically, see Fig.10 As shown, assuming that the number of pixels is sorted from small to large, the number of pixels of the i-th tissue is NNMi (0NiNN, N∈N3, 4N), then NNM1N…NNMi…NNNMN, the sum of pixels of all categories is NNMsum, then if the number of tissue categories is three, then whether the ratio of NNM2 / NNM3 is less than the preset value a, if less than a, the retained tissue category is the tissue category corresponding to NNM3, if greater than a, the retained tissue category is the tissue category corresponding to NNM2 and NNM3. If the number of tissue categories is four, first determine whether the ratio of NNM3 / NNMsum is less than the preset value b, if less than b, the retained tissue category is the tissue category corresponding to NNM4; if greater than b, then continue to determine whether the ratio of NNM2 / NNMsum is less than the preset value c, if less than c, the retained tissue category is the tissue category corresponding to NNM3 and NNM4, if greater than c, the retained tissue category is the tissue category corresponding to NNM2, NNM3 and NNM4.

[0135] S603, merging the retained tissue categories into one category, and performing morphological repair on the merged one category of tissue to obtain a three-dimensional mask image of the target organ in the image to be processed.

[0136] Please continue to see Fig.10 As shown, after the tissue categories to be retained are determined, the retained tissue categories are merged into one category. The merged tissue category is then morphologically repaired, for example, corrosion operation, maximum connected domain algorithm to remove redundant residue blocks, expansion operation, etc., which are not limited in this embodiment. After morphological repair, a three-dimensional mask image of the brain in the image to be processed can be obtained.

[0137] In the organ segmentation method provided in this embodiment, the finely segmented image is post-processed by post-processing, and during the post-processing, the different tissue categories to be retained are determined in combination with the number of finely segmented tissue categories and the number of pixels of each tissue. In this way, targeted post-processing is performed, which greatly ensures the accuracy of the final segmented three-dimensional mask image.

[0138] In one embodiment, Fig.11 As shown, the embodiment of the present application provides an organ segmentation method, taking the cerebral cortex segmentation process as an example, the embodiment includes:

[0139] S701, using a template registration method, based on the mutual information between the template image and the image to be processed, registering the template image and the image to be processed to obtain a brain region of the image to be processed;

[0140] S702, downsampling the brain region of the image to be processed;

[0141] S703, performing non-uniform field correction on the brain region of the sampled image to be processed and performing rough segmentation on each tissue in the brain region to obtain a primary segmentation image;

[0142] S704, upsampling the first segmented image;

[0143] S705, determining the modality of the brain region of the image to be processed, and determining the number of categories to be precisely segmented according to the modality;

[0144] S706, performing fine segmentation on each tissue in the upsampled initial segmentation image based on the number of categories to be finely segmented;

[0145] S707, post-processing the finely segmented image to obtain a final three-dimensional mask image of the brain.

[0146] The implementation principles and technical effects of each step in the organ segmentation method provided in this embodiment are similar to those in the previous organ segmentation method embodiments, and will not be repeated here. Fig.11The implementation method of each step in the embodiment is only an example, and the implementation method is not limited. The order of each step can be adjusted in actual application as long as the purpose of each step can be achieved.

[0147] It should be understood that although Figure 2-11 The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Moreover, Figure 2-11 At least part of the steps may include multiple steps or multiple stages. These steps or stages are not necessarily performed at the same time, but can be performed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed in turn or alternately with other steps or at least part of the steps or stages in other steps.

[0148] In one embodiment, Fig.12 As shown, an organ segmentation device is provided, comprising: a registration module 10, a processing module 11 and a correction segmentation module 12, wherein:

[0149] A registration module 10 is used to register the template image and the image to be processed based on the mutual information between the template image and the image to be processed to obtain a registration result; the template image is determined according to any modality;

[0150] A processing module 11 is used to process the target organ mask in the template image according to the registration result to obtain the target organ region in the image to be processed;

[0151] The correction and segmentation module 12 is used to perform non-uniform field correction on the target organ region of the image to be processed and to segment the tissue in the target organ to obtain a three-dimensional mask image of the target organ in the image to be processed.

[0152] In one embodiment, the above-mentioned registration module 10 is specifically used to perform rigid registration on the template image and the image to be processed to obtain an affine matrix, and to perform elastic registration on the template image and the image to be processed to obtain an elastic deformation field; the above-mentioned processing module 11 is specifically used to apply the affine matrix and the elastic deformation field to the target organ mask in the template image to obtain the target organ area in the image to be processed.

[0153] In one embodiment, the correction and segmentation module 12 includes:

[0154] A correction and segmentation unit is used to perform non-uniform field correction on the target organ region of the image to be processed based on the grayscale difference between tissues, and to perform rough segmentation on the tissues in the target organ to obtain a primary segmentation image of the target organ in the image to be processed;

[0155] The fine segmentation unit is used to perform fine segmentation on the tissue in the primary segmentation image to obtain a three-dimensional mask image of the target organ in the image to be processed.

[0156] In one embodiment, the correction and segmentation unit includes:

[0157] A downsampling subunit, used for downsampling the target organ region of the image to be processed by a preset multiple to obtain the target organ region of the image to be processed after downsampling;

[0158] A construction subunit is used to construct an objective function including a category grayscale vector, a membership matrix and an inhomogeneous field, and obtain the category grayscale vector, the membership matrix and the inhomogeneous field corresponding to the minimum value of the objective function; the category grayscale vector represents the central grayscale value of each category of tissue, and the membership represents the probability of each pixel point in the corresponding category of tissue;

[0159] The correction and segmentation subunit is used to perform non-uniform field correction on the target organ area of ​​the downsampled image to be processed according to the non-uniform field, and to perform rough segmentation on the tissue in the target organ area according to the membership matrix to obtain a primary segmentation image of the target organ in the downsampled image to be processed.

[0160] In one embodiment, the above-mentioned construction subunit is specifically used to obtain a model expression of a real image of a target organ and a model expression of an inhomogeneous field; the model expression of the real image of the target organ includes a category grayscale vector and a membership matrix; an image model expression including a multiplicative inhomogeneous field is established based on the model expression of the real image of the target organ and the model expression of the inhomogeneous field; and an objective function is constructed based on the image model expression including a multiplicative inhomogeneous field.

[0161] In one embodiment, the correction and segmentation module further includes:

[0162] A first determining unit is used to determine the number of pixels of the tissue with the smallest grayscale value and the number of pixels of the tissue with the largest grayscale value in the target organ of the image to be processed according to the category grayscale vector;

[0163] A comparison unit, used for comparing the number of pixels of the tissue with the smallest gray value with the number of pixels of the designated tissue in the target organ, and the number of pixels of the tissue with the largest gray value with the number of pixels of the designated tissue in the target organ;

[0164] The second determination unit is used to determine that the tissue in the initial segmentation image needs to be finely segmented into the first value category number if the number of pixels of the tissue with the smallest grayscale value belongs to the range of the number of pixels of the specified tissue in the target organ; if the number of pixels of the tissue with the largest grayscale value belongs to the range of the number of pixels of the specified tissue in the target organ, then determine that the tissue in the initial segmentation image needs to be finely segmented into the second value category number.

[0165] In one embodiment, the fine segmentation unit comprises:

[0166] An upsampling subunit, used for upsampling the primary segmented image by a preset multiple to obtain an upsampled primary segmented image;

[0167] A fine segmentation subunit is used for, if the tissue in the initial segmentation image needs to be finely segmented into a first value of category quantity, taking the center of each category in the category grayscale vector as the initial category center for fine segmentation, and taking the membership matrix after upsampling the membership matrix by a preset multiple as the initial membership matrix for fine segmentation, and performing iterative clustering on the histogram of the upsampled initial segmentation image to finely segment the tissue in the upsampled initial segmentation image; if the tissue in the initial segmentation image needs to be finely segmented into a second value of category quantity, taking the grayscale values ​​of the second value at an average interval from the category grayscale vector as the initial category center for fine segmentation, taking the membership of the second value at an average interval within the range of the membership matrix as the initial membership matrix for fine segmentation, and performing iterative clustering on the histogram of the upsampled initial segmentation image to finely segment the tissue in the upsampled initial segmentation image.

[0168] In one embodiment, the fine segmentation unit further includes:

[0169] An acquisition subunit is used to perform fine segmentation on the tissues in the initial segmentation image and obtain the number of pixels of each category of tissues in the fine segmentation image;

[0170] A determination subunit is used to determine the tissue category to be retained according to the number of tissue categories and the number of pixels of each category of tissue in the finely segmented image;

[0171] The processing subunit is used to merge the retained tissue categories into one category, and perform morphological repair on the merged one category of tissue to obtain a three-dimensional mask image of the target organ in the image to be processed.

[0172] The specific definition of the organ segmentation device can be found in the definition of the organ segmentation method above, which will not be repeated here. Each module in the above-mentioned organ segmentation device can be implemented in whole or in part by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0173] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as described above. Figure 1As shown. The computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, an organ segmentation method is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covered on the display screen, or a button, trackball or touchpad set on the computer device housing, or an external keyboard, touchpad or mouse, etc.

[0174] Those skilled in the art will appreciate that the above Figure 1 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0175] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the following steps are implemented:

[0176] Based on the mutual information between the template image and the image to be processed, the template image and the image to be processed are registered to obtain a registration result; the template image is determined according to any modality;

[0177] According to the registration result, the target organ mask in the template image is processed to obtain the target organ region in the image to be processed;

[0178] The target organ region of the image to be processed is subjected to non-uniform field correction, and the tissue in the target organ is segmented to obtain a three-dimensional mask image of the target organ in the image to be processed.

[0179] The above embodiment provides a computer device, whose implementation principle and technical effect are similar to those of the above method embodiment, and will not be repeated here.

[0180] In one embodiment, a computer readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:

[0181] Based on the mutual information between the template image and the image to be processed, the template image and the image to be processed are registered to obtain a registration result; the template image is determined according to any modality;

[0182] According to the registration result, the target organ mask in the template image is processed to obtain the target organ region in the image to be processed;

[0183] The target organ region of the image to be processed is subjected to non-uniform field correction, and the tissue in the target organ is segmented to obtain a three-dimensional mask image of the target organ in the image to be processed.

[0184] The above embodiment provides a computer-readable storage medium, whose implementation principle and technical effect are similar to those of the above method embodiment, and will not be repeated here.

[0185] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory or optical memory, etc. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0186] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0187] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A method for segmenting an organ, characterized in that: The method comprises: Based on the mutual information between the template image and the image to be processed, registering the template image and the image to be processed to obtain a registration result; the template image is determined according to any modality; According to the registration result, the target organ mask in the template image is processed to obtain the target organ region in the image to be processed; Based on the grayscale difference between tissues, non-uniform field correction is performed on the target organ region of the image to be processed, and the tissue in the target organ is roughly segmented to obtain a primary segmentation image of the target organ in the image to be processed; After fine segmentation of the tissue in the initially segmented image, the number of pixels of each category of tissue in the fine segmented image is obtained; Determining the tissue categories to be retained according to the number of tissue categories in the image after fine segmentation and the number of pixels of each tissue category; The retained tissue categories are merged into one category, and the merged one category of tissue is morphologically repaired to obtain a three-dimensional mask image of the target organ in the image to be processed.

2. The method according to claim 1, characterized in that: The registering the template image and the image to be processed to obtain a registration result includes: Performing rigid registration on the template image and the image to be processed to obtain an affine matrix, and performing elastic registration on the template image and the image to be processed to obtain an elastic deformation field; Then, according to the registration result, the target organ mask in the template image is processed to obtain the target organ region in the image to be processed, including: The affine matrix and the elastic deformation field are applied to the target organ mask in the template image to obtain the target organ region in the image to be processed.

3. The method according to claim 1 or 2, characterized in that: The mutual information represents the correlation between the template image and the image to be processed, and the amount of information they contain.

4. The method according to claim 1 or 2, characterized in that: The method of performing non-uniform field correction on the target organ region of the image to be processed based on the grayscale difference between tissues and performing rough segmentation on the tissue in the target organ to obtain a primary segmentation image of the target organ in the image to be processed includes: Downsampling the target organ region of the image to be processed by a preset multiple to obtain the downsampled target organ region of the image to be processed; Constructing an objective function including a category grayscale vector, a membership matrix and an inhomogeneous field, and obtaining the category grayscale vector, the membership matrix and the inhomogeneous field corresponding to the minimum value of the objective function; the category grayscale vector represents the central grayscale value of each category of tissue, and the membership degree represents the probability of each pixel point in the corresponding category of tissue; The target organ region of the downsampled image to be processed is subjected to uneven field correction according to the uneven field, and the tissue in the target organ region is roughly segmented according to the membership matrix to obtain a primary segmentation image of the target organ in the downsampled image to be processed.

5. The method according to claim 4, characterized in that The construction includes a category grayscale vector, a membership matrix and an objective function of an inhomogeneous field, including: Acquire a model expression of the target organ's real image and an inhomogeneous field model expression; the model expression of the target organ's real image includes the category grayscale vector and the membership matrix; Establishing an image model expression including a multiplicative inhomogeneous field according to the model expression of the target organ's real image and the inhomogeneous field model expression; The objective function is constructed according to the image model expression including the multiplicative inhomogeneous field.

6. The method according to claim 4, characterized in that Before fine segmentation of the primary segmented image of the target organ in the image to be processed, the method further comprises: According to the category grayscale vector, determining the number of pixels of the tissue with the smallest grayscale value and the number of pixels of the tissue with the largest grayscale value in the target organ of the image to be processed; comparing the number of pixels of the tissue with the smallest gray value with the number of pixels of the designated tissue in the target organ, and comparing the number of pixels of the tissue with the largest gray value with the number of pixels of the designated tissue in the target organ; If the number of pixels of the tissue with the smallest grayscale value belongs to the range of the number of pixels of the specified tissue in the target organ, it is determined that the tissue in the initial segmentation image needs to be finely segmented into the first value category number; if the number of pixels of the tissue with the largest grayscale value belongs to the range of the number of pixels of the specified tissue in the target organ, it is determined that the tissue in the initial segmentation image needs to be finely segmented into the second value category number.

7. The method according to claim 6, characterized in that The fine segmentation of the tissue in the primary segmentation image comprises: Upsampling the initial segmented image by the preset multiple to obtain an upsampled initial segmented image; If the tissue in the initially segmented image needs to be finely segmented into a first value of categories, the center of each category in the category grayscale vector is used as the initial category center for fine segmentation, and the membership matrix after upsampling the membership matrix by the preset multiple is used as the initial membership matrix for fine segmentation, and iterative clustering is performed on the histogram of the upsampled initially segmented image to perform fine segmentation on the tissue in the upsampled initially segmented image; If the tissue in the initially segmented image needs to be finely segmented into a second value number of categories, then the second value number of grayscale values ​​are taken at an average interval from the category grayscale vector as the initial category center for fine segmentation, the second value number of memberships are taken at an average interval from the range of the membership matrix as the initial membership matrix for fine segmentation, and iterative clustering is performed on the histogram of the upsampled initially segmented image to perform fine segmentation on the tissue in the upsampled initially segmented image.

8. The method according to claim 1 or 2, characterized in that: The template image is a standard image containing the target organ tissue segmentation result.

9. An organ segmentation device, characterized in that: The device comprises: A registration module, used for registering the template image and the image to be processed based on the mutual information between the template image and the image to be processed to obtain a registration result; the template image is determined according to any modality; A processing module, used for processing the target organ mask in the template image according to the registration result to obtain the target organ region in the image to be processed; The correction and segmentation module is used to perform non-uniform field correction on the target organ area of ​​the image to be processed based on the grayscale difference between tissues, and to roughly segment the tissue in the target organ to obtain a primary segmentation image of the target organ in the image to be processed; after finely segmenting the tissue in the primary segmentation image, the number of pixels of each category of tissue in the finely segmented image is obtained; according to the number of categories of tissue in the finely segmented image and the number of pixels of each category of tissue, the category of tissue to be retained is determined; the retained tissue categories are merged into one category, and the merged one category of tissue is morphologically repaired to obtain a three-dimensional mask image of the target organ in the image to be processed.

10. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 8 are implemented.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Images