Method and apparatus for segmenting brain tumor images
By generating multi-scale three-dimensional brain images and performing tilt correction, combined with convolutional neural networks and lesion discrimination models, the problem of low accuracy in brain tumor image segmentation was solved, and the accuracy of brain tumor image segmentation was improved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SHENZHEN INSTITUTE OF INFORMATION TECHNOLOGY
- Filing Date
- 2024-03-29
- Publication Date
- 2026-06-09
Smart Images
Figure CN118279584B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of medical imaging technology, and in particular to a method, apparatus, computer device, storage medium, and computer program product for segmenting brain tumor images. Background Technology
[0002] With the continuous advancement of medical imaging technology, obtaining tomographic images of internal body structures through computed tomography (CT) and positron emission tomography (PET) and using these images for disease diagnosis and treatment has become an important tool in modern medicine.
[0003] In related technologies, brain tumor image segmentation typically relies on experienced physicians manually identifying tumor regions in brain images. To improve the efficiency of brain tumor image segmentation, neural network-based segmentation schemes have been applied to this task. However, the current accuracy rate for brain tumor image segmentation is low, and there is an urgent need to improve it. Summary of the Invention
[0004] Therefore, it is necessary to provide a brain tumor image segmentation method, apparatus, computer equipment, storage medium, and computer program product that can improve the accuracy of brain tumor image segmentation in response to the above-mentioned technical problems.
[0005] In a first aspect, this application provides a method for segmenting brain tumor images, comprising:
[0006] Acquire a set of brain computed tomography (CT) images;
[0007] Generate at least two scales of three-dimensional brain images based on the brain tomography image set;
[0008] The midsagittal plane of each of the three-dimensional brain images was obtained;
[0009] Tilt correction is performed on the corresponding three-dimensional brain image based on the median sagittal plane;
[0010] Brain tumor images were extracted from tilt-corrected 3D brain images.
[0011] In one embodiment, acquiring the midsagittal plane of the at least two scales of three-dimensional brain images includes:
[0012] Based on the geometric center of the three-dimensional brain image, an initial sagittal plane is obtained;
[0013] Extract a predetermined number of pairs of three-dimensional image blocks with the initial sagittal plane as the plane of symmetry from the three-dimensional brain image;
[0014] Obtain the similarity of each pair of 3D image patches separately;
[0015] The symmetry of the initial sagittal plane is obtained based on the similarity of each pair of three-dimensional image blocks;
[0016] Based on the symmetry of the initial sagittal plane, the genetic algorithm and the Powell algorithm are used to obtain the extreme value of symmetry, and the sagittal plane corresponding to the extreme value of symmetry is determined as the midsagittal plane of the three-dimensional brain image.
[0017] In one embodiment, extracting a predetermined number of pairs of three-dimensional image blocks symmetrical about the initial sagittal plane from the three-dimensional brain image includes:
[0018] The preset number of three-dimensional image patches with the initial sagittal plane as the plane of symmetry are extracted on both sides of the initial sagittal plane using the Poisson sampling model.
[0019] In one embodiment, obtaining the similarity of each pair of 3D image patches includes:
[0020] The similarity of each pair of 3D image patches is obtained based on a similarity acquisition model;
[0021] The similarity acquisition model is a model obtained by training a multi-scale three-dimensional convolutional neural network model based on the first sample data. The first sample data includes multiple pairs of sample three-dimensional image blocks and the similarity of each pair of sample three-dimensional image blocks.
[0022] In one embodiment, the tilt correction based on the median sagittal plane of the at least two scales of the three-dimensional brain images includes:
[0023] Based on the midsagittal plane of the three-dimensional brain image and the expression of the sagittal plane in the three-dimensional coordinate system, the head and neck lateral flexion angle and head and neck rotation angle of the three-dimensional brain image are obtained.
[0024] The tilt correction of the three-dimensional brain image is performed based on the head and neck lateral flexion angle and the head and neck rotation angle.
[0025] In one embodiment, the extraction of brain tumor images based on the tilt-corrected three-dimensional brain images at at least two scales includes:
[0026] Based on the at least two scales of three-dimensional brain images and the corresponding image segmentation models at the at least two scales, candidate brain tumor regions in the at least two scales of three-dimensional brain images are obtained; the image segmentation model is a model obtained by training a three-dimensional convolutional neural network based on second sample data, the second sample data including: multiple sample three-dimensional brain images and masks of brain tumor regions in each sample three-dimensional brain image;
[0027] By fusing candidate brain tumor regions from at least two scales of three-dimensional brain images, an initial brain tumor image is obtained.
[0028] In one embodiment, after obtaining an initial brain tumor image by fusing candidate brain tumor regions from the at least two scales of three-dimensional brain images, the method further includes:
[0029] The initial brain tumor image is segmented into three-dimensional image blocks of a preset size;
[0030] The lesion discrimination model is used to determine whether each three-dimensional image block belongs to a brain tumor lesion. The lesion discrimination model is a model obtained by training the Visual Geometry Group (VGG) model based on third sample data. The third sample data includes multiple sample three-dimensional image blocks and the label information of each sample three-dimensional image block. The label information of the sample three-dimensional image block is used to indicate whether the sample three-dimensional image block belongs to a brain tumor lesion.
[0031] Combine three-dimensional image blocks belonging to brain tumor lesions to obtain the target brain tumor image.
[0032] Secondly, this application also provides a segmentation device for brain tumor images, comprising:
[0033] The acquisition module is used to acquire a set of brain computed tomography (CT) images.
[0034] A generation module is used to generate at least two scales of three-dimensional brain images based on the brain tomography image set;
[0035] The correction module is used to acquire the midsagittal plane of the at least two scales of the three-dimensional brain images, and to perform tilt correction on the at least two scales of the three-dimensional brain images based on the midsagittal plane of the at least two scales of the three-dimensional brain images.
[0036] A segmentation module is used to extract brain tumor images based on the tilt-corrected three-dimensional brain images at at least two scales.
[0037] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:
[0038] Acquire a set of brain computed tomography (CT) images;
[0039] Generate at least two scales of three-dimensional brain images based on the brain tomography image set;
[0040] The midsagittal plane of each of the three-dimensional brain images was obtained;
[0041] Tilt correction is performed on the corresponding three-dimensional brain image based on the median sagittal plane;
[0042] Brain tumor images were extracted from tilt-corrected 3D brain images.
[0043] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the following steps:
[0044] Acquire a set of brain computed tomography (CT) images;
[0045] Generate at least two scales of three-dimensional brain images based on the brain tomography image set;
[0046] The midsagittal plane of each of the three-dimensional brain images was obtained;
[0047] Tilt correction is performed on the corresponding three-dimensional brain image based on the median sagittal plane;
[0048] Brain tumor images were extracted from tilt-corrected 3D brain images.
[0049] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, performs the following steps:
[0050] Acquire a set of brain computed tomography (CT) images;
[0051] Generate at least two scales of three-dimensional brain images based on the brain tomography image set;
[0052] The midsagittal plane of each of the three-dimensional brain images was obtained;
[0053] Tilt correction is performed on the corresponding three-dimensional brain image based on the median sagittal plane;
[0054] Brain tumor images were extracted from tilt-corrected 3D brain images.
[0055] The aforementioned brain tumor image segmentation method, apparatus, computer equipment, storage medium, and computer program product, after acquiring a set of brain tomographic scan images, first generate at least two scales of three-dimensional brain images based on the brain tomographic scan image set. Then, the midsagittal plane of each of the three-dimensional brain images is acquired, and the corresponding three-dimensional brain images are tilt-corrected based on the midsagittal plane. Finally, the brain tumor image is extracted based on the tilt-corrected three-dimensional brain images. Because this embodiment first acquires the midsagittal plane of the three-dimensional brain image and performs tilt correction on the three-dimensional brain image based on the midsagittal plane before extracting the brain tumor image from the three-dimensional brain image, this embodiment can reduce or avoid the influence of the tilt angle of the three-dimensional brain image on the segmentation of the brain tumor image, thereby improving the accuracy of tumor image segmentation. Attached Figure Description
[0056] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0057] Figure 1 This is a flowchart of the steps of a brain tumor image segmentation method in one embodiment;
[0058] Figure 2 This is a schematic diagram of a brain tomography image set in one embodiment;
[0059] Figure 3 This is a flowchart of the steps of a brain tumor image segmentation method in another embodiment;
[0060] Figure 4 This is a flowchart of the steps of a brain tumor image segmentation method in another embodiment;
[0061] Figure 5 This is a structural block diagram of a brain tumor image segmentation device in one embodiment;
[0062] Figure 6 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation
[0063] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.
[0064] The tilt angle of brain CT scan images significantly impacts the accuracy of extracted brain tumor images. If the head is tilted during the acquisition of brain CT scan images, the acquired images will be asymmetrical, severely affecting the accuracy of brain tumor images obtained through convolutional neural networks. Given that the tilt angle of the three-dimensional brain image affects the accuracy of the extracted brain tumor image, this embodiment of the application, after acquiring a set of brain CT scan images and generating at least two scales of three-dimensional brain images based on the set, first acquires the midsagittal plane of each of the three-dimensional brain images, and then performs tilt correction on the corresponding three-dimensional brain images based on the midsagittal plane. Finally, the brain tumor image is extracted based on the tilt-corrected three-dimensional brain image. Therefore, this embodiment of the application can reduce or avoid the impact of the tilt angle of the three-dimensional brain image on the segmentation of brain tumor images, thereby improving the accuracy of tumor image segmentation.
[0065] The execution entity of the brain tumor image segmentation method provided in this application embodiment can be a brain tumor image segmentation device. This brain tumor image segmentation device includes, but is not limited to, various personal computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, smart vehicle devices, etc. Portable wearable devices can include smartwatches, smart bracelets, head-mounted devices, etc.
[0066] In one exemplary embodiment, such as Figure 1 As shown, a method for segmenting brain tumor images is provided, which includes the following steps 101 to 105:
[0067] Step 101: Obtain a set of brain computed tomography (CT) images.
[0068] The brain computed tomography image set in this application embodiment may include at least one of the following: an image set obtained by scanning the brain using computed tomography (CT) technology, an image set obtained by scanning the brain using positron emission tomography (PET) technology, an image set obtained by scanning the brain using nuclear magnetic resonance imaging (NMR) technology, and an image set obtained by scanning the brain using single photon emission computed tomography (SPECT) technology.
[0069] For example, refer to Figure 2As shown, the brain tomography image set 200 includes multiple brain tomography images 201, which are scan images of different locations of the brain. Figure 2 The example shown is a brain tomography image set 200 containing 20 brain tomography images 201. However, the embodiments of this application are not limited to this. In some embodiments, the spacing when performing tomography scans on the brain can be reduced, thereby increasing the number of brain tomography images in the brain tomography image set. Alternatively, the spacing when performing tomography scans on the brain can be increased, thereby reducing the number of brain tomography images in the brain tomography image set.
[0070] Step 102: Generate at least two scales of three-dimensional brain images based on the brain tomography image set.
[0071] In some embodiments, generating at least two scales of three-dimensional brain images based on the brain computed tomography image set includes the following steps 102a and 102b:
[0072] Step 102a: Scale the brain tomography images in the brain tomography image set to obtain at least two preprocessed image sets.
[0073] Brain tomography images within the same preprocessed image set have the same scale, while brain tomography images within different preprocessed image sets have different scales.
[0074] For example, if the resolution of the brain tomography images in the brain tomography image set obtained in step 101 is 1024*1024, then each brain tomography image in the brain tomography image set can be downsampled to an image with a resolution of 768*768 to obtain a first preprocessed image set, each brain tomography image in the brain tomography image set can be downsampled to an image with a resolution of 512*512 to obtain a second preprocessed image set, and the brain tomography image set can be used as a third preprocessed image set, for a total of three preprocessed image sets.
[0075] Step 102b: Construct three-dimensional brain images based on the at least two preprocessed image sets to obtain three-dimensional brain images at the at least two scales.
[0076] Continuing with the example above, the at least two preprocessed image sets include: a first preprocessed image set where each image has a resolution of 512*512, a second preprocessed image set where each image has a resolution of 768*768, and a third preprocessed image set where each image has a resolution of 1024*1024. Then, brain 3D images are constructed based on the at least two preprocessed image sets to obtain brain 3D images at at least two scales, including:
[0077] A first three-dimensional brain image is obtained by constructing a three-dimensional brain image based on a first preprocessed image set, a second three-dimensional brain image is obtained by constructing a three-dimensional brain image based on a second preprocessed image set, and a third three-dimensional brain image is obtained by constructing a three-dimensional brain image based on a third preprocessed image set.
[0078] Since the resolution of each image in the first preprocessed image set is 512*512, the resolution of the first three-dimensional brain image can be 512*512*512. Since the resolution of each image in the second preprocessed image set is 768*768, the resolution of the second three-dimensional brain image can be 768*768*768. Since the resolution of each image in the third preprocessed image set is 1024*1024, the resolution of the third three-dimensional brain image can be 1024*1024*1024.
[0079] In some embodiments, constructing a three-dimensional image of the brain based on a preprocessed image set includes: constructing a three-dimensional image of the brain based on surface reconstruction. The construction of a three-dimensional image of the brain based on surface reconstruction is a method of describing the structure of a three-dimensional object by geometrically stitching together and fitting the object's surface. This may include: firstly, using image segmentation techniques to segment the contour curves of the brain structure from each image in the preprocessed image set, and then performing image processing to obtain the three-dimensional structure of the brain, thereby acquiring a three-dimensional image of the brain.
[0080] In some embodiments, constructing a three-dimensional image of the brain based on a preprocessed image set includes: constructing the three-dimensional image of the brain based on voxel reconstruction. The method of constructing the three-dimensional image of the brain based on voxel reconstruction is a method of projecting volume pixels onto a display plane with certain colors and transparency, and may include: generating a three-dimensional volume data set based on the preprocessed image set, and reconstructing the three-dimensional image of the brain based on the three-dimensional volume data set.
[0081] Step 103: Obtain the midsagittal plane of each of the three-dimensional brain images.
[0082] The sagittal plane is an anatomical term, specifically referring to the longitudinal section that divides the human body into left and right halves. The median sagittal plane, on the other hand, is the sagittal plane that divides the anatomical object into two symmetrical parts. The median sagittal plane of the brain, specifically, is the sagittal plane that divides the brain into the left and right hemispheres.
[0083] In some embodiments, step 103 (obtaining the midsagittal plane of a three-dimensional image of the brain) is implemented in the following ways:
[0084] Step 103a: Obtain the initial sagittal plane based on the geometric center of the three-dimensional brain image.
[0085] That is, the sagittal plane passing through the center of gravity of the brain is determined as the initial sagittal plane.
[0086] Step 103b: Extract a preset number of three-dimensional image blocks from the three-dimensional brain image, with the initial sagittal plane as the plane of symmetry.
[0087] In some embodiments, extracting a predetermined number of pairs of three-dimensional image patches symmetrical about the initial sagittal plane from the three-dimensional brain image includes:
[0088] The preset number of three-dimensional image patches with the initial sagittal plane as the plane of symmetry are extracted on both sides of the initial sagittal plane using the Poisson sampling model.
[0089] Specifically, the Poisson sampling model is a statistical model used to describe the probability distribution of events occurring within a given time or space. The Poisson sampling model assumes that events occur independently and that their probability of occurrence is the same at any given time or space.
[0090] By using the Poisson sampling model to extract a preset number of three-dimensional image patches on both sides of the initial sagittal plane, with the initial sagittal plane as the plane of symmetry, the distribution of the extracted three-dimensional image patches in the three-dimensional brain image can be made more uniform.
[0091] Step 103c: Obtain the similarity of each pair of 3D image blocks.
[0092] In some embodiments, obtaining the similarity of each pair of 3D image patches includes:
[0093] The similarity of each pair of 3D image patches is obtained based on a similarity acquisition model;
[0094] The similarity acquisition model is a model obtained by training a multi-scale three-dimensional convolutional neural network model based on the first sample data. The first sample data includes multiple pairs of sample three-dimensional image blocks and the similarity of each pair of sample three-dimensional image blocks.
[0095] In some embodiments, the training process of the similarity acquisition model may include: first, inputting a pair of sample 3D image blocks into a 3D convolutional neural network model and obtaining the similarity of the pair of sample 3D image blocks predicted by the 3D convolutional neural network model; then, calculating a loss value based on the similarity of the pair of sample 3D image blocks predicted by the 3D convolutional neural network model, the similarity of the pair of sample 3D image blocks, and a preset loss function; and adjusting the model parameters predicted by the 3D convolutional neural network model based on the loss value.
[0096] Step 103d: Obtain the symmetry of the initial sagittal plane based on the similarity of each pair of three-dimensional image blocks.
[0097] In some embodiments, obtaining the symmetry of the initial sagittal plane based on the similarity of each pair of three-dimensional image blocks includes: calculating the sum of the similarities of each pair of three-dimensional image blocks to obtain the symmetry of the initial sagittal plane.
[0098] That is, the symmetry of the initial sagittal plane is denoted as S, and the similarity of the i-th pair of three-dimensional image patches is denoted as s. i Then we have:
[0099]
[0100] Step 104e: Based on the symmetry of the initial sagittal plane, the genetic algorithm and the Powell algorithm are used to obtain the extreme value of symmetry, and the sagittal plane corresponding to the extreme value of symmetry is determined as the midsagittal plane of the three-dimensional brain image.
[0101] In some embodiments, the extreme value of symmetry is obtained based on the symmetry of the initial sagittal plane, a genetic algorithm, and a Powell algorithm, and the sagittal plane corresponding to the extreme value of symmetry is determined as the midsagittal plane of the three-dimensional brain image, including:
[0102] A genetic algorithm is used to coarsely locate the median sagittal plane of the three-dimensional brain image, and then the Powell algorithm is used to optimize the median sagittal plane of the three-dimensional brain image to obtain the median sagittal plane of the three-dimensional brain image.
[0103] The expression for the midsagittal plane π in a three-dimensional brain image in a three-dimensional coordinate system is:
[0104] π = Ax + By + Cz + D
[0105] A = cos(α) + cos(β)
[0106] B = sin(α) + cos(β)
[0107] C = sin(α)
[0108] Where α is the angle between the normal vector of the median sagittal plane π and the xy plane, β is the angle between the projection of the median sagittal plane π onto the xy plane and the x-axis, and D is the distance from the origin of the three-dimensional coordinate system to the median sagittal plane π.
[0109] When using a genetic algorithm to coarsely locate the median sagittal plane of the three-dimensional brain image and using the Powell algorithm to optimize the median sagittal plane of the three-dimensional brain image, α, β, and D can be used as parameters to calculate the extreme value of the symmetry of the sagittal plane, and the optimization stops when the symmetry of the sagittal plane reaches the maximum value.
[0110] Step 104: Perform tilt correction on the corresponding three-dimensional brain image based on the median sagittal plane.
[0111] In some embodiments, tilt correction is performed on the three-dimensional brain image based on the median sagittal plane, including:
[0112] Step 104a: Based on the midsagittal plane of the three-dimensional brain image and the expression of the sagittal plane in the three-dimensional coordinate system, obtain the head and neck lateral flexion angle and head and neck rotation angle of the three-dimensional brain image.
[0113] As shown above, the expression of the sagittal plane in the three-dimensional coordinate system is an equation with α, β and D as parameters. α is the angle between the normal vector of the median sagittal plane π and the xy plane, and β is the angle between the projection of the median sagittal plane π onto the xy plane and the x-axis. Therefore, α is the lateral flexion angle of the head and neck, and β is the rotation angle of the head and neck.
[0114] Step 104b: The three-dimensional image of the brain is tilted according to the head and neck lateral flexion angle and the head and neck rotation angle.
[0115] That is, rotate the three-dimensional image of the brain by -α degrees so that the normal vector of the median sagittal plane is parallel to the xy plane, and rotate the three-dimensional image of the brain by -β degrees so that the projection of the median sagittal plane onto the xy plane is parallel to the x-axis.
[0116] Step 105: Extract brain tumor images based on the tilt-corrected 3D brain images.
[0117] The brain tumor image segmentation method provided in this application, after acquiring a set of brain tomographic images, first generates at least two scales of three-dimensional brain images based on the brain tomographic image set. Then, it acquires the midsagittal plane of each of the three-dimensional brain images and performs tilt correction on the corresponding three-dimensional brain images based on the midsagittal plane. Finally, it extracts the brain tumor image based on the tilt-corrected three-dimensional brain images. Because this application first acquires the midsagittal plane of the three-dimensional brain image and performs tilt correction on the three-dimensional brain image based on the midsagittal plane before extracting the brain tumor image from the three-dimensional brain image, this application can reduce or avoid the influence of the tilt angle of the three-dimensional brain image on the segmentation of the brain tumor image, thereby improving the accuracy of tumor image segmentation.
[0118] As an extension and refinement of the above embodiments, refer to Figure 3 As shown, in an exemplary embodiment, such as Figure 3 As shown, the brain tumor image segmentation method provided in the above embodiment includes the following steps 301 to 310:
[0119] Step 301: Obtain a set of brain computed tomography (CT) images.
[0120] Step 302: Scale the brain tomography images in the brain tomography image set to obtain at least two preprocessed image sets.
[0121] Brain tomography images within the same preprocessed image set have the same scale, while brain tomography images within different preprocessed image sets have different scales.
[0122] Step 303: Construct three-dimensional brain images based on the at least two preprocessed image sets to obtain three-dimensional brain images at at least two scales.
[0123] Step 304: Obtain the midsagittal plane of each of the three-dimensional brain images.
[0124] Step 305: Perform tilt correction on the corresponding three-dimensional brain images based on the median sagittal plane.
[0125] Step 306: Obtain brain tumor candidate regions in the brain three-dimensional images at the at least two scales based on the brain three-dimensional images at the at least two scales and the image segmentation models corresponding to the at least two scales, respectively.
[0126] The image segmentation model is a model obtained by training a three-dimensional convolutional neural network based on second sample data. The second sample data includes: multiple sample three-dimensional brain images and masks of brain tumor regions in each sample three-dimensional brain image.
[0127] Step 307: Obtain the overlap information of brain tumor candidate regions in the at least two scales of three-dimensional brain images.
[0128] Step 308: Based on the overlap information of the brain tumor candidate regions in the at least two scales of three-dimensional brain images, determine the brain tumor candidate regions belonging to the same region in the at least two scales of three-dimensional brain images.
[0129] In some embodiments, determining brain tumor candidate regions belonging to the same region in the at least two scales of three-dimensional brain images based on overlap information of brain tumor candidate regions in the at least two scales includes: determining whether the overlap rate of a first brain tumor candidate region and a second brain tumor candidate region in the at least two scales of three-dimensional brain images is greater than a threshold overlap rate; if so, determining that the first brain tumor candidate region and the second brain tumor candidate region belong to the same region. Wherein, the first brain tumor candidate region and the second brain tumor candidate region are brain tumor candidate regions in three-dimensional brain images of different scales.
[0130] Step 309: Obtain the initial brain tumor image based on the confidence level of the candidate brain tumor regions belonging to the same region.
[0131] In some embodiments, the initial brain tumor image is obtained based on the confidence level of the candidate brain tumor regions belonging to the same region, including the following steps 309a to 309d:
[0132] Step 309a: Obtain the confidence level of the region based on the confidence level of the candidate brain tumor regions belonging to the same region.
[0133] The confidence level of the region is the sum of the confidence levels of each brain tumor candidate region belonging to the region.
[0134] For example, if brain tumor candidate region A in a first-scale three-dimensional brain image, brain tumor candidate region B in a second-scale three-dimensional brain image, and brain tumor candidate region C in a third-scale three-dimensional brain image belong to the same region, and the confidence scores of brain tumor candidate region A, brain tumor candidate region B, and brain tumor candidate region C are x, y, and z respectively, then the confidence score of the region is x+y+z.
[0135] Step 309b: Determine whether the confidence level of the region is greater than a preset threshold.
[0136] In step 309b above, if the confidence level of the region is less than the preset threshold, the probability that the region is a brain tumor region is low, so the region is not identified as a brain tumor region. However, if the confidence level of the region is greater than the preset threshold, the probability that the region is a brain tumor region is high, so step 309c is executed as follows:
[0137] Step 309c: Determine that the area is a brain tumor area.
[0138] Step 309d: Extract the image corresponding to the brain tumor region to obtain the initial brain tumor image.
[0139] In some embodiments, extracting the image corresponding to the brain tumor region includes: extracting the image corresponding to the brain tumor region from the largest (highest resolution) three-dimensional brain image among the at least two scales of three-dimensional brain images.
[0140] It should be noted that, since this application involves extracting brain tumor images from three-dimensional brain images, the initial brain tumor images obtained are also three-dimensional images.
[0141] In some embodiments, refer to Figure 4 As shown above, in the above Figure 3 Based on the illustrated embodiments, the brain tumor image segmentation method provided in this application further includes the following steps 401 to 403:
[0142] Step 401: Segment the initial brain tumor image into three-dimensional image blocks of a preset size.
[0143] In some embodiments, the initial brain tumor image can be divided into three-dimensional cubic image blocks with a size of 1.0mm*1.0mm*1.0mm.
[0144] Step 402: Determine whether each three-dimensional image block belongs to a brain tumor lesion.
[0145] In some embodiments, determining whether each three-dimensional image block belongs to a brain tumor lesion includes:
[0146] Based on the lesion discrimination model, each three-dimensional image block is determined to be a brain tumor lesion.
[0147] The lesion discrimination model is a model obtained by training a Visual Geometry Group (VGG) model based on third sample data. The third sample data includes multiple sample three-dimensional image patches and label information of each sample three-dimensional image patch. The label information of the sample three-dimensional image patch is used to indicate whether the sample three-dimensional image patch belongs to a brain tumor lesion.
[0148] In some embodiments, a multi-scale lesion discrimination model can also be established to accurately segment brain tumor images output by image segmentation models at different scales, and finally, image blocks suspected of being brain cancer can be determined through ensemble learning.
[0149] Step 403: Combine the three-dimensional image blocks belonging to the brain tumor lesion to obtain the target brain tumor image.
[0150] Steps 401 to 403 above can further refine the initial brain tumor image to obtain a more accurate target brain tumor image.
[0151] Furthermore, compared to directly segmenting the three-dimensional image of the brain into multiple three-dimensional image blocks and then determining whether each three-dimensional image block belongs to a brain tumor lesion, the above embodiment first obtains an initial brain tumor image and then segments the three-dimensional image of the brain into multiple three-dimensional image blocks. Therefore, the above embodiment can reduce the number of three-dimensional image blocks that need to be determined, thereby reducing the amount of computation in the process of extracting brain tumor images and improving the efficiency of brain tumor image extraction.
[0152] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.
[0153] Based on the same inventive concept, this application also provides a brain tumor image segmentation apparatus for implementing the brain tumor image segmentation method described above. The solution provided by this apparatus is similar to the implementation described in the above method; therefore, the specific limitations in one or more brain tumor image segmentation apparatus embodiments provided below can be found in the limitations of the brain tumor image segmentation method described above, and will not be repeated here.
[0154] In one exemplary embodiment, such as Figure 5 As shown, a brain tumor image segmentation device is provided, comprising: an acquisition module 51, a generation module 52, a correction module 53, and a segmentation module 54. Wherein:
[0155] The acquisition module 51 is used to acquire a set of brain tomographic scan images.
[0156] The generation module 52 is used to generate at least two scales of three-dimensional brain images based on the brain tomography image set;
[0157] The correction module 53 is used to acquire the midsagittal plane of the at least two scales of the three-dimensional brain images respectively, and to perform tilt correction on the at least two scales of the three-dimensional brain images based on the midsagittal plane of the at least two scales of the three-dimensional brain images respectively.
[0158] The segmentation module 54 is used to extract brain tumor images based on the tilt-corrected three-dimensional brain images at at least two scales.
[0159] In some embodiments, the correction module 53 is specifically used to obtain an initial sagittal plane based on the geometric center of the three-dimensional brain image; extract a preset number of pairs of three-dimensional image blocks from the three-dimensional brain image that are symmetrical about the initial sagittal plane; obtain the similarity of each pair of three-dimensional image blocks; obtain the symmetry of the initial sagittal plane based on the similarity of each pair of three-dimensional image blocks; obtain the extreme value of the symmetry based on the symmetry of the initial sagittal plane, a genetic algorithm, and a Powell algorithm, and determine the sagittal plane corresponding to the extreme value of the symmetry as the median sagittal plane of the three-dimensional brain image.
[0160] In some embodiments, the correction module 53 is specifically used to extract the preset number of three-dimensional image blocks with the initial sagittal plane as the plane of symmetry on both sides of the initial sagittal plane using a Poisson sampling model.
[0161] In some embodiments, the correction module 53 is specifically used to obtain the similarity of each pair of three-dimensional image patches based on a similarity acquisition model;
[0162] The similarity acquisition model is a model obtained by training a multi-scale three-dimensional convolutional neural network model based on the first sample data. The first sample data includes multiple pairs of sample three-dimensional image blocks and the similarity of each pair of sample three-dimensional image blocks.
[0163] In some embodiments, the correction module 53 is specifically used to obtain the head and neck lateral flexion angle and head and neck rotation angle of the three-dimensional brain image based on the midsagittal plane of the three-dimensional brain image and the expression of the sagittal plane in the three-dimensional coordinate system; and to perform tilt correction on the three-dimensional brain image based on the head and neck lateral flexion angle and the head and neck rotation angle.
[0164] In some embodiments, the segmentation module 54 is specifically used to obtain brain tumor candidate regions in the at least two scales of the three-dimensional brain images and the corresponding image segmentation models at the at least two scales; and to fuse the brain tumor candidate regions in the at least two scales of the three-dimensional brain images to obtain an initial brain tumor image; wherein the image segmentation model is a model obtained by training a three-dimensional convolutional neural network based on second sample data, and the second sample data includes: multiple sample three-dimensional brain images and masks of brain tumor regions in each sample three-dimensional brain image.
[0165] In some embodiments, the segmentation module 54 is further configured to, after fusing candidate brain tumor regions from at least two scales of three-dimensional brain images to obtain an initial brain tumor image, segment the initial brain tumor image into three-dimensional image blocks of a preset size; determine whether each three-dimensional image block belongs to a brain tumor lesion based on a lesion discrimination model; and combine the three-dimensional image blocks belonging to the brain tumor lesion to obtain a target brain tumor image. The lesion discrimination model is a model obtained by training a Visual Geometry Group (VGG) model based on third sample data. The third sample data includes multiple sample three-dimensional image blocks and label information for each sample three-dimensional image block. The label information of the sample three-dimensional image blocks is used to indicate whether the sample three-dimensional image block belongs to a brain tumor lesion.
[0166] Each module in the aforementioned brain tumor image segmentation device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the operations corresponding to each module.
[0167] In one exemplary embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 6As shown, the computer device includes a processor, memory, input / output interface, communication interface, display unit, and input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interface. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The input / output interface is used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, NFC (Near Field Communication), or other technologies. When the computer program is executed by the processor, it implements the brain tumor image segmentation method provided in the above embodiment. The display unit of the computer device forms a visually visible image and can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the casing of the computer device, or external keyboards, touchpads, or mice, etc.
[0168] Those skilled in the art will understand that Figure 6 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.
[0169] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:
[0170] Acquire a set of brain computed tomography (CT) images;
[0171] Generate at least two scales of three-dimensional brain images based on the brain tomography image set;
[0172] The midsagittal plane of each of the three-dimensional brain images was obtained;
[0173] Tilt correction is performed on the corresponding three-dimensional brain image based on the median sagittal plane;
[0174] Brain tumor images were extracted from tilt-corrected 3D brain images.
[0175] In one embodiment, the processor, when executing a computer program, also performs the following steps:
[0176] Based on the geometric center of the three-dimensional brain image, an initial sagittal plane is obtained;
[0177] Extract a predetermined number of pairs of three-dimensional image blocks with the initial sagittal plane as the plane of symmetry from the three-dimensional brain image;
[0178] Obtain the similarity of each pair of 3D image patches separately;
[0179] The symmetry of the initial sagittal plane is obtained based on the similarity of each pair of three-dimensional image blocks;
[0180] Based on the symmetry of the initial sagittal plane, the genetic algorithm and the Powell algorithm are used to obtain the extreme value of symmetry, and the sagittal plane corresponding to the extreme value of symmetry is determined as the midsagittal plane of the three-dimensional brain image.
[0181] In one embodiment, the processor, when executing a computer program, also performs the following steps:
[0182] The preset number of three-dimensional image patches with the initial sagittal plane as the plane of symmetry are extracted on both sides of the initial sagittal plane using the Poisson sampling model.
[0183] In one embodiment, the processor, when executing a computer program, also performs the following steps:
[0184] The similarity of each pair of 3D image patches is obtained based on a similarity acquisition model;
[0185] The similarity acquisition model is a model obtained by training a multi-scale three-dimensional convolutional neural network model based on the first sample data. The first sample data includes multiple pairs of sample three-dimensional image blocks and the similarity of each pair of sample three-dimensional image blocks.
[0186] In one embodiment, the processor, when executing a computer program, also performs the following steps:
[0187] Based on the midsagittal plane of the three-dimensional brain image and the expression of the sagittal plane in the three-dimensional coordinate system, the head and neck lateral flexion angle and head and neck rotation angle of the three-dimensional brain image are obtained.
[0188] The tilt correction of the three-dimensional brain image is performed based on the head and neck lateral flexion angle and the head and neck rotation angle.
[0189] In one embodiment, the processor, when executing a computer program, also performs the following steps:
[0190] Based on the at least two scales of three-dimensional brain images and the corresponding image segmentation models at the at least two scales, candidate brain tumor regions in the at least two scales of three-dimensional brain images are obtained; the image segmentation model is a model obtained by training a three-dimensional convolutional neural network based on second sample data, the second sample data including: multiple sample three-dimensional brain images and masks of brain tumor regions in each sample three-dimensional brain image;
[0191] By fusing candidate brain tumor regions from at least two scales of three-dimensional brain images, an initial brain tumor image is obtained.
[0192] In one embodiment, the processor, when executing a computer program, also performs the following steps:
[0193] The initial brain tumor image is segmented into three-dimensional image blocks of a preset size;
[0194] The lesion discrimination model is used to determine whether each three-dimensional image block belongs to a brain tumor lesion. The lesion discrimination model is a model obtained by training the Visual Geometry Group (VGG) model based on third sample data. The third sample data includes multiple sample three-dimensional image blocks and the label information of each sample three-dimensional image block. The label information of the sample three-dimensional image block is used to indicate whether the sample three-dimensional image block belongs to a brain tumor lesion.
[0195] Combine three-dimensional image blocks belonging to brain tumor lesions to obtain the target brain tumor image.
[0196] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor:
[0197] Acquire a set of brain computed tomography (CT) images;
[0198] Generate at least two scales of three-dimensional brain images based on the brain tomography image set;
[0199] The midsagittal plane of each of the three-dimensional brain images was obtained;
[0200] Tilt correction is performed on the corresponding three-dimensional brain image based on the median sagittal plane;
[0201] Brain tumor images were extracted from tilt-corrected 3D brain images.
[0202] In one embodiment, when the computer program is executed by a processor, it further performs the following steps:
[0203] Based on the geometric center of the three-dimensional brain image, an initial sagittal plane is obtained;
[0204] Extract a predetermined number of pairs of three-dimensional image blocks with the initial sagittal plane as the plane of symmetry from the three-dimensional brain image;
[0205] Obtain the similarity of each pair of 3D image patches separately;
[0206] The symmetry of the initial sagittal plane is obtained based on the similarity of each pair of three-dimensional image blocks;
[0207] Based on the symmetry of the initial sagittal plane, the genetic algorithm and the Powell algorithm are used to obtain the extreme value of symmetry, and the sagittal plane corresponding to the extreme value of symmetry is determined as the midsagittal plane of the three-dimensional brain image.
[0208] In one embodiment, when the computer program is executed by a processor, it further performs the following steps:
[0209] The preset number of three-dimensional image patches with the initial sagittal plane as the plane of symmetry are extracted on both sides of the initial sagittal plane using the Poisson sampling model.
[0210] In one embodiment, the processor, when executing a computer program, also performs the following steps:
[0211] The similarity of each pair of 3D image patches is obtained based on a similarity acquisition model;
[0212] The similarity acquisition model is a model obtained by training a multi-scale three-dimensional convolutional neural network model based on the first sample data. The first sample data includes multiple pairs of sample three-dimensional image blocks and the similarity of each pair of sample three-dimensional image blocks.
[0213] In one embodiment, when the computer program is executed by a processor, it further performs the following steps:
[0214] Based on the midsagittal plane of the three-dimensional brain image and the expression of the sagittal plane in the three-dimensional coordinate system, the head and neck lateral flexion angle and head and neck rotation angle of the three-dimensional brain image are obtained.
[0215] The tilt correction of the three-dimensional brain image is performed based on the head and neck lateral flexion angle and the head and neck rotation angle.
[0216] In one embodiment, when the computer program is executed by a processor, it further performs the following steps:
[0217] Based on the at least two scales of three-dimensional brain images and the corresponding image segmentation models at the at least two scales, candidate brain tumor regions in the at least two scales of three-dimensional brain images are obtained; the image segmentation model is a model obtained by training a three-dimensional convolutional neural network based on second sample data, the second sample data including: multiple sample three-dimensional brain images and masks of brain tumor regions in each sample three-dimensional brain image;
[0218] By fusing candidate brain tumor regions from at least two scales of three-dimensional brain images, an initial brain tumor image is obtained.
[0219] In one embodiment, when the computer program is executed by a processor, it further performs the following steps:
[0220] The initial brain tumor image is segmented into three-dimensional image blocks of a preset size;
[0221] The lesion discrimination model is used to determine whether each three-dimensional image block belongs to a brain tumor lesion. The lesion discrimination model is a model obtained by training the Visual Geometry Group (VGG) model based on third sample data. The third sample data includes multiple sample three-dimensional image blocks and the label information of each sample three-dimensional image block. The label information of the sample three-dimensional image block is used to indicate whether the sample three-dimensional image block belongs to a brain tumor lesion.
[0222] Combine three-dimensional image blocks belonging to brain tumor lesions to obtain the target brain tumor image.
[0223] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, performs the following steps:
[0224] Acquire a set of brain computed tomography (CT) images;
[0225] Generate at least two scales of three-dimensional brain images based on the brain tomography image set;
[0226] The midsagittal plane of each of the three-dimensional brain images was obtained;
[0227] Tilt correction is performed on the corresponding three-dimensional brain image based on the median sagittal plane;
[0228] Brain tumor images were extracted from tilt-corrected 3D brain images.
[0229] In one embodiment, when the computer program is executed by a processor, it further performs the following steps:
[0230] Based on the geometric center of the three-dimensional brain image, an initial sagittal plane is obtained;
[0231] Extract a predetermined number of pairs of three-dimensional image blocks with the initial sagittal plane as the plane of symmetry from the three-dimensional brain image;
[0232] Obtain the similarity of each pair of 3D image patches separately;
[0233] The symmetry of the initial sagittal plane is obtained based on the similarity of each pair of three-dimensional image blocks;
[0234] Based on the symmetry of the initial sagittal plane, the genetic algorithm and the Powell algorithm are used to obtain the extreme value of symmetry, and the sagittal plane corresponding to the extreme value of symmetry is determined as the midsagittal plane of the three-dimensional brain image.
[0235] In one embodiment, when the computer program is executed by a processor, it further performs the following steps:
[0236] The preset number of three-dimensional image patches with the initial sagittal plane as the plane of symmetry are extracted on both sides of the initial sagittal plane using the Poisson sampling model.
[0237] In one embodiment, the processor, when executing a computer program, also performs the following steps:
[0238] The similarity of each pair of 3D image patches is obtained based on a similarity acquisition model;
[0239] The similarity acquisition model is a model obtained by training a multi-scale three-dimensional convolutional neural network model based on the first sample data. The first sample data includes multiple pairs of sample three-dimensional image blocks and the similarity of each pair of sample three-dimensional image blocks.
[0240] In one embodiment, when the computer program is executed by a processor, it further performs the following steps:
[0241] Based on the midsagittal plane of the three-dimensional brain image and the expression of the sagittal plane in the three-dimensional coordinate system, the head and neck lateral flexion angle and head and neck rotation angle of the three-dimensional brain image are obtained.
[0242] The tilt correction of the three-dimensional brain image is performed based on the head and neck lateral flexion angle and the head and neck rotation angle.
[0243] In one embodiment, when the computer program is executed by a processor, it further performs the following steps:
[0244] Based on the at least two scales of three-dimensional brain images and the corresponding image segmentation models at the at least two scales, candidate brain tumor regions in the at least two scales of three-dimensional brain images are obtained; the image segmentation model is a model obtained by training a three-dimensional convolutional neural network based on second sample data, the second sample data including: multiple sample three-dimensional brain images and masks of brain tumor regions in each sample three-dimensional brain image;
[0245] By fusing candidate brain tumor regions from at least two scales of three-dimensional brain images, an initial brain tumor image is obtained.
[0246] In one embodiment, when the computer program is executed by a processor, it further performs the following steps:
[0247] The initial brain tumor image is segmented into three-dimensional image blocks of a preset size;
[0248] The lesion discrimination model is used to determine whether each three-dimensional image block belongs to a brain tumor lesion. The lesion discrimination model is a model obtained by training the Visual Geometry Group (VGG) model based on third sample data. The third sample data includes multiple sample three-dimensional image blocks and the label information of each sample three-dimensional image block. The label information of the sample three-dimensional image block is used to indicate whether the sample three-dimensional image block belongs to a brain tumor lesion.
[0249] Combine three-dimensional image blocks belonging to brain tumor lesions to obtain the target brain tumor image.
[0250] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, 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, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.
[0251] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0252] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.
Claims
1. A method for segmenting brain tumor images, characterized in that, The method includes: Acquire a set of brain computed tomography (CT) images; Generate at least two scales of three-dimensional brain images based on the brain tomography image set; The midsagittal plane of the at least two scales of the brain three-dimensional images is obtained respectively; Tilt correction is performed on the at least two scales of the three-dimensional brain images based on the midsagittal plane of the at least two scales of the three-dimensional brain images; Brain tumor images were extracted based on the tilt-corrected three-dimensional brain images at at least two scales. The step of obtaining the median sagittal plane of the at least two scales of the three-dimensional brain images includes: obtaining an initial sagittal plane based on the geometric center of the three-dimensional brain images; extracting a predetermined number of pairs of three-dimensional image blocks from the three-dimensional brain images that are symmetrical about the initial sagittal plane; obtaining the similarity of each pair of three-dimensional image blocks; obtaining the symmetry of the initial sagittal plane based on the similarity of each pair of three-dimensional image blocks; obtaining the extreme value of the symmetry based on the symmetry of the initial sagittal plane, a genetic algorithm, and a Powell algorithm, and determining the sagittal plane corresponding to the extreme value of the symmetry as the median sagittal plane of the three-dimensional brain images. The tilt correction based on the median sagittal plane of the at least two scales of the three-dimensional brain images includes: obtaining the head and neck lateral flexion angle and head and neck rotation angle of the three-dimensional brain images based on the median sagittal plane of the at least two scales of the three-dimensional brain images and the expression of the sagittal plane in the three-dimensional coordinate system; and performing tilt correction on the three-dimensional brain images based on the head and neck lateral flexion angle and the head and neck rotation angle.
2. The method according to claim 1, characterized in that, The step of extracting a predetermined number of pairs of three-dimensional image blocks from the three-dimensional brain image, with the initial sagittal plane as the plane of symmetry, includes: The preset number of three-dimensional image patches with the initial sagittal plane as the plane of symmetry are extracted on both sides of the initial sagittal plane using the Poisson sampling model.
3. The method according to claim 1, characterized in that, The step of obtaining the similarity of each pair of 3D image patches includes: The similarity of each pair of 3D image patches is obtained based on a similarity acquisition model; The similarity acquisition model is a model obtained by training a multi-scale three-dimensional convolutional neural network model based on the first sample data. The first sample data includes multiple pairs of sample three-dimensional image blocks and the similarity of each pair of sample three-dimensional image blocks.
4. The method according to claim 1, characterized in that, The extraction of brain tumor images based on the tilt-corrected, at least two-scale, three-dimensional brain images includes: Based on the at least two scales of three-dimensional brain images and the corresponding image segmentation models at the at least two scales, candidate brain tumor regions in the at least two scales of three-dimensional brain images are obtained; the image segmentation model is a model obtained by training a three-dimensional convolutional neural network based on second sample data, the second sample data including: multiple sample three-dimensional brain images and masks of brain tumor regions in each sample three-dimensional brain image; By fusing candidate brain tumor regions from at least two scales of three-dimensional brain images, an initial brain tumor image is obtained.
5. The method according to claim 4, characterized in that, After obtaining an initial brain tumor image by fusing candidate brain tumor regions from at least two scales of three-dimensional brain images, the method further includes: The initial brain tumor image is segmented into three-dimensional image blocks of a preset size; The lesion discrimination model is used to determine whether each three-dimensional image block belongs to a brain tumor lesion. The lesion discrimination model is a model obtained by training the Visual Geometry Group (VGG) model based on third sample data. The third sample data includes multiple sample three-dimensional image blocks and the label information of each sample three-dimensional image block. The label information of the sample three-dimensional image block is used to indicate whether the sample three-dimensional image block belongs to a brain tumor lesion. Combine three-dimensional image blocks belonging to brain tumor lesions to obtain the target brain tumor image.
6. A segmentation device for brain tumor images, characterized in that, include: The acquisition module is used to acquire a set of brain computed tomography (CT) images. A generation module is used to generate at least two scales of three-dimensional brain images based on the brain tomography image set; The correction module is used to acquire the midsagittal plane of the at least two scales of the three-dimensional brain images, and to perform tilt correction on the at least two scales of the three-dimensional brain images based on the midsagittal plane of the at least two scales of the three-dimensional brain images. A segmentation module is used to extract brain tumor images based on the tilt-corrected three-dimensional brain images at at least two scales. The step of obtaining the median sagittal plane of the at least two scales of the three-dimensional brain images includes: obtaining an initial sagittal plane based on the geometric center of the three-dimensional brain images; extracting a predetermined number of pairs of three-dimensional image blocks from the three-dimensional brain images that are symmetrical about the initial sagittal plane; obtaining the similarity of each pair of three-dimensional image blocks; obtaining the symmetry of the initial sagittal plane based on the similarity of each pair of three-dimensional image blocks; obtaining the extreme value of the symmetry based on the symmetry of the initial sagittal plane, a genetic algorithm, and a Powell algorithm, and determining the sagittal plane corresponding to the extreme value of the symmetry as the median sagittal plane of the three-dimensional brain images. The tilt correction based on the median sagittal plane of the at least two scales of the three-dimensional brain images includes: obtaining the head and neck lateral flexion angle and head and neck rotation angle of the three-dimensional brain images based on the median sagittal plane of the at least two scales of the three-dimensional brain images and the expression of the sagittal plane in the three-dimensional coordinate system; and performing tilt correction on the three-dimensional brain images based on the head and neck lateral flexion angle and the head and neck rotation angle.
7. 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, it implements the steps of the method according to any one of claims 1 to 5.
8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 5.
Citation Information
Patent Citations
Left and right cerebral hemisphere segmentation method
CN107203997A
Brain image correction method, computer equipment and readable storage medium
CN111105362A