Medical image annotation method, segmentation method, segmentation model training method, electronic device and medium

Abstract

The present invention provides a medical image annotation method, a segmentation method, a segmentation model training method, an electronic device, and a storage medium. The annotation method includes: acquiring a medical image to be annotated; registering a pre-acquired standard medical image with the medical image to be annotated to obtain a registration matrix between the standard medical image and the medical image to be annotated; and transforming the medical image to be annotated onto a pre-acquired mask image of a target organ and tissue corresponding to the standard medical image based on the registration matrix to obtain an annotation result for the medical image to be annotated. The present invention can automatically annotate target organs and tissues on medical images, thereby rapidly generating a medical image sample set. The generated medical image sample set can be used to train a segmentation model. The trained segmentation model can achieve efficient and rapid segmentation of target organs and tissues based on a neural network model.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to a medical image annotation method, a segmentation method, a segmentation model training method, an electronic device, and a storage medium. Background Art

[0002] The basal ganglia, also known as the basal ganglia, is a collective term for a number of nuclei within the medulla of the cerebral hemispheres. It is composed of the caudate nucleus, putamen, globus pallidus (collectively known as the striatum), claustrum, and amygdala. The striatum, along with the subthalamic nucleus and substantia nigra, form a subcortical circuit that regulates movement, collaborating with the cerebral cortex and cerebellum to regulate voluntary movement, muscle tone, and postural reflexes. Therefore, the ability to accurately identify various basal ganglia in medical imaging is of great significance to doctors in treating motor neurological diseases, as well as assisting in surgical planning and postoperative evaluation.

[0003] Existing basal ganglia segmentation techniques are primarily performed manually by doctors or based on automated registration methods. For example, these methods calculate the deformation fields of different MRI images and perform registration based on the deformation field vectors, or use B-spline surfaces as deformation models to achieve registration of different MRI images, indirectly achieving target nucleus segmentation. Manual segmentation by doctors is not only time-consuming but also requires extensive experience. Automated registration methods require iteration and are also time-consuming. Summary of the Invention

[0004] The purpose of the present invention is to provide a medical image annotation method, a segmentation method, a segmentation model training method, an electronic device and a storage medium, which can generate samples required for training a neural network model for medical image segmentation, so as to achieve efficient segmentation of target organ tissues based on the neural network model.

[0005] To achieve the above objectives, the present invention provides a medical image annotation method, comprising:

[0006] Obtaining medical images to be annotated;

[0007] Using a pre-acquired standard medical image as a reference image and the medical image to be annotated as an image to be registered, registering the standard medical image and the medical image to be annotated to obtain a registration matrix between the standard medical image and the medical image to be annotated;

[0008] According to the registration matrix, the medical image to be annotated is transformed onto a pre-acquired target organ tissue mask image corresponding to the standard medical image to obtain an annotation result of the medical image to be annotated.

[0009] Optionally, the transforming the medical image to be annotated onto a pre-acquired target organ tissue mask image corresponding to the standard medical image to obtain an annotation result of the medical image to be annotated includes:

[0010] Transforming the medical image to be annotated onto a pre-acquired target organ tissue mask image corresponding to the standard medical image, and obtaining a preliminary annotation result of the medical image to be annotated in the coordinate system of the target organ tissue mask image;

[0011] According to the registration matrix, the preliminary annotation result is transformed into the coordinate system of the medical image to be annotated to obtain a final annotation result of the medical image to be annotated.

[0012] Optionally, registering the standard medical image and the medical image to be annotated to obtain a registration matrix between the standard medical image and the medical image to be annotated includes:

[0013] Converting the standard medical image and the medical image to be annotated from a pixel coordinate system to a physical coordinate system;

[0014] Using the translation matrix between the centroid of the standard medical image and the centroid of the medical image to be annotated as an initial transformation matrix for registration;

[0015] Based on the initial transformation matrix, transforming the medical image to be annotated into the physical coordinate system of the standard medical image;

[0016] The mutual information between the transformed medical image to be annotated and the standard medical image is calculated, and an iterative search is performed to obtain a registration matrix between the standard medical image and the medical image to be annotated.

[0017] To achieve the above object, the present invention also provides a segmentation model training method, comprising:

[0018] Acquire a sample set, the sample set including multiple samples, the samples including medical sample images and medical label images corresponding to the medical sample images, the medical label images being obtained by labeling the medical sample images using the medical image labeling method described above;

[0019] The pre-built neural network model is trained according to the sample set to obtain a segmentation model.

[0020] Optionally, before training the pre-built neural network model according to the sample set, the training method includes:

[0021] The medical sample image and the medical label image in the sample are preprocessed to crop the medical sample image and the medical label image into a plurality of medical sample sub-images and medical label sub-images with preset sizes, wherein the medical label sub-images correspond one-to-one to the medical sample sub-images.

[0022] Optionally, the preprocessing of the medical sample image and the medical label image in the sample to crop the medical sample image and the medical label image into a plurality of medical sample sub-images and medical label sub-images having preset sizes includes:

[0023] scaling the medical sample image and the medical label image to a target size;

[0024] performing truncation processing on the medical sample image scaled to a target size, so as to adjust the pixel value of each pixel point in the medical sample image to within a preset range;

[0025] Normalizing the pixel value of each pixel in the truncated medical sample image;

[0026] The normalized medical sample image and the medical label image scaled to a target size are cropped to crop the medical sample image and the medical label image into a plurality of medical sample sub-images and medical label sub-images having preset sizes.

[0027] Optionally, normalizing the pixel value of each pixel in the truncated medical sample image includes:

[0028] The pixel value of each pixel in the truncated medical sample image is normalized according to the following formula:

[0029]

[0030] Among them, P' i is the pixel value of pixel i in the medical sample image after normalization, P i is the pixel value of the pixel point i in the medical sample image after truncation, is the average value of the pixel values ​​of the medical sample image after truncation, and σ is the standard deviation of the pixel values ​​of the medical sample image after truncation.

[0031] Optionally, the step of cropping the normalized medical sample image and the medical label image scaled to a target size to crop the medical sample image and the medical label image into a plurality of medical sample sub-images and medical label sub-images having preset sizes includes:

[0032] Cropping the normalized medical sample image and the medical label image scaled to a target size to obtain a valid medical sample image and a valid medical label image;

[0033] The valid medical sample image and the valid medical label image are gradually cropped to obtain a plurality of medical sample sub-images and medical label sub-images with preset sizes.

[0034] To achieve the above object, the present invention further provides a medical image segmentation method, comprising:

[0035] Acquire a medical image to be segmented;

[0036] The medical image to be segmented is segmented using a pre-trained segmentation model to obtain a target organ tissue segmentation image, wherein the segmentation model is trained using the training method described above.

[0037] Optionally, before using a pre-trained segmentation model to segment the medical image to be segmented, the segmentation method includes:

[0038] Cutting the medical image to be segmented into a plurality of medical sub-images to be segmented with preset sizes;

[0039] The method of using a pre-trained segmentation model to segment the medical image to be segmented to obtain a target organ tissue segmentation image includes:

[0040] Using a pre-trained neural network model to segment the plurality of medical sub-images to be segmented, respectively, to obtain a plurality of target organ tissue sub-segmentation images;

[0041] A target organ tissue segmentation image is acquired according to the multiple target organ tissue sub-segmentation images.

[0042] Optionally, the step of cropping the medical image to be segmented into a plurality of medical sub-images to be segmented having preset sizes includes:

[0043] Scaling the medical image to be segmented to a target size;

[0044] performing truncation processing on the medical image to be segmented that has been scaled to a target size, so as to adjust the pixel value of each pixel point in the medical image to be segmented to within a preset range;

[0045] performing normalization processing on the pixel value of each pixel point in the medical image to be segmented after the truncation processing;

[0046] The normalized medical image to be segmented is cropped to crop the medical image to be segmented into a plurality of medical sub-images to be segmented with preset sizes.

[0047] Optionally, acquiring a target organ tissue segmentation image based on the multiple target organ tissue sub-segmentation images includes:

[0048] splicing the plurality of target organ tissue sub-segmentation images according to a preset weight template to obtain an initial target organ tissue segmentation image;

[0049] The size of the initial target organ tissue segmentation image is adjusted to the original size of the medical image to be segmented to obtain a final target organ tissue segmentation image.

[0050] To achieve the above-mentioned objectives, the present invention also provides an electronic device, comprising a processor and a memory, wherein a computer program is stored on the memory, and when the computer program is executed by the processor, the medical image annotation method, segmentation model training method or medical image segmentation method described above is implemented.

[0051] To achieve the above-mentioned objectives, the present invention also provides a readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the medical image annotation method, segmentation model training method or medical image segmentation method described above.

[0052] Compared with the prior art, the medical image annotation method, segmentation method, segmentation model training method, electronic device and storage medium provided by the present invention have the following advantages: the present invention obtains a medical image to be annotated; then uses a pre-acquired standard medical image as a reference image and the medical image to be annotated as an image to be registered, and registers the standard medical image and the medical image to be annotated to obtain a registration matrix between the standard medical image and the medical image to be annotated; then, based on the registration matrix, the medical image to be annotated is transformed to a pre-acquired target organ tissue mask image corresponding to the standard medical image, and obtains the annotation result of the medical image to be annotated. Thus, by adopting the annotation method provided by the present invention, the target organ tissue can be automatically annotated on the medical image (for example, the basal ganglia can be annotated on the brain image), so that a medical image sample set can be quickly generated. The generated medical image sample set can be used to train the segmentation model. The trained segmentation model can achieve efficient and rapid segmentation of the target organ tissue (for example, the basal ganglia can be annotated) based on the neural network model. In addition, the medical image segmentation method provided by the present invention can achieve end-to-end rapid segmentation, solving the tedious manual segmentation operations of doctors and the time-consuming problem of segmentation based on the registration method, and can better assist doctors in completing pre-operative surgical planning and post-operative evaluation. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 Schematic diagram of the process of a medical image annotation method in one embodiment of the present invention;

[0054] Figure 2 is a specific example of a standard medical image;

[0055] Figure 3 for Figure 2 The corresponding target organ tissue mask image;

[0056] Figure 4 Schematic diagram of the process of the mutual information registration method in one embodiment of the present invention;

[0057] Figure 5 is a specific example of a medical image to be annotated in the present invention;

[0058] Figure 6 for Figure 5 The annotation results of

[0059] Figure 7 Schematic diagram of the process of a segmentation model training method in one embodiment of the present invention;

[0060] Figure 8 Schematic diagram of the cropping process of medical sample images and medical label images in one embodiment of the present invention;

[0061] Figure 9 A schematic diagram of a process for training a neural network model in one embodiment of the present invention;

[0062] Figure 10 Schematic diagram of the flow of a medical image segmentation method in one embodiment of the present invention;

[0063] Figure 11 Schematic diagram of the cropping process of a medical image to be segmented in one embodiment of the present invention;

[0064] Figure 12 Schematic diagram of the block structure of an electronic device in one embodiment of the present invention;

[0065] The accompanying drawings are numerals as follows:

[0066] STN nucleus-11; GPi nucleus-12;

[0067] Processor-21; communication interface-22; memory-23; communication bus-24. DETAILED DESCRIPTION

[0068] The following is combined with Figures 1 to 12The medical image annotation method, segmentation method, segmentation model training method, electronic device and storage medium proposed in the present invention are further described in detail in the following embodiments. The advantages and features of the present invention will become clearer according to the following description. It should be noted that the drawings are in a very simplified form and are not in precise proportions, which are only used to conveniently and clearly assist in explaining the purpose of the embodiments of the present invention. In order to make the purpose, features and advantages of the present invention more obvious and easy to understand, please refer to the drawings. It should be noted that the structures, proportions, sizes, etc. illustrated in the drawings of this specification are only used to match the contents disclosed in the specification for people familiar with this technology to understand and read, and are not used to limit the conditions for the implementation of the present invention. Any modification of the structure, change in the proportional relationship or adjustment of the size should still fall within the scope of the technical content disclosed in the present invention when the effect produced by the present invention and the purpose that can be achieved are the same or similar.

[0069] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply the existence of any such actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or device comprising the element.

[0070] The core idea of ​​the present invention is to provide a medical image annotation method, segmentation method, segmentation model training method, electronic device and storage medium, which can generate samples required for training a neural network model for medical image segmentation, so as to achieve efficient segmentation of target organ tissues based on the neural network model.

[0071] It should be noted that the electronic device of the embodiment of the present invention can be a personal computer, a mobile terminal, etc., and the mobile terminal can be a hardware device with various operating systems such as a mobile phone and a tablet computer. In addition, it should be noted that although the present invention is described by taking the segmentation (or labeling) of the basal ganglia (including the STN ganglia and the GPi ganglia) from the brain image as an example, as those skilled in the art will understand, the present invention can also be used to segment (or label) other brain tissues from the brain image, and to segment other organ tissues from other medical images, and the present invention is not limited to this.

[0072] To realize the above idea, the present invention provides a medical image annotation method, please refer to Figure 1 , which schematically shows a flow chart of a medical image annotation method provided by one embodiment of the present invention. Figure 1 As shown, the medical image annotation method includes the following steps:

[0073] Step S110: Acquire a medical image to be annotated.

[0074] The medical image to be annotated can be a brain image or an image of another organ, and the present invention is not limited to this. The size of the medical image to be annotated can be set according to the specific situation, and the present invention is not limited to this. For example, the size of the medical image to be annotated can be 256×256×176 pixels. The medical image to be segmented can be collected by an image acquisition device, such as a CT or MRI imaging device, or can be collected through the Internet or scanned by a scanning device.

[0075] Step S120: Using a pre-acquired standard medical image as a reference image and the medical image to be annotated as an image to be registered, the standard medical image and the medical image to be annotated are registered to obtain a registration matrix between the standard medical image and the medical image to be annotated.

[0076] Wherein, the standard medical image is selected from the organ atlas template recognized by the industry. Taking the brain image as an example of the medical image to be annotated, in the field of brain nerves, there is a public brain atlas template recognized by the International Brain Mapping Association. The brain atlas template is imaged under 7T MRI and manually segmented and confirmed by multiple neurologists. Various brain tissues are segmented in the brain atlas, such as various basal nerve nuclei, including STN nerve nuclei and GPi nerve nuclei. The present invention can use the brain atlas template recognized by the International Brain Mapping Association as the standard medical image, and use the target organ tissue (such as STN nerve nuclei and GPi nerve nuclei) image segmented in the brain atlas as the target organ tissue mask image below. Please refer to Figure 2 and Figure 3 ,in Figure 2 A specific example of a standard medical image is given schematically. Figure 3 Schematically given Figure 2 The target organ tissue mask image corresponding to the standard medical image shown is Figure 3 Manual comparison of multiple neurologists Figure 2 The standard medical image shown is segmented to obtain mask images of the STN neural nucleus and the GPi neural nucleus.

[0077] Therefore, by using the standard medical image as a reference image and the medical image to be annotated as an image to be registered, the standard medical image and the medical image to be annotated are registered, and a registration matrix between the standard medical image and the medical image to be annotated can be obtained. The registration matrix can be represented by a 4*4 rotation and translation matrix M.

[0078] Specifically, the standard medical image and the medical image to be annotated may be registered based on a mutual information registration method. It should be noted that, as those skilled in the art will appreciate, in other embodiments, other image registration methods in the prior art may be employed, such as template matching-based image registration, model-based image registration, feature-based image registration, and the like, and the present invention is not limited thereto.

[0079] For further information, please refer to Figure 4 , which schematically shows a flow chart of the mutual information registration method provided by one embodiment of the present invention. Figure 4 As shown, the mutual information registration method is used to register the standard medical image and the medical image to be annotated, including:

[0080] Converting the standard medical image and the medical image to be annotated from a pixel coordinate system to a physical coordinate system;

[0081] Using the translation matrix between the centroid of the standard medical image and the centroid of the medical image to be annotated as an initial transformation matrix for registration;

[0082] Based on the initial transformation matrix, transforming the medical image to be annotated into the physical coordinate system of the standard medical image;

[0083] The mutual information between the transformed medical image to be annotated and the standard medical image is calculated, and an iterative search is performed to obtain a registration matrix between the standard medical image and the medical image to be annotated.

[0084] Specifically, the standard medical image can be transformed from a pixel coordinate system to a physical coordinate system based on its tag information (including the spacing value for converting a single pixel to millimeters, the offset value between the origin of the pixel coordinate system and the origin of the physical coordinate system, and the angle value representing the direction of the coordinate system). Similarly, the medical image to be annotated can be transformed from a pixel coordinate system to a physical coordinate system based on its tag information.

[0085] It should be noted that, as those skilled in the art will appreciate, the centroid of an image is also referred to as the center of the image, which is equivalent to the center of the image grayscale. There are multiple methods for obtaining the centroid. The present invention can adopt any centroid calculation method in the prior art, such as moment calculation, to calculate the centroid of the standard medical image and the medical image to be annotated. The specific calculation process can refer to the prior art, and the present invention will not elaborate on this. After calculating the centroids of the standard medical image and the medical image to be annotated, the translation matrix between the centroid of the standard medical image and the centroid of the medical image to be annotated can be calculated based on the position coordinates of the two centroids. This translation matrix is ​​the initial transformation matrix for registration.

[0086] Specifically, the medical image to be annotated can be transformed into the physical coordinate system of the standard medical image based on the initial transformation matrix. Thus, by performing initial alignment through the initial transformation matrix, unnecessary iterations can be reduced. Since both the medical image to be annotated and the standard medical image are discrete image data, the coordinates of the transformed image pixels are obtained after the coordinate transformation of the medical image to be annotated using the initial transformation matrix, and their positions do not necessarily fall exactly on integer coordinates. Therefore, it is necessary to interpolate such pixels on the image obtained after the transformation, and estimate the grayscale value of the point based on the grayscale and spatial information of the surrounding pixels. Specifically, interpolation methods in the prior art can be used, such as the nearest neighbor interpolation method, the trilinear interpolation method, the PV interpolation method (partial volume interpolation method), and the like.

[0087] After transforming the medical image to be annotated into the physical coordinate system of the standard medical image, the number of occurrences of corresponding pixel pairs (i, j) in the overlapping area between the standard medical image and the medical image to be annotated can be counted to obtain a joint grayscale histogram H AB , according to the joint grayscale histogram H AB The mutual information between the standard medical image and the medical image to be annotated can be calculated.

[0088] Assume that the grayscale value probability densities of the medical image to be annotated A and the standard medical image B are P A (i) P B (j), the joint probability density of pixel pair (i, j) is P AB (i, j), the mutual information I(A, B) between the medical image to be annotated A and the standard medical image B is defined as:

[0089]

[0090]

[0091] P A (i)=∑ j PAB (i,j)

[0092] P B (j) = ∑ i P AB (i,j)

[0093] H AB (i,j) is the joint histogram H AB The elements of , which represent the number of pixel pairs with grayscale value i in image A and grayscale value j in image B.

[0094] By performing an iterative search based on mutual information, a registration matrix can be obtained that maximizes the mutual information between the standard medical image and the medical image to be annotated. Specifically, a stochastic gradient descent method can be used for iterative search. By performing an iterative search, the transformation matrix between the standard medical image and the medical image to be annotated can be continuously iteratively updated. When a preset iteration termination condition is met, such as when the iteration error is less than a preset threshold, the iteration is terminated, and the transformation matrix at this time is used as the registration matrix between the standard medical image and the medical image to be annotated. Regarding how to perform an iterative search using the stochastic gradient descent method, reference can be made to the relevant description in the prior art, and the present invention will not elaborate on this.

[0095] In an exemplary embodiment, before executing step S120, the marking method further includes:

[0096] Smoothing the medical image to be annotated to filter out noise in the medical image to be annotated.

[0097] Correspondingly, registering the standard medical image and the medical image to be annotated includes:

[0098] The standard medical image and the smoothed medical image to be annotated are registered.

[0099] Therefore, by smoothing the medical image to be annotated, for example, using a Gaussian filter for smoothing, the noise information in the medical image to be annotated can be effectively filtered out. Since the measurement function of the mutual information is a non-smooth unimodal function, there will be many local optimal values. By smoothing, the iteration can be prevented from falling into the local optimal value.

[0100] Step S130 : transforming the medical image to be annotated onto a pre-acquired target organ tissue mask image corresponding to the standard medical image according to the registration matrix, and obtaining an annotation result of the medical image to be annotated.

[0101] Before transforming the medical image to be annotated onto the target organ tissue mask image, the target organ tissue mask image needs to be converted from the pixel coordinate system to the physical coordinate system. Suppose the coordinates of the pixel point on the medical image to be annotated are (X i ,Y i ,Z i ), then the coordinates (X j ,Y j ,Z j ), can be obtained by the following formula:

[0102]

[0103] in, is the registration matrix M between the standard medical image and the medical image to be annotated.

[0104] Therefore, according to the registration matrix, the medical image to be annotated can be transformed into the target organ tissue mask image, that is, the pixel value of each pixel point on the target tissue mask image can be assigned to the corresponding pixel point on the medical image to be annotated, so that the segmentation result of the target organ tissue area of ​​the medical image to be annotated can be obtained (the pixel value of the area where the target organ tissue is located is non-0 (for example, the pixel value of the area where the STN nerve nucleus is located is 1, and the pixel value of the area where the GPi nerve nucleus is located is 2), and the pixel value of other areas is 0), and the annotation result of the medical image to be annotated can be obtained based on the segmentation result.

[0105] Specifically, the step of transforming the medical image to be annotated onto a pre-acquired target organ tissue mask image corresponding to the standard medical image to obtain an annotation result of the medical image to be annotated may include:

[0106] Transforming the medical image to be annotated onto a pre-acquired target organ tissue mask image corresponding to the standard medical image, and obtaining a preliminary annotation result of the medical image to be annotated in the coordinate system of the target organ tissue mask image;

[0107] According to the registration matrix, the preliminary annotation result is transformed into the coordinate system of the medical image to be annotated to obtain a final annotation result of the medical image to be annotated.

[0108] Since what is obtained through step S130 is the preliminary annotation result of the medical image to be annotated in the coordinate system of the target organ tissue mask image (i.e., the coordinate system of the standard medical image), it is necessary to perform an inverse transformation according to the registration matrix to inversely transform the preliminary annotation result to the coordinate system of the medical image to be annotated to obtain the final annotation result of the medical image to be annotated.

[0109] Please refer to Figure 5 and Figure 6 ,in, Figure 5 A specific example of a medical image to be annotated in the present invention is schematically given. Figure 6 Schematically given Figure 5 The annotation results of the medical image shown in . Figure 5 and Figure 6 As shown, by adopting the medical image annotation method described in the present invention, the target organ tissue area can be quickly and accurately annotated in the medical image, for example, the area where the STN neural nucleus 11 and the GPi neural nucleus 12 are located can be annotated in the brain image, thereby providing support for the rapid generation of samples required for training the neural network model for medical image segmentation, and laying the foundation for realizing the segmentation of medical images based on the neural network model.

[0110] Based on the same inventive concept, the present invention also provides a segmentation model training method, please refer to Figure 7 , which schematically shows a flow chart of a segmentation model training method provided by one embodiment of the present invention. Figure 7 As shown, the training method includes:

[0111] Step S210: Acquire a sample set, where the sample set includes multiple samples, and the samples include medical sample images and medical label images corresponding to the medical sample images;

[0112] Step S220: training the pre-built neural network model according to the sample set to obtain a segmentation model.

[0113] Specifically, the medical label image is obtained by annotating the medical sample image using the medical image annotation method described above. The number of samples in the sample set can be set according to specific circumstances. For example, the sample set may include 200 samples, each of which includes a medical sample image and a medical label image corresponding to the medical sample image (the pixel value in the area where the target organ tissue is located is non-zero (for example, the pixel value in the area where the STN neural nucleus is located is 1, and the pixel value in the area where the GPi neural nucleus is located is 2), and the pixel values ​​in other areas are 0), wherein the medical label image in each sample is obtained by annotating the medical sample image using the annotation method described above.

[0114] The neural network model in this embodiment adopts a 3D V-Net network structure. The V-Net network structure adopts an encoder-decoder structure. The encoder consists of a convolutional layer and a pooling layer, and the decoder consists of a convolutional layer and a deconvolutional layer. V-Net uses a residual learning structure in the convolutional layer, that is, the input of the convolutional layer is added to the final output of the convolutional layer, and then subsequent calculations are performed, thereby solving the problem of vanishing gradients. The pooling layer of the V-Net network adopts a convolutional pooling method, which can effectively reduce the memory usage during training.

[0115] When training the neural network model, the loss function L used in the present invention is determined by the Dice loss L1 and the binary cross entropy loss L2. Specifically:

[0116] L=L1+L2

[0117] Therefore, the binary cross entropy loss L2 can be used to improve the situation where training fluctuation instability is prone to occur when only Dice loss is used as the loss function, while the Dice loss L1 can improve the situation where the target organ tissue (such as the target basal ganglia) occupies a smaller proportion of the entire background.

[0118] Preferably, before training the pre-built neural network model according to the sample set, the training method includes:

[0119] The medical sample image and the medical label image in the sample are preprocessed to crop the medical sample image and the medical label image into a plurality of medical sample sub-images and medical label sub-images with preset sizes, wherein the medical label sub-images correspond one-to-one to the medical sample sub-images.

[0120] Therefore, by cropping the medical sample image into multiple medical sample sub-images with preset sizes (for example, 64×64×64), and correspondingly cropping the medical label image into multiple medical label sub-images with preset sizes (for example, 64×64×64), each sample can be cropped into small areas and input into the network model for training the network model, effectively avoiding the problem of not being able to directly put the entire sample data in due to processor video memory limitations.

[0121] For further information, please refer to Figure 8 , which schematically shows a schematic diagram of the cropping process of medical sample images and medical label images provided by an embodiment of the present invention. Figure 8 As shown, the preprocessing of the medical sample image and the medical label image in the sample to crop the medical sample image and the medical label image into a plurality of medical sample sub-images and medical label sub-images having preset sizes may specifically include:

[0122] scaling the medical sample image and the medical label image to a target size;

[0123] performing truncation processing on the medical sample image scaled to a target size, so as to adjust the pixel value of each pixel point in the medical sample image to within a preset range;

[0124] Normalizing the pixel value of each pixel in the truncated medical sample image;

[0125] The normalized medical sample image and the medical label image scaled to a target size are cropped to crop the medical sample image and the medical label image into a plurality of medical sample sub-images and medical label sub-images having preset sizes.

[0126] Due to the differences between different medical imaging devices (such as MRI scanning devices), the original resolutions of medical images collected by different medical imaging devices will be slightly different, and the original image sizes of the samples are also different. Therefore, it is necessary to scale the medical sample image and the medical label image in each sample. Specifically, the original resolution values ​​of the medical sample image in the X, Y, and Z directions of each sample in the sample set can be counted separately and sorted, and then the median of the resolution in the X, Y, and Z directions is selected as the target resolution value in the corresponding direction, and then the target size of each medical sample image is calculated based on the original size, original resolution, and calculated target resolution in each direction. Specifically, the target size target_size of the medical sample image can be calculated according to the following formula:

[0127] target_size=origin_spacing*origin_size / target_spacing

[0128] Wherein, origin_spacing represents the original resolution of the medical sample image, origin_size represents the original size of the medical sample image, and target_spacing represents the target resolution.

[0129] It should be noted that, as those skilled in the art will appreciate, when calculating the target size of the medical sample image, the target sizes of the medical sample image in the X, Y, and Z directions are calculated using the above-mentioned target size calculation formulas, specifically:

[0130] target_size(X)=origin_spacing(X)*origin_size(X) / target_spacing(X)

[0131] target_size(Y)=origin_spacing(Y)*origin_size(Y) / target_spacing(Y)

[0132] target_size(Z)=origin_spacing(Z)*origin_size(Z) / target_spacing(Z)

[0133] Wherein, target_size(X) represents the target size of the medical sample image in the X direction, origin_spacing(X) represents the original resolution of the medical sample image in the X direction, origin_size(X) represents the original size of the medical sample image in the X direction, and target_spacing(X) represents the target resolution of the medical sample image in the X direction; target_size(Y) represents the target size of the medical sample image in the Y direction, origin_spacing(Y) represents the original resolution of the medical sample image in the Y direction, origin_size(Y) represents the original size of the medical sample image in the Y direction, and target_spacing(Y) represents the target resolution of the medical sample image in the Y direction; target_size(Z) represents the target size of the medical sample image in the Z direction, origin_spacing(Z) represents the original resolution of the medical sample image in the Z direction, origin_size(Z) represents the original size of the medical sample image in the Z direction, and target_spacing(Z) represents the target resolution of the medical sample image in the Z direction.

[0134] Finally, each sample (including the medical sample image and the corresponding medical label image) is resized to its corresponding target size. Specifically, the medical sample image can be resized to the target size using spline interpolation (trilinear interpolation), and the medical label image can be resized to the target size using nearest neighbor interpolation. Of course, as will be appreciated by those skilled in the art, other interpolation methods can also be used to resize the medical sample image and the medical label image to the target size.

[0135] After resizing the sample to the target size, the medical sample image scaled to the target size is truncated to adjust the pixel values ​​of each pixel in the medical sample image to within a preset range, thereby facilitating subsequent normalization processing. Specifically, the pixel values ​​of pixels in the medical sample image scaled to the target size with pixel values ​​less than 100 can be set to 100, the pixel values ​​of pixels with pixel values ​​greater than 800 can be set to 800, and the pixel values ​​of pixels in the range [100, 800] can remain unchanged.

[0136] By normalizing the pixel values ​​of each pixel in the truncated medical sample image, significant gradient changes during network model training can be effectively avoided, thereby achieving better iterative convergence of the neural network. Specifically, the pixel values ​​of each pixel in the truncated medical sample image can be normalized according to the following formula:

[0137]

[0138] Among them, P' i is the pixel value of pixel i in the medical sample image after normalization, P i is the pixel value of the pixel point i in the medical sample image after truncation, is the average value of the pixel values ​​of the medical sample image after truncation, and σ is the standard deviation of the pixel values ​​of the medical sample image after truncation.

[0139] Furthermore, the step of cropping the normalized medical sample image and the medical label image scaled to a target size to crop the medical sample image and the medical label image into a plurality of medical sample sub-images and medical label sub-images having preset sizes may specifically include:

[0140] Cropping the normalized medical sample image and the medical label image scaled to a target size to obtain a valid medical sample image and a valid medical label image;

[0141] The valid medical sample image and the valid medical label image are gradually cropped to obtain a plurality of medical sample sub-images and medical label sub-images with preset sizes.

[0142] Specifically, the normalized medical sample image can be cropped at a preset position range in the X, Y, and Z directions to crop out a valid medical sample image. Similarly, the same cropping operation is performed on the corresponding medical label image to crop out a valid medical label image. The preset position range can be set according to the specific situation. For example, when the medical sample image is a brain image and the target organ tissue to be segmented is the basal ganglia, since the approximate position of the basal ganglia in the brain is fixed, the [1 / 4, 3 / 4] position in the X, Y, and Z directions of the medical sample image can be selected for cropping to crop out a valid medical sample image. Similarly, the [1 / 4, 3 / 4] position in the X, Y, and Z directions of the medical label image can be selected for cropping to crop out a valid medical label image. Then, according to the preset size, the valid medical sample image is gradually cropped with a preset step size (for example, a step size of 32) to crop out a plurality of medical sample sub-images with a preset size (for example, 64×64×64). Similarly, the same cropping operation is performed on the valid medical sample image to crop out a plurality of medical label sub-images with a preset size (for example, 64×64×64). Thus, when the neural network model is finally trained, the medical sample sub-image and the corresponding medical label image can be used as a pair of training sub-samples and sent to the network model for training. It should be noted that, as can be understood by those skilled in the art, the preset size can be set according to the specific situation, and the present invention is not limited to this.

[0143] Please continue to refer to Figure 9 , which schematically shows a flow chart of the neural network model training provided by one embodiment of the present invention. Figure 9 As shown, in order to further improve the segmentation accuracy of the segmentation model, the pre-built neural network model is trained according to the sample set to obtain a segmentation model, which may specifically include:

[0144] From the sample set, determining a first proportion of samples as a training set, and determining a second proportion of samples as a test set;

[0145] Using the training set to train the pre-built neural network model to obtain an initial segmentation model;

[0146] Using the test set to test the output accuracy of the initial segmentation model;

[0147] If the output accuracy is less than the preset accuracy, the neural network model is retrained.

[0148] Specifically, the first ratio and the second ratio can be set according to specific circumstances. For example, 75% of the samples in the sample set can be selected as the training set, and the remaining 25% of the samples can be selected as the test set. For example, 150 samples out of 200 samples can be used for model training, and the other 50 samples can be used for model testing. The preset accuracy rate is set according to specific circumstances, for example, the preset accuracy rate is set to 95%. Thus, when the output accuracy rate is less than the preset accuracy rate, the neural network model can be retrained by increasing the number of samples in the sample set or readjusting the ratio of the training set and the test set (i.e., the first ratio and the second ratio) until the output accuracy rate is greater than or equal to the preset accuracy rate.

[0149] Based on the same inventive concept, the present invention also provides a medical image segmentation method, please refer to Figure 10 , which schematically shows a flow chart of a medical image segmentation method provided by one embodiment of the present invention. Figure 10 As shown, the medical image segmentation method includes the following steps:

[0150] Step S310: obtaining a medical image to be segmented;

[0151] Step S320: Segment the medical image to be segmented using a pre-trained segmentation model to obtain a target organ tissue segmentation image.

[0152] The segmentation model is trained using the training method described above. Thus, by using a pre-trained segmentation model to segment acquired medical images to be segmented, the present invention can rapidly segment target organ and tissue images, achieving an end-to-end algorithmic process and better assisting physicians in pre-operative surgical planning and post-operative evaluation.

[0153] Furthermore, before using a pre-trained segmentation model to segment the medical image to be segmented, the segmentation method includes:

[0154] The medical image to be segmented is cropped into a plurality of medical sub-images to be segmented with preset sizes.

[0155] For further information, please refer to Figure 11 , which schematically shows a schematic diagram of the cropping process of the medical image to be segmented provided by one embodiment of the present invention. Figure 11 As shown, the step of cropping the medical image to be segmented into a plurality of medical sub-images to be segmented having preset sizes specifically includes:

[0156] Scaling the medical image to be segmented to a target size;

[0157] performing truncation processing on the medical image to be segmented that has been scaled to a target size, so as to adjust the pixel value of each pixel point in the medical image to be segmented to within a preset range;

[0158] performing normalization processing on the pixel value of each pixel point in the medical image to be segmented after the truncation processing;

[0159] The normalized medical image to be segmented is cropped to crop the medical image to be segmented into a plurality of medical sub-images to be segmented with preset sizes.

[0160] Regarding how to crop the normalized medical image to be segmented so as to crop the medical image to be segmented into multiple medical sub-images to be segmented with preset sizes, reference can be made to the above description of how to crop the normalized medical sample image into multiple medical sample sub-images with preset sizes, so this will not be repeated here.

[0161] Correspondingly, the use of a pre-trained segmentation model to segment the medical image to be segmented to obtain a target organ tissue segmentation image includes:

[0162] Using a pre-trained neural network model to segment the plurality of medical sub-images to be segmented, respectively, to obtain a plurality of target organ tissue sub-segmentation images;

[0163] A target organ tissue segmentation image is acquired based on the plurality of target organ tissue sub-segmentation images.

[0164] Furthermore, acquiring a target organ tissue segmentation image based on the plurality of target organ tissue sub-segmentation images includes:

[0165] splicing the plurality of target organ tissue sub-segmentation images according to a preset weight template to obtain an initial target organ tissue segmentation image;

[0166] The size of the initial target organ tissue segmentation image is adjusted to the original size of the medical image to be segmented to obtain a final target organ tissue segmentation image.

[0167] Specifically, because the cropped areas may overlap, it is generally believed that the accuracy of the results predicted by the neural network in the central area is greater than that in the edge area. Therefore, a weight template of the same size as the target organ and tissue segmentation image is defined, for example, a weight template of 64×64×64. Within the weight template, the weights are Gaussian distributed, that is, the weights in the central area are large and the weights in the edge area are small. Each of the target organ and tissue sub-segmentation images is multiplied by the template weight (that is, each pixel in the target organ and tissue segmentation sub-image is multiplied by the corresponding weight), and then spliced ​​to obtain the initial target organ and tissue segmentation image. Since the initial target organ and tissue segmentation image is an image obtained by segmenting the effective area cropped from the medical image to be segmented, it is necessary to adjust the size of the initial target organ and tissue segmentation image to the original size of the medical image to be segmented. Specifically, the nearest neighbor interpolation method can be used to adjust the size of the initial target organ and tissue segmentation image to the original size of the medical image to be segmented to obtain the final target organ and tissue segmentation image.

[0168] Based on the same inventive concept, the present invention also provides an electronic device, please refer to Figure 12 , which schematically shows a block diagram of an electronic device provided by one embodiment of the present invention. Figure 12 As shown, the electronic device includes a processor 21 and a memory 23, and the memory 23 stores a computer program. When the computer program is executed by the processor 21, the medical image annotation method, segmentation model training method or medical image segmentation method described above is implemented.

[0169] like Figure 12 As shown, the electronic device further includes a communication interface 22 and a communication bus 24, wherein the processor 21, the communication interface 22, and the memory 23 communicate with each other via the communication bus 24. The communication bus 24 may be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The communication bus 24 may be divided into an address bus, a data bus, a control bus, etc. For ease of illustration, only one thick line is used in the figure, but this does not mean that there is only one bus or one type of bus. The communication interface 22 is used for communication between the electronic device and other devices.

[0170] The processor 21 referred to in the present invention may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor. The processor 21 is the control center of the electronic device, connecting various parts of the entire electronic device using various interfaces and lines.

[0171] The memory 23 may be used to store the computer program. The processor 21 implements various functions of the electronic device by running or executing the computer program stored in the memory 23 and calling the data stored in the memory 23.

[0172] The memory 23 may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM).

[0173] The present invention also provides a readable storage medium, which stores a computer program. When the computer program is executed by a processor, it can implement the medical image annotation method, segmentation model training method or medical image segmentation method described above.

[0174] The readable storage medium of the embodiment of the present invention can adopt any combination of one or more computer-readable media. The readable medium can be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium can be, for example, but not limited to, a system, device or component of electricity, magnetism, light, electromagnetic, infrared or semiconductor, or any combination thereof. More specific examples (non-exhaustive list) of computer-readable storage media include: an electrical connection with one or more wires, a portable computer hard disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In this article, a computer-readable storage medium can be any tangible medium containing or storing a program, which can be used by an instruction execution system, device or device or used in combination with it.

[0175] A computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, which carries computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium that can transmit, propagate, or transport a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0176] Computer program code for performing the operations of the present invention may be written in one or more programming languages, or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., through the Internet using an Internet service provider).

[0177] Compared with the prior art, the medical image annotation method, segmentation method, segmentation model training method, electronic device and storage medium provided by the present invention have the following advantages: the present invention obtains a medical image to be annotated; then uses a pre-acquired standard medical image as a reference image and the medical image to be annotated as an image to be registered, and registers the standard medical image and the medical image to be annotated to obtain a registration matrix between the standard medical image and the medical image to be annotated; then, based on the registration matrix, the medical image to be annotated is transformed to a pre-acquired target organ tissue mask image corresponding to the standard medical image, and obtains the annotation result of the medical image to be annotated. Thus, by adopting the annotation method provided by the present invention, the target organ tissue can be automatically annotated on the medical image (for example, the basal ganglia can be annotated on the brain image), so that a medical image sample set can be quickly generated. The generated medical image sample set can be used to train the segmentation model. The trained segmentation model can achieve efficient and rapid segmentation of the target organ tissue (for example, the basal ganglia can be annotated) based on the neural network model. In addition, the medical image segmentation method provided by the present invention can achieve end-to-end rapid segmentation, solving the tedious manual segmentation operations of doctors and the time-consuming problem of segmentation based on the registration method, and can better assist doctors in completing pre-operative surgical planning and post-operative evaluation.

[0178] It should be noted that the devices and methods disclosed in the embodiments of this document may also be implemented in other ways. The device embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings illustrate the possible architectures, functions, and operations of the devices, methods, and computer program products according to the various embodiments of this document. In this regard, each box in the flowchart or block diagram may represent a module, program, or portion of code, wherein the module, program segment, or portion of code contains one or more executable instructions for implementing a specified logical function, and the module, program segment, or portion of code contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the boxes may also occur in an order different from that marked in the accompanying drawings. For example, two consecutive boxes may actually be executed substantially in parallel, or they may sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart, may be implemented by a dedicated hardware-based system for performing the specified function or action, or may be implemented by a combination of dedicated hardware and computer instructions.

[0179] In addition, the functional modules in the various embodiments of this document may be integrated together to form an independent part, or each module may exist independently, or two or more modules may be integrated to form an independent part.

[0180] The above description is merely a description of preferred embodiments of the present invention and does not limit the scope of the present invention. Any changes or modifications made by persons skilled in the art based on the above disclosure are within the scope of protection of the present invention. Obviously, various modifications and variations may be made by persons skilled in the art without departing from the spirit and scope of the present invention. Thus, if such modifications and variations fall within the scope of the present invention and its equivalents, the present invention is intended to include such modifications and variations.

Claims

1. A medical image annotation method for generating samples required for training a neural network model for medical image segmentation, characterized in that: include: Obtaining medical images to be annotated; Using a pre-acquired standard medical image as a reference image and the medical image to be annotated as an image to be registered, registering the standard medical image and the medical image to be annotated to obtain a registration matrix between the standard medical image and the medical image to be annotated; transforming the medical image to be annotated onto a pre-acquired target organ tissue mask image corresponding to the standard medical image according to the registration matrix, and obtaining an annotation result of the medical image to be annotated; The registering the standard medical image and the medical image to be annotated to obtain a registration matrix between the standard medical image and the medical image to be annotated includes: Converting the standard medical image and the medical image to be annotated from a pixel coordinate system to a physical coordinate system; Using the translation matrix between the centroid of the standard medical image and the centroid of the medical image to be annotated as an initial transformation matrix for registration; Based on the initial transformation matrix, transforming the medical image to be annotated into the physical coordinate system of the standard medical image; Calculating the mutual information between the transformed medical image to be annotated and the standard medical image, and performing an iterative search to obtain a registration matrix between the standard medical image and the medical image to be annotated; The step of transforming the medical image to be annotated onto a pre-acquired target organ tissue mask image corresponding to the standard medical image according to the registration matrix to obtain an annotation result of the medical image to be annotated includes: transforming the medical image to be annotated onto a pre-acquired target organ and tissue mask image corresponding to the standard medical image according to the registration matrix, so as to assign the pixel value of each pixel point on the target organ and tissue mask image to the corresponding pixel point on the medical image to be annotated, thereby obtaining a preliminary annotation result of the medical image to be annotated in the coordinate system of the target organ and tissue mask image; According to the inverse transformation of the registration matrix, the preliminary annotation result is transformed into the coordinate system of the medical image to be annotated to obtain a final annotation result of the medical image to be annotated.

2. A segmentation model training method, characterized in that: include: Acquire a sample set, the sample set including multiple samples, the samples including medical sample images and medical label images corresponding to the medical sample images, the medical label images being obtained by labeling the medical sample images using the medical image labeling method according to claim 1; The pre-built neural network model is trained according to the sample set to obtain a segmentation model.

3. The segmentation model training method according to claim 2, characterized in that Before training the pre-built neural network model according to the sample set, the training method includes: The medical sample image and the medical label image in the sample are preprocessed to crop the medical sample image and the medical label image into a plurality of medical sample sub-images and medical label sub-images with preset sizes, wherein the medical label sub-images correspond one-to-one to the medical sample sub-images.

4. The segmentation model training method according to claim 3, characterized in that The preprocessing of the medical sample image and the medical label image in the sample to crop the medical sample image and the medical label image into a plurality of medical sample sub-images and medical label sub-images having preset sizes includes: scaling the medical sample image and the medical label image to a target size; performing truncation processing on the medical sample image scaled to a target size, so as to adjust the pixel value of each pixel point in the medical sample image to within a preset range; Normalizing the pixel value of each pixel in the truncated medical sample image; The normalized medical sample image and the medical label image scaled to a target size are cropped to crop the medical sample image and the medical label image into a plurality of medical sample sub-images and medical label sub-images having preset sizes.

5. The segmentation model training method according to claim 4, characterized in that The normalizing of the pixel value of each pixel point in the truncated medical sample image includes: The pixel value of each pixel in the truncated medical sample image is normalized according to the following formula: Among them, P i ' is the pixel value of pixel i in the medical sample image after normalization, P i is the pixel value of the pixel point i in the medical sample image after truncation, is the average value of the pixel values ​​of the medical sample image after truncation, and σ is the standard deviation of the pixel values ​​of the medical sample image after truncation.

6. The segmentation model training method according to claim 4, characterized in that The step of cropping the normalized medical sample image and the medical label image scaled to a target size to crop the medical sample image and the medical label image into a plurality of medical sample sub-images and medical label sub-images having preset sizes includes: Cropping the normalized medical sample image and the medical label image scaled to a target size to obtain a valid medical sample image and a valid medical label image; The valid medical sample image and the valid medical label image are gradually cropped to obtain a plurality of medical sample sub-images and medical label sub-images with preset sizes.

7. A medical image segmentation method, characterized in that: include: Acquire a medical image to be segmented; The medical image to be segmented is segmented using a pre-trained segmentation model to obtain a target organ tissue segmentation image, wherein the segmentation model is trained using the training method described in any one of claims 2 to 6.

8. The medical image segmentation method according to claim 7, characterized in that: Before using a pre-trained segmentation model to segment the medical image to be segmented, the segmentation method includes: Cutting the medical image to be segmented into a plurality of medical sub-images to be segmented with preset sizes; The method of using a pre-trained segmentation model to segment the medical image to be segmented to obtain a target organ tissue segmentation image includes: Using a pre-trained neural network model to segment the plurality of medical sub-images to be segmented, respectively, to obtain a plurality of target organ tissue sub-segmentation images; A target organ tissue segmentation image is acquired according to the multiple target organ tissue sub-segmentation images.

9. The medical image segmentation method according to claim 8, characterized in that: The step of cutting the medical image to be segmented into a plurality of medical sub-images to be segmented having preset sizes comprises: Scaling the medical image to be segmented to a target size; performing truncation processing on the medical image to be segmented that has been scaled to a target size, so as to adjust the pixel value of each pixel point in the medical image to be segmented to within a preset range; performing normalization processing on the pixel value of each pixel point in the medical image to be segmented after the truncation processing; The normalized medical image to be segmented is cropped to crop the medical image to be segmented into a plurality of medical sub-images to be segmented with preset sizes.

10. The medical image segmentation method according to claim 8, characterized in that: The step of acquiring a target organ tissue segmentation image based on the plurality of target organ tissue sub-segmentation images includes: splicing the plurality of target organ tissue sub-segmentation images according to a preset weight template to obtain an initial target organ tissue segmentation image; The size of the initial target organ tissue segmentation image is adjusted to the original size of the medical image to be segmented to obtain a final target organ tissue segmentation image.

11. An electronic device, characterized in that: The invention comprises a processor and a memory, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the labeling method according to the claim or the training method according to any one of claims 2 to 6 or the segmentation method according to any one of claims 7 to 10 is implemented.

12. A readable storage medium, characterized in that: The readable storage medium stores a computer program, and when the computer program is executed by the processor, it implements the labeling method according to claim 1, the training method according to any one of claims 2 to 6, or the segmentation method according to any one of claims 7 to 10.

Images