SWI-based brain micro-hemorrhage segmentation method and device, medium and product
Through the SWI-based brain microbleed segmentation method, the CeMiS-Net model and SAM model combined with convolutional neural networks are used to solve the problem of low reliability in the evaluation of brain microbleed in the existing technology, and more accurate brain microbleed detection is achieved and analysis improves the reliability of segmentation and the detection ability of micro lesions.
Patent Information
- Application Number
- CN202411056329.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-02
- Publication Date
- 2024-11-08
AI Technical Summary
The existing cerebral microbleed evaluation methods have low reliability in quantity and location, lack effective auxiliary tools, and it is difficult to accurately detect and analyze cerebral microbleed.
Using the SWI-based brain microbleed segmentation method, the target magnetically sensitive weighted image is obtained, and the CeMiS-Net model is used for preprocessing and training, and combining the SAM model and convolutional neural network to construct a multi-level jump structure cerebral microbleed segmentation model. Dice loss function, Focal loss function and cross entropy loss function are used for training to improve the reliability of segmentation.
It improves the reliability of cerebral microbleeding segmentation and the detection ability of models for micro lesions, enhances the accurate assessment of cerebral microbleeding, and reduces social and medical burdens.
Smart Images

Figure CN118918121A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of neural network technology, and in particular to a SWI-based cerebral microbleed segmentation method, device, medium and product. Background Art
[0002] As the global population aging problem becomes increasingly serious, related brain diseases, such as cerebral microbleeds, are gradually bringing an increasing burden to society. Cerebral microbleeds are common small blood vessel lesions in the brain of the elderly. With age, the structure and function of the brain undergo some subtle changes, which may aggravate and lead to the occurrence of cerebral microbleeds. The occurrence of cerebral microbleeds is closely related to multiple factors such as genetics, environmental factors, underlying diseases, and trauma, which may significantly accelerate the aging process of the brain. Quantitative assessment of cerebral microbleeds is particularly important because it can not only help understand the speed and stage of brain aging, but also has important significance for the early diagnosis, prevention and treatment of related neurodegenerative diseases. With the help of advanced brain imaging technology and biomarkers, the microbleeding status of the brain can be more accurately assessed, and then the pathological changes that may occur in this aging process can be monitored and intervened to reduce the social and medical burden caused by it. Accurate and effective detection and analysis of cerebral microbleeds has important clinical significance for in-depth understanding of its pathological mechanism and prevention of related complications.
[0003] At present, the assessment of cerebral microbleeds mainly adopts visual scoring systems, such as the Microbleed Anatomical Rating Scale (MARS) and the Brain Observer Microbleed Scale (BOMBS). In current clinical practice, these methods have relatively low reliability in assessing the number and location of cerebral microbleeds without corresponding auxiliary assessment tools. However, these methods have relatively low reliability in assessing the number and location of cerebral microbleeds unless professional assessment tools are used. Summary of the invention
[0004] The purpose of this application is to provide a SWI-based cerebral microbleed segmentation method, device, medium and product, which improves the reliability of cerebral microbleed segmentation.
[0005] To achieve the above objectives, this application provides the following solutions:
[0006] In a first aspect, the present application provides a method for segmenting cerebral microbleeds based on SWI, comprising:
[0007] Acquire a target susceptibility weighted image; the target susceptibility weighted image is a susceptibility weighted image of cerebral microbleeds in the brain to be segmented;
[0008] Based on the target susceptibility weighted image and the cerebral microbleeding segmentation model, the predicted value of the corresponding cerebral microbleeding segmentation image mask is determined; the cerebral microbleeding segmentation model is obtained by training CeMiS-Net using a training set; the training set includes: a plurality of preprocessed cerebral microbleeding susceptibility weighted images and the actual values of the corresponding cerebral microbleeding segmentation image mask; the CeMiS-Net is constructed based on the SAM model and the convolutional neural network.
[0009] Optionally, determining a prediction value of a corresponding cerebral microbleeding segmentation image mask based on the target susceptibility weighted image and the cerebral microbleeding segmentation model includes:
[0010] Preprocessing the target magnetic susceptibility weighted image to obtain a preprocessed target magnetic susceptibility weighted image; the preprocessing includes: standardization, data enhancement, oversampling and noise removal;
[0011] The preprocessed target susceptibility weighted image is input into the cerebral microbleeding segmentation model to obtain the prediction value of the corresponding cerebral microbleeding segmentation image mask.
[0012] Optionally, the training process of the cerebral microbleed segmentation model includes:
[0013] Obtaining the training set;
[0014] Constructing the CeMiS-Net based on the SAM model and convolutional neural network;
[0015] The CeMiS-Net is trained using the training set and the total loss function to obtain the cerebral microbleed segmentation model; the total loss function is the sum of the Dice loss function, the Focal loss function and the cross entropy loss function.
[0016] Optionally, obtaining the training set includes:
[0017] Acquire multiple cerebral microbleed susceptibility weighted images and corresponding actual values of cerebral microbleed segmentation image masks;
[0018] Preprocessing each of the cerebral microbleed susceptibility weighted images respectively to obtain a corresponding preprocessed cerebral microbleed susceptibility weighted image;
[0019] The actual values of all preprocessed cerebral microbleed susceptibility weighted images and the corresponding cerebral microbleed segmentation image masks were determined as the initial data set;
[0020] The initial data set is divided according to a set ratio to obtain the training set, validation set and test set.
[0021] Optionally, the CeMiS-Net comprises: a SAM encoder, a convolutional neural network encoder, a feature fusion module and a convolutional neural network decoder; the CeMiS-Net adopts a multi-stage skip structure;
[0022] The SAM encoder is used to obtain brain image structure information corresponding to an input image; the input image is a pre-processed cerebral microbleed susceptibility weighted image or a pre-processed target susceptibility weighted image;
[0023] The convolutional neural network encoder is used to obtain local information of cerebral microbleeds corresponding to the input image;
[0024] The feature fusion module is used to fuse the brain image structure information and the local information of cerebral microbleeds corresponding to the input image to obtain a fusion feature map corresponding to the input image;
[0025] The convolutional neural network decoder is used to obtain a corresponding cerebral microbleeding segmentation image mask based on a fused feature map corresponding to an input image.
[0026] Optionally, the SAM encoder comprises: a linear change layer, an embedded patch layer, a normalization layer, a multi-head attention layer and a multi-layer perceptron;
[0027] The convolutional neural network encoder comprises: a plurality of convolutional neural network encoding blocks;
[0028] The feature fusion module includes: a plurality of feature fusion submodules; the number of the feature fusion submodules is equal to the number of the convolutional neural network encoding blocks;
[0029] The convolutional neural network decoder includes: a plurality of convolutional neural network decoding blocks; the number of the convolutional neural network decoding blocks is 1 less than the number of the feature fusion submodules.
[0030] Optionally, the convolutional neural network encoding block is a decoding block formed by repeatedly stacking a first convolutional layer, a batch normalization layer, a ReLu activation function layer, and a second convolutional layer; the convolutional neural network decoding block is a decoding block formed by repeatedly stacking a deconvolutional layer, a batch normalization layer, a ReLu activation function layer, and a third convolutional layer;
[0031] The convolution of the first convolutional layer is 3×3×3, and the stride is 2×2×2;
[0032] The convolution of the second convolutional layer is 3×3×3, and the stride is 1×1×1;
[0033] The convolution of the deconvolution layer is 3×3×3 and the stride is 2×2×2;
[0034] The convolution of the third convolutional layer is 3×3×3 and the stride is 1×1×1.
[0035] In a second aspect, the present application provides a computer device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement any of the above-mentioned SWI-based cerebral microbleed segmentation methods.
[0036] In a third aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the computer program implements any of the above-mentioned SWI-based cerebral microbleed segmentation methods.
[0037] In a fourth aspect, the present application provides a computer program product, including a computer program, which, when executed by a processor, implements any of the above-mentioned SWI-based cerebral microbleed segmentation methods.
[0038] According to the specific embodiments provided in this application, this application discloses the following technical effects:
[0039] The present application discloses a SWI-based cerebral microbleed segmentation method, device, medium and product. First, a target susceptibility weighted image is obtained; the target susceptibility weighted image is a cerebral microbleed susceptibility weighted image of the brain to be segmented; then, based on the target susceptibility weighted image and the cerebral microbleed segmentation model, the predicted value of the corresponding cerebral microbleed segmentation image mask is determined; the cerebral microbleed segmentation model is obtained by training CeMiS-Net using a training set; the training set includes: a plurality of pre-processed cerebral microbleed susceptibility weighted images and the actual values of the corresponding cerebral microbleed segmentation image mask; CeMiS-Net is constructed based on a SAM model and a convolutional neural network. This application integrates the SAM model pre-trained based on brain medical images into the convolutional neural network, and fuses the information extracted by SAM with the information extracted by the convolutional neural network, so as to provide a better information extraction capability for the model and improve the reliability of cerebral microbleeding segmentation; according to the characteristics of cerebral microbleeding, this application adopts a multi-level jump method between the encoding layer and the decoding layer to compensate for the information loss caused by the deepening of the network layers, and further enhance the model's ability to detect tiny lesions; when training the model, the cross-entropy loss (CE Loss) is used for weighted operation with the Dice loss function and the Focal loss function, and training is performed according to the characteristics of small lesions, thereby improving the reliability of cerebral microbleeding segmentation. BRIEF DESCRIPTION OF THE DRAWINGS
[0040] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0041] Figure 1 A schematic diagram of a flow chart of a method for segmenting cerebral microbleeds based on SWI provided in one embodiment of the present application;
[0042] Figure 2 This is a flow chart of the brain microbleed segmentation method based on brain susceptibility weighted imaging;
[0043] Figure 3 This is the framework diagram of the cerebral microbleed segmentation model;
[0044] Figure 4 This is the specific structure diagram of the cerebral microbleed segmentation model;
[0045] Figure 5 This is the structure diagram of the multi-head attention mechanism;
[0046] Figure 6 This is the structural diagram of the embedded feature processing module;
[0047] Figure 7 Encode block structure diagram for convolutional neural network;
[0048] Figure 8 Decoding block diagram for convolutional neural network;
[0049] Fig. 9 A schematic diagram of the structure of a computer device provided in one embodiment of the present application. DETAILED DESCRIPTION
[0050] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.
[0051] The purpose of this application is to provide a SWI-based cerebral microbleed segmentation method, device, medium and product, aiming to improve the reliability of cerebral microbleed segmentation.
[0052] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below in conjunction with the accompanying drawings and specific implementation methods.
[0053] In an exemplary embodiment, Figure 1 and Figure 2 As shown, the SWI-based cerebral microbleed segmentation method in this embodiment includes:
[0054] Step 1: Acquire a target susceptibility weighted image; the target susceptibility weighted image is a susceptibility weighted image of cerebral microbleeds in the brain to be segmented.
[0055] Step 2: Based on the target susceptibility weighted image and the cerebral microbleeding segmentation model, determine the prediction value of the corresponding cerebral microbleeding segmentation image mask.
[0056] Among them, the cerebral microbleed segmentation model is obtained by training CeMiS-Net using the training set; the training set includes: multiple pre-processed cerebral microbleed susceptibility weighted imaging (SWI) and the actual values of the corresponding cerebral microbleed segmentation image masks; CeMiS-Net is built based on the SAM model and the Convolutional Neural Network (CNN). The full name of CeMiS-Net in English is A CNN-SAM Integrated Model for Cerebral Microbleed Segmentation, and its Chinese name is CNN-SAM Integrated Model for Cerebral Microbleed Segmentation.
[0057] As an optional implementation, step 2 includes:
[0058] Step 201: preprocessing the target susceptibility weighted image to obtain a preprocessed target susceptibility weighted image; the preprocessing includes: standardization, data enhancement, oversampling and noise removal.
[0059] Step 202: input the preprocessed target susceptibility weighted image into the cerebral microbleeding segmentation model to obtain the corresponding predicted value of the cerebral microbleeding segmentation image mask.
[0060] As an optional implementation, the training process of the cerebral microbleed segmentation model includes:
[0061] Step 211: Obtain a training set.
[0062] As an optional implementation, step 211 includes:
[0063] Step 2111: Acquire multiple cerebral microbleed susceptibility weighted images and actual values of corresponding cerebral microbleed segmentation image masks.
[0064] Specifically, the susceptibility-weighted images of cerebral microbleeds are derived from susceptibility-weighted images of cerebral microbleeds in susceptibility-weighted imaging data of 1,217 patients from three hospitals in Beijing.
[0065] Step 2112: preprocess each cerebral microbleed susceptibility weighted image respectively to obtain the corresponding preprocessed cerebral microbleed susceptibility weighted image.
[0066] Specifically, during preprocessing, the processing standards of susceptibility-weighted images of cerebral microbleeds were first unified by calculating the median of the voxel size and size in all susceptibility-weighted images of cerebral microbleeds to achieve standardization. Then, in order to enhance the diversity and richness of the data, data enhancement techniques were applied, including rotation, scaling, cropping and color transformation of susceptibility-weighted images of cerebral microbleeds. Thirdly, in view of the extremely low proportion of positive samples in the overall image, less than 1%, oversampling was implemented to increase the representativeness of positive areas by repeated sampling, thereby alleviating the class imbalance in the data set. Finally, in order to further improve the focus of the cerebral microbleed segmentation model on key areas, background noise with a field strength below a specific threshold was removed in three directions of the susceptibility-weighted images of cerebral microbleeds. This series of comprehensive preprocessing steps aims to improve the generalization ability and segmentation accuracy of the cerebral microbleed segmentation model.
[0067] Preprocessing includes: data enhancement, oversampling and noise removal.
[0068] Step 2113: Determine the actual values of all pre-processed cerebral microbleed susceptibility weighted images and corresponding cerebral microbleed segmentation image masks as the initial data set.
[0069] Step 2114: Divide the initial data set according to the set ratio to obtain a training set, a validation set, and a test set.
[0070] Specifically, the initial data set is divided into a training set, a validation set, and a test set in a ratio of 6:2:2. The training set is then used to train CeMiS-Net through back propagation and gradient descent algorithms, and model parameters with high prediction accuracy and strong generalization performance are selected and saved.
[0071] When training CeMiS-Net, the validation set is used as input, and the training process is evaluated based on the results of the validation set, and finally the cerebral microbleed segmentation model is obtained.
[0072] The cerebral microbleed segmentation model is applied to generate end-to-end image masks for the cerebral microbleed locations of the test set.
[0073] Step 212: Construct CeMiS-Net based on the SAM model and the convolutional neural network.
[0074] As an optional implementation, CeMiS-Net includes: a SAM encoder, a convolutional neural network encoder, a feature fusion module and a convolutional neural network decoder; CeMiS-Net adopts a multi-stage skip structure.
[0075] The SAM encoder is used to obtain brain image structure information corresponding to the input image; the input image is a preprocessed cerebral microbleed susceptibility weighted image or a preprocessed target susceptibility weighted image.
[0076] The convolutional neural network encoder is used to obtain the local information of cerebral microbleeds corresponding to the input image.
[0077] The feature fusion module is used to fuse the brain image structure information and the local information of cerebral microbleeds corresponding to the input image to obtain a fused feature map corresponding to the input image.
[0078] The convolutional neural network decoder is used to obtain the corresponding cerebral microbleed segmentation image mask based on the fused feature map corresponding to the input image.
[0079] As an optional implementation, the SAM encoder includes: a linear change layer, an embedded patch layer, a normalization layer, a multi-head attention layer and a multi-layer perceptron.
[0080] The convolutional neural network encoder comprises: a plurality of convolutional neural network encoding blocks.
[0081] The feature fusion module includes: a plurality of feature fusion sub-modules; the number of the feature fusion sub-modules is equal to the number of the convolutional neural network encoding blocks.
[0082] The convolutional neural network decoder includes: a plurality of convolutional neural network decoding blocks; the number of the convolutional neural network decoding blocks is one less than the number of the feature fusion submodules.
[0083] As an optional implementation, the convolutional neural network encoding block is a decoding block composed of a first convolutional layer, a batch normalization layer, a ReLu activation function layer, and a second convolutional layer repeatedly stacked together; the convolutional neural network decoding block is a decoding block composed of a deconvolutional layer, a batch normalization layer, a ReLu activation function layer, and a third convolutional layer repeatedly stacked together.
[0084] The convolution of the first convolutional layer is 3×3×3 and the stride is 2×2×2.
[0085] The convolution of the second convolutional layer is 3×3×3 and the stride is 1×1×1.
[0086] The deconvolution layer has a convolution size of 3×3×3 and a stride of 2×2×2.
[0087] The convolution of the third convolutional layer is 3×3×3 and the stride is 1×1×1.
[0088] Specifically, Figure 3 As shown in the figure, the overall architecture of CeMiS-Net is divided into four parts: the first is the 3D SAM image encoder (i.e., SAM encoder) obtained by the pre-trained three-dimensional (3D) SAM model, which obtains the brain image structure information through the pre-training data set; the second is the 3D CNN encoder (i.e., convolutional neural network encoder) composed of convolutional neural network parts, which is used to obtain the local information of cerebral microbleeds; the third is the feature fusion part (i.e., feature fusion module), which is used to fuse the brain image structure information with the local information of cerebral microbleeds to obtain a fused feature map; the fourth is the 3D CNN decoder (i.e., convolutional neural network decoder) composed of a 3D convolutional neural network, which uses a multi-level skip connection part to reduce the feature disappearance problem caused by the deepening of the network layer. The 3D CNN decoder is used to obtain the image mask based on the fused feature map.
[0089] Before training begins, the present application first obtains a SAM model pre-trained based on brain medical images as a SAM image encoder.
[0090] During the training process, the SAM encoder and the convolutional neural network encoder process the input image at the same time. After the information obtained by the SAM image encoder part and the convolutional neural network encoder part is feature fused, it is input into the convolutional neural network decoder to finally obtain the corresponding image mask.
[0091] During the test, the performance of the model was evaluated based on the test set, and the preprocessed cerebral microbleed susceptibility weighted images in the test set were input into the cerebral microbleed segmentation model to obtain the corresponding image mask.
[0092] Further, if Figure 4 As shown in the figure, CeMiS-Net specifically includes:
[0093] (1) Coding part
[0094] The encoding part includes: 3D SAM image encoder (i.e. SAM encoder), 3D CNN encoder (i.e. convolutional neural network encoder) and feature fusion module.
[0095] The image size in the 3D CNN encoder is (H c , W c , D c ), this image size maintains the original image space size ratio, the size selection setting is determined by the speed of the overall susceptibility weighted image and the image size. However, due to the limitations of the SAM preprocessing model, when input to the 3D SAM image encoder, the image size will be limited to (H s , W s, D s ), (H s , W s , D s ) is right (H c , W c , D c ) is obtained by trilinear interpolation.
[0096] Since the input of the 3D SAM image encoder is a 3D voxel image x∈R H×W×D×C Split into multiple one-dimensional sequences and input into the model (H, W, D) represents the size of the image, and C represents the non-overlapping image blocks x1∈R that are split N×(P,P,P,C) , P in (P, P, P) is the size of each image block of the image, N represents the length of the one-dimensional sequence.
[0097] The 3D SAM image encoder then projects these image patches into a K-dimensional embedding space through a linear transformation layer, but still keeps the transformer layer structure unchanged.
[0098] In such Figure 6 After the embedded feature processing block shown, a series of transformation blocks are used, including Figure 5 The Multi-Head Attention (MSA) layer and the Multilayer Perception (MLP) shown. The MLP consists of two linear layers with a Gaussian Error Linear Unit (GELU) activation layer.
[0099] z′ i =MSA(Norm(z i-1 ))+z i-1 ,i=1,…,L.
[0100] z i =MLP(Nornm(z′ i ))+z i ,i=1,…,L.
[0101]
[0102] Among them, z' i is the output of the i-th multi-head attention layer; Norm is the normalization layer; z i-1 is the output of the i-1th multilayer perceptron; L is the number of transformer layers; z i is the output of the i-th multilayer perceptron; GELU is the Gaussian error linear unit activation layer; is the cumulative function of the Gaussian normal distribution of x.
[0103] By obtaining z i (i∈{3,6,9,12}), with a size of Then reshape it into Tensor. This application passes all embedded features through multiple layers of 3×3×3 continuous 3D convolutions, performs batch normalization on the embedded features, and adjusts the resolution of the embedded features to the same resolution size as the corresponding convolutional neural network layers.
[0104] After obtaining the feature map output by the SAM encoder (i.e., brain image structure information) and the feature map output by the convolutional neural network encoder (i.e., local information of cerebral microbleeds), the two are fused to obtain a fused feature map, which is then input into the decoder part.
[0105] ReLu(x)=max(0,x).
[0106] Among them, ReLu is the ReLu activation function layer; max(0,x) is the maximum value between 0 and x.
[0107] like Figure 7 As shown in the figure, the 3D convolutional neural network encoder is an encoding block composed of a first convolutional layer, a batch normalization layer, a ReLu activation function layer, a second convolutional layer, etc., which is repeatedly stacked to process the fused feature map and the image input of the previous layer, and output the encoded feature map of the next layer after processing the fused feature map.
[0108] (2) Decoder part.
[0109] The decoder part includes: 3D CNN decoder (i.e. convolutional neural network decoder).
[0110] This application superimposes the multi-layer features of the encoder part on the channel dimension by interpolation, combines them with the low-dimensional features of the decoder, and inputs them into the 3D convolution block of the next layer to enhance the model's ability to detect local subtle features. Therefore, in the decoder part, the multi-level jump structural features are used to input the multi-level feature maps of the image into the convolutional neural network structure of each layer. For 3D feature maps of different dimensions, this model uses trilinear interpolation to pass the feature map to the decoder part. After multiple layers of 3D convolutional neural network operations, the output mask is finally obtained.
[0111] like Figure 8 As shown in the figure, the 3D convolutional neural network decoder is a decoding block composed of a deconvolution layer, batch normalization, ReLu activation function, and a third convolution layer, which is repeatedly stacked to process the input transmitted by the encoder and output the decoding feature map of the next layer.
[0112] Step 213: CeMiS-Net is trained using the training set and the total loss function to obtain a cerebral microbleed segmentation model; the total loss function is the sum of the Dice loss function, the Focal loss function, and the cross entropy loss function.
[0113] For the loss function of the model, this application chooses to use the Dice loss function Loss for the case where positive samples (positive sample voxels account for less than 1% of the total voxels) are rare. Dice , Focal loss function Loss CE and cross entropy (CE) loss function Loss Focal The sum of the two is used as the total loss function of the model Loss Total , the formula is as follows:
[0114] Loss Total =Loss Dice +Loss CE +Loss Focal .
[0115]
[0116] Among them, y j is the actual value of the cerebral microbleed segmentation image mask corresponding to the j-th preprocessed cerebral microbleed susceptibility weighted image in the training set; is the predicted value of the cerebral microbleed segmentation image mask corresponding to the j-th preprocessed cerebral microbleed susceptibility weighted image in the training set; J is the number of preprocessed cerebral microbleed susceptibility weighted images in the training set.
[0117] According to the definition of the Dice loss function, the loss calculation is based on the overlap between the predicted value of the cerebral microbleeding segmentation image mask and the true value of the cerebral microbleeding segmentation image mask.
[0118] In an exemplary embodiment, a computer device is provided, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement a cerebral microbleed segmentation method based on SWI.
[0119] In an exemplary embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, a method for segmenting cerebral microbleeds based on SWI is implemented.
[0120] In an exemplary embodiment, a computer program product is provided, including a computer program, which implements a cerebral microbleed segmentation method based on SWI when executed by a processor.
[0121] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Fig. 9 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a cerebral microbleed segmentation method based on SWI is implemented.
[0122] Those skilled in the art will understand that Fig. 9 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.
[0123] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.
[0124] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present 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 may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).
[0125] The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., but is not limited thereto. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but is not limited thereto.
[0126] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0127] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.
Claims
1. A method for segmenting cerebral microbleeds based on SWI, characterized in that: The SWI-based cerebral microbleed segmentation method comprises: Acquire a target susceptibility weighted image; the target susceptibility weighted image is a susceptibility weighted image of cerebral microbleeds in the brain to be segmented; Based on the target susceptibility weighted image and the cerebral microbleeding segmentation model, the predicted value of the corresponding cerebral microbleeding segmentation image mask is determined; the cerebral microbleeding segmentation model is obtained by training CeMiS-Net using a training set; the training set includes: a plurality of preprocessed cerebral microbleeding susceptibility weighted images and the actual values of the corresponding cerebral microbleeding segmentation image mask; the CeMiS-Net is constructed based on the SAM model and the convolutional neural network.
2. The SWI-based cerebral microbleed segmentation method according to claim 1, characterized in that: Determining a prediction value of a corresponding cerebral microbleeding segmentation image mask based on the target susceptibility weighted image and the cerebral microbleeding segmentation model includes: Preprocessing the target magnetic susceptibility weighted image to obtain a preprocessed target magnetic susceptibility weighted image; the preprocessing includes: standardization, data enhancement, oversampling and noise removal; The preprocessed target susceptibility weighted image is input into the cerebral microbleeding segmentation model to obtain the prediction value of the corresponding cerebral microbleeding segmentation image mask.
3. The SWI-based cerebral microbleed segmentation method according to claim 1, characterized in that: The training process of the cerebral microbleeding segmentation model includes: Obtaining the training set; Constructing the CeMiS-Net based on the SAM model and convolutional neural network; The CeMiS-Net is trained using the training set and the total loss function to obtain the cerebral microbleed segmentation model; the total loss function is the sum of the Dice loss function, the Focal loss function and the cross entropy loss function.
4. The SWI-based cerebral microbleed segmentation method according to claim 2, characterized in that: Obtaining the training set includes: Acquire multiple cerebral microbleed susceptibility weighted images and corresponding actual values of cerebral microbleed segmentation image masks; Preprocessing each of the cerebral microbleed susceptibility weighted images respectively to obtain a corresponding preprocessed cerebral microbleed susceptibility weighted image; The actual values of all preprocessed cerebral microbleed susceptibility weighted images and the corresponding cerebral microbleed segmentation image masks were determined as the initial data set; The initial data set is divided according to a set ratio to obtain the training set, validation set and test set.
5. The SWI-based cerebral microbleed segmentation method according to claim 2, characterized in that: The CeMiS-Net includes: a SAM encoder, a convolutional neural network encoder, a feature fusion module and a convolutional neural network decoder; the CeMiS-Net adopts a multi-level skip structure; The SAM encoder is used to obtain brain image structure information corresponding to an input image; the input image is a pre-processed cerebral microbleed susceptibility weighted image or a pre-processed target susceptibility weighted image; The convolutional neural network encoder is used to obtain local information of cerebral microbleeds corresponding to the input image; The feature fusion module is used to fuse the brain image structure information and the local information of cerebral microbleeds corresponding to the input image to obtain a fusion feature map corresponding to the input image; The convolutional neural network decoder is used to obtain a corresponding cerebral microbleeding segmentation image mask based on a fused feature map corresponding to an input image.
6. The SWI-based cerebral microbleed segmentation method according to claim 5, characterized in that: The SAM encoder includes: a linear change layer, an embedded patch layer, a normalization layer, a multi-head attention layer and a multi-layer perceptron; The convolutional neural network encoder comprises: a plurality of convolutional neural network encoding blocks; The feature fusion module includes: a plurality of feature fusion submodules; the number of the feature fusion submodules is equal to the number of the convolutional neural network encoding blocks; The convolutional neural network decoder includes: a plurality of convolutional neural network decoding blocks; the number of the convolutional neural network decoding blocks is 1 less than the number of the feature fusion submodules.
7. The SWI-based cerebral microbleed segmentation method according to claim 6, characterized in that: The convolutional neural network encoding block is a decoding block formed by repeatedly stacking a first convolutional layer, a batch normalization layer, a ReLu activation function layer, and a second convolutional layer; the convolutional neural network decoding block is a decoding block formed by repeatedly stacking a deconvolutional layer, a batch normalization layer, a ReLu activation function layer, and a third convolutional layer; The convolution of the first convolutional layer is 3×3×3, and the stride is 2×2×2; The convolution of the second convolutional layer is 3×3×3, and the stride is 1×1×1; The convolution of the deconvolution layer is 3×3×3 and the stride is 2×2×2; The convolution of the third convolutional layer is 3×3×3 and the stride is 1×1×1.
8. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the SWI-based cerebral microbleed segmentation method according to any one of claims 1 to 7.
9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the SWI-based cerebral microbleed segmentation method described in any one of claims 1 to 7 is implemented.
10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the SWI-based cerebral microbleed segmentation method described in any one of claims 1 to 7 is implemented.
