WO2021151275A1

WO2021151275A1 - Image segmentation method and apparatus, device, and storage medium

Info

Publication number: WO2021151275A1
Application number: PCT/CN2020/098975
Authority: WO
Inventors: 章古月
Original assignee: 平安科技（深圳）有限公司
Priority date: 2020-05-20
Filing date: 2020-06-29
Publication date: 2021-08-05
Also published as: CN111696082A

Abstract

An image segmentation method, relating to the field of artificial intelligence, the method comprising: converting abdominal CT image data in DICOM format into an abdominal image in JPG format (S100); inputting the JPG format abdominal image into a generative network model constructed by a Vnet-based network model (S102); by means of the generative network model, generating six channels of predicted segmentation tags, the six channels of predicted segmentation tags comprising subcutaneous fat, musculature, bone, visceral fat, internal organ, and background predicted segmentation tags (S104); and on the basis of the six channels of predicted segmentation tags, obtaining predicted segmentation result images, the predicted segmentation result image comprising a subcutaneous fat image, a musculature image, a bone image, a visceral fat image, an internal organ image, and a background image (S106). In addition, the present method further relates to blockchain technology, and the DICOM format abdominal CT image data, the JPG format abdominal image can be stored in a blockchain. Thus, the effect of segmenting an abdominal musculature image and a fat image can be improved.

Description

Image segmentation method, device, equipment and storage medium

This application claims the priority of a Chinese patent application filed with the Chinese Patent Office on May 20, 2020, the application number is CN202010431606.6, and the invention title is "Image segmentation methods, devices, electronic equipment, and computer-readable storage media". The entire content is incorporated into this application by reference.

Technical field

This application relates to the field of artificial intelligence data processing, and in particular to an image segmentation method, device, equipment, and storage medium.

Background technique

The analysis of human body components such as fat and skeletal muscle is an important method of medical research. The content of fat and skeletal muscle in the human body is an important basis for evaluating individual nutritional status, and is important in clinical aspects such as patient diagnosis, treatment and prognosis. Guiding significance. At present, quantitative analysis of fat and skeletal muscle based on imaging techniques such as Computed Tomography (CT) is a widely recognized evaluation method. In particular, the skeletal muscle area, visceral fat area, subcutaneous fat area, and total abdominal fat volume in CT images of the umbilical plane have important clinical value.

The inventor found that the current common method for doctors is to segment the visceral fat and subcutaneous fat according to the threshold of the abdominal umbilical plane image, and then manually mark the dividing line of the muscle to segment the muscle image and the fat image. However, manually marking the dividing line of the muscle is very time-consuming, and the accuracy of the dividing line is not good, resulting in the problem that the segmentation of the abdominal muscle image and the fat image takes a long time and the segmentation effect is poor.

Therefore, how to provide an image processing solution based on CT abdominal images while overcoming the above shortcomings has become an urgent technical problem to be solved.

Summary of the invention

In view of this, this application proposes an image segmentation method, device, computer equipment, and computer-readable storage medium to solve the problem of long time-consuming segmentation of abdominal muscle images and fat images and low segmentation accuracy in the prior art.

First of all, in order to achieve the above objective, this application proposes an image segmentation method, which includes the steps:

Convert abdominal CT image data in DICOM format to abdominal image in JPG format;

Constructing a generation network model based on the Vnet network model, and inputting the abdominal image in JPG format into the generation network model;

Generate 6-channel predicted segmentation labels through the generation network model, where the 6-channel predicted segmentation labels include subcutaneous fat, muscle, bone, visceral fat, internal organs, and background predicted segmentation labels;

The predicted segmentation result image is obtained according to the 6-channel predicted segmentation label, where the predicted segmentation result image includes subcutaneous fat image, muscle image, bone image, visceral fat image, internal organ image, and background image.

To achieve the above objective, the present application also provides an image segmentation device, including:

Conversion module: used to convert abdominal CT image data in DICOM format into abdominal image in JPG format;

Processing module: used to construct a generation network model based on the Vnet network model, and input the abdominal image in JPG format into the generation network model;

Generating module: used to generate 6-channel predicted segmentation labels through the generation network model, where the 6-channel predicted segmentation labels include subcutaneous fat, muscle, bone, visceral fat, internal organs, and background predicted segmentation labels;

Obtaining module: used to obtain predicted segmentation result images according to the 6-channel predicted segmentation tags, where the predicted segmentation result images include subcutaneous fat images, muscle images, bone images, visceral fat images, internal organs images, and background images.

In order to achieve the above object, the present application also provides a computer device, including a memory, a processor, and a computer program stored in the memory and running on the processor, and the processor executes the computer program when the computer program is executed. The following steps:

To achieve the foregoing objective, the present application also provides a computer-readable storage medium on which a computer program is stored, wherein the computer program is executed by a processor to implement the following steps:

Compared with the prior art, the image segmentation method, device, computer equipment, and computer-readable storage medium proposed in this application input the JPG format abdominal image into the generation network model constructed based on the Vnet network model; Generate a network model to generate 6-channel predicted segmentation labels; obtain predicted segmentation result images according to the 6-channel predicted segmentation labels, where the predicted segmentation result images include subcutaneous fat images, muscle images, bone images, visceral fat images, and internal organs images , Background image. In this way, it is possible to obtain more accurate abdominal muscle images and fat images without manual labeling, reduce the time for segmentation of abdominal muscle images and fat images, and improve the segmentation effect of abdominal muscle images and fat images.

Description of the drawings

Fig. 1 is a schematic flowchart of a first embodiment of an image segmentation method according to the present application;

FIG. 2 is a schematic flowchart of step S102 of the image segmentation method of the present application;

FIG. 3 is a schematic flowchart of step S104 of the image segmentation method of the present application;

4 is a schematic diagram of an embodiment of the CA module of the image segmentation device of the present application;

5 is a schematic diagram of an embodiment of a predicted segmentation result image of the image segmentation device of the present application;

Fig. 6 is a schematic diagram of an embodiment of a gold standard image of the image segmentation device of the present application;

FIG. 7 is a schematic diagram of an embodiment of a discriminant network model of the image segmentation device according to the present application;

FIG. 8 is a schematic diagram of program modules of the first embodiment of the image segmentation device of the present application;

FIG. 9 is a schematic diagram of an embodiment of a processing module of the image segmentation device of the present application;

FIG. 10 is a schematic diagram of an embodiment of a generating module of the image segmentation device of the present application;

Fig. 11 is a schematic diagram of an optional hardware architecture of the computer device of the present application.

The realization, functional characteristics, and advantages of the purpose of this application will be further described in conjunction with the embodiments and with reference to the accompanying drawings.

Detailed ways

It should be understood that the specific embodiments described here are only used to explain the present application, and are not used to limit the present application.

This application proposes an image segmentation method. Referring to FIG. 1, it is a schematic flowchart of an image segmentation method provided by an embodiment of this application. The method can be executed by a device, and the device can be implemented by software and/or hardware.

In this embodiment, the image segmentation method includes:

In step S100, the abdominal CT image data in DICOM format is converted into an abdominal image in JPG format.

In this embodiment, the CT abdominal image data in the Digital Imaging and Communications in Medicine (DICOM) format is set to a specific window width and window level for the abdominal image, and then the CT image in DICOM format is converted through the format conversion program The data is converted to abdomen image in JPG format, and the abdomen image in JPG format is saved. It should be emphasized that in order to further ensure the privacy and security of the aforementioned DICOM format abdominal CT image data and JPG format abdominal map, the aforementioned DICOM format abdominal CT image data and JPG format abdominal map can also be stored in a blockchain In the node.

In this embodiment, the specific window width and window level for the abdominal image can be set to a window width of 400 HU and a window level of 10 HU. It is understandable that the abdominal CT image data in the DICOM format contains protected health information (PHI) of the patient, such as name, gender, age, and other image-related information, such as captured and generated images Equipment information, some medical context related information, etc. The DICOM format abdominal CT image data carries a lot of information, which can be divided into the following four categories: (a) Patient information, (b) Examination and Study information, (c) Series information, (d) Image Image information. Patient information includes patient name, patient ID, patient gender, patient weight, etc. Study information includes: inspection number, inspection instance number, inspection date, inspection time, inspection location, inspection description, etc. Series information includes serial number, inspection mode, image location, inspection description and description, image orientation, image location, layer thickness, layer-to-layer spacing, actual relative position, and body position. Image information includes information such as the time the image was taken, pixel spacing, image code, and sampling rate on the image. According to the pixel spacing, the conversion parameters between the pixel points and the physical space area can be obtained, and the actual area of the physical space corresponding to the pixel area can be calculated according to the conversion parameters.

In step S102, a generation network model is constructed based on the Vnet network model, and the abdominal image in the JPG format is input into the generation network model.

Optionally, referring to FIG. 2, the step S102 includes the following steps:

Step S1021: Set the convolution kernel in the encoding stage of the Vnet network model to a two-dimensional convolution kernel;

Step S1022, replacing the deconvolution in the decoding stage of the Vnet network model with bilinear interpolation to obtain a modified Vnet network model;

Step S1023, access the channel attention CA module in the modified Vnet network model to obtain the generative network model, where the CA module is used to obtain the encoding stage and the decoding stage of the modified Vnet network Generate semantic information of the high-level feature map, and select pixel information belonging to the high-level feature map from the low-level feature map according to the semantic information;

Wherein, the high-level feature map and the low-level feature map are determined according to the sequence of obtaining feature maps in the encoding stage and the decoding stage. Among the adjacent encoding layers in the encoding stage, the feature map obtained by the next encoding layer is higher than that of the previous encoding. The feature map obtained by the layer is higher-level; in the adjacent coding layers of the decoding stage, the feature map obtained by the previous decryption layer is lower than the feature map obtained by the next decryption layer.

In this embodiment, the Vnet network model is Fausto Milletari, Nasir Nawab, Said Ahmed Ahmad ( The medical imaging Vnet network model proposed by Seyed-Ahmad (Ahmadi) and others. The Vnet network model is a typical encoding-decoding network model. In the Vnet network model, the coding stage includes multiple coding layers, and each coding layer includes a convolutional layer, an activation layer, and a down-sampling layer. The decoding stage includes multiple decoding layers, and each decoding layer includes a deconvolution layer, an activation layer, and an upsampling layer.

The convolution kernel in the encoding stage of the Vnet network model is based on a three-dimensional convolution kernel, but the three-dimensional data is unreliable due to the thicker CT data scanning layer. In this embodiment, the convolution kernel in the encoding stage of the Vnet network model is set as a two-dimensional convolution kernel, and segmentation is performed separately based on the two-dimensional image. In this embodiment, in order to reduce the amount of learnable parameters, the deconvolution in the decoding stage of the Vnet network model is replaced with bilinear interpolation.

Step S104: Generate 6-channel predicted segmentation labels through the generation network model, where the 6-channel predicted segmentation labels include subcutaneous fat, muscle, bone, visceral fat, internal organs, and background predicted segmentation labels.

Optionally, referring to FIG. 3, the step S104 includes the following steps:

Step S1041: Obtain a feature map of each coding layer through the coding stage of the generated network model;

Step S1042: Obtain a feature map of each decoding layer through the decoding stage of the generating network model;

Step S1043: In the encoding stage, the CA module is used to channelize and activate the high-level features of the h*w*2c dimension of the next layer of the adjacent encoding layer in the encoding stage to obtain the first weight results of different channels , Multiplying the first weight results of the different channels with the low-level features of the upper 2h*2w*c dimension of the adjacent coding layer to obtain the first feature map of the 2h*2w*c dimension;

Step S1044, in the decoding stage, perform channelization operation and activation operation on the advanced features of the upper 2h*2w*c dimension of the adjacent decoding layer in the decoding stage through the CA module to obtain the second weight results of different channels Multiplying the second weight results of the different channels with the low-level features of the 2h*2w*c dimension of the next layer of the adjacent coding layer to obtain a second feature map of the 2h*2w*c dimension;

Step S1045: Obtain the 6-channel prediction segmentation according to the feature map obtained in each layer of the encoding stage, the feature map obtained in each layer of the decoding stage, the first feature map, and the second feature map. Label.

In this embodiment, in the coding stage of the generated network model, the convolution operation is performed on the convolutional layer to extract features from the input abdominal CT image. After each layer of the coding stage is completed, an appropriate stride is used to reduce the resolution. If the resolution of the previous layer is 2h*2w, the resolution of the next layer is reduced to h*w. In this embodiment, the features of the next layer in the encoding stage of the generative network model are doubled compared to the features of the previous layer. If the number of features of the previous layer in the encoding stage of the generative network model is c, then the next layer The number of features is 2c.

In this embodiment, the feature map of each coding layer is obtained through the coding stage of the generating network model. In the coding stage, the feature map obtained by the next coding layer of the adjacent coding layer is higher than the feature map obtained by the previous coding layer. To be advanced. The high-level features acquired in the next layer of the adjacent coding layer in the coding stage of the generative network model are high-level features of h*w*2c dimensions, where h represents the height of the graph, w represents the width of the graph, and 2c represents the number of features . The low-level features obtained by the upper layer of the adjacent coding layer in the coding stage of the generating network model are low-level features of dimension 2h*2w*c, 2h represents the height of the graph, 2w represents the width of the graph, and c represents the number of features.

In this embodiment, in the decoding stage of the generated network model, each input voxel is projected to a larger area through the kernel through the deconvolution layer to increase the data size. If the resolution of the previous layer is h*w, then The resolution of the next layer is increased to 2h*2w. In this embodiment, the features of the next layer in the decoding stage of the generative network model are twice as small as the features of the previous layer. If the number of features of the previous layer in the encoding stage of the generative network model is 2c, then the next layer The number of features is c.

In this embodiment, the feature map of each decoding layer is obtained through the decoding stage of the generative network model. Among the adjacent coding layers in the decoding stage, the feature map obtained by the upper decryption layer is higher than that obtained by the next decryption layer. The feature map should be low-level. In this embodiment, the high-level features acquired by the upper layer of the adjacent decoding layer in the decoding stage of the generative network model are high-level features of h*w*2c dimensions, where h represents the height of the graph, and w represents the height of the graph. Wide, 2c represents the number of features. The low-level features acquired in the next layer of the adjacent decoding layer in the decoding code stage of the generating network model are low-level features of dimension 2h*2w*c, where 2h represents the height of the graph, 2w represents the width of the graph, and c represents the number of features.

It should be noted that as the coding process continues to deepen, the obtained feature expressions gradually become richer. However, due to multiple convolution processes and the application of non-linear functions, a large amount of position information in the high-level feature map is lost, resulting in the phenomenon of misclassification of a large number of pixels. A Channel-Attention (CA) module is connected to the modified Vnet network, and the misclassified pixels are corrected through the CA module.

In step S1043, channelizing and activating the high-level features of the next layer of h*w*2c dimensions of the adjacent coding layer in the coding stage through the CA module to obtain the first weight results of different channels includes The following steps:

The high-level features of the h*w*2c dimension of the next layer of the adjacent coding layer are passed through the global average pooling, 1*1 convolution, batch normalization (BN) algorithm model, and nonlinearity of the CA module. Rectified Linear Units, ReLu) activation function to obtain 1*1*c feature channel, c represents the number of features; pass the 1*1*c feature channel through the fully connected layer and sigmoid activation function to obtain the first of different channels Weighted result.

In step S1044, channelizing and activating the advanced features of the upper 2h*2w*c dimension of the adjacent decoding layer of the decoding stage through the CA module to obtain the second weight results of different channels includes The following steps: The advanced features of the h*w*2c dimension of the upper layer of the adjacent decoding layer are obtained through the global average pooling, 1*1 convolution, BN algorithm model, and ReL activation function of the CA module to obtain 1* 1*c feature channel, c represents the number of features; pass the 1*1*c feature channel through the fully connected layer and the sigmoid activation function to obtain the second weight results of different channels.

Referring to FIG. 4, the processing flow of the CA module mainly includes channelization operation, activation operation, and weight assignment reweighting operation. In the encoding stage, the CA module is used to channelize the advanced features of the next layer of the adjacent encoding layer in the encoding stage, where the channelization operation includes: converting the advanced features of the next layer of the adjacent encoding layer Through the global average pooling of the CA module, the 1*1 convolution, the BN algorithm model, and the ReLu activation function, a feature channel of 1*1*c is obtained, and c represents the number of features; the 1*1*c The feature channel performs an activation operation, where the activation operation includes: passing the 1*1*c feature channel through a fully connected layer and a sigmoid activation function to obtain the weight results of different channels; and comparing the weight results of the different channels with The low-level features of the upper layer of adjacent coding layers are multiplied to obtain a first feature map, and the first feature map is a 2h*2w*c dimension feature map.

In the decoding stage, the CA module is used to channelize the advanced features of the upper layer of the adjacent decoding layer in the decoding stage, where the channelization operation includes: Through the global average pooling of the CA module, the 1*1 convolution, the BN algorithm model, and the ReLu activation function, a feature channel of 1*1*c is obtained, and c represents the number of features; the 1*1*c The feature channel performs an activation operation, where the activation operation includes: passing the 1*1*c feature channel through a fully connected layer and a sigmoid activation function to obtain the weight results of different channels; and comparing the weight results of the different channels with The low-level features of the next layer of adjacent coding layers are multiplied to obtain a second feature map. The second feature map is a 2h*2w*c dimension feature map.

Step S106: Obtain a predicted segmentation result image according to the 6-channel predicted segmentation label, where the predicted segmentation result image includes subcutaneous fat images, muscle images, bone images, visceral fat images, internal organs images, and background images.

In this embodiment, the predicted segmentation labels of the 6-channel respectively represent the predicted segmentation labels of subcutaneous fat, muscle, bone, visceral fat, internal organs, and background, which are filled with different colors to obtain the predicted segmentation result image. For example, you can use Red draws subcutaneous fat, green draws muscle, yellow draws bones, blue draws visceral fat, pink draws internal organs, and black draws background. Please refer to Figure 5. In Figure 5, different gray-scale colors are used to represent the six categories of subcutaneous fat, muscle, bone, visceral fat, internal organs, and background.

In this way, it is possible to obtain more accurate abdominal muscle images and fat images without manual labeling, reduce the segmentation time of abdominal muscle images and fat images, and improve the segmentation effect of abdominal muscle images and fat images.

Optionally, the image segmentation method further includes:

Determine the number of pixels in the subcutaneous fat region, visceral fat region, and muscle region from the predicted segmentation result image, and determine the subcutaneous fat, visceral fat, and muscle based on the determined number of pixels and pre-acquired physical space conversion parameters The actual area.

In this embodiment, the number of pixels in the subcutaneous fat area, visceral fat area, and muscle area is determined from the predicted segmentation result image, and the difference between the pixel points and the physical space area is obtained from the CT image data in the DICOM format. The conversion parameter is to determine the actual area of subcutaneous fat, visceral fat, and muscle based on the number of pixels in the subcutaneous fat area, visceral fat, and muscle area multiplied by the square of the conversion parameter.

It is further explained that the image information of the CT image data in the DICOM format includes information such as the time when the image was taken, the pixel spacing, the image code, and the sampling rate on the image. According to the pixel spacing, the conversion parameters between the pixel points and the physical space area can be obtained, and the actual areas of subcutaneous fat, visceral fat, and muscle can be calculated according to the following formula (1). Formula (1) s=n*x^2, where s represents the actual area of subcutaneous fat, visceral fat, and muscle, n represents the total number of pixels in the subcutaneous fat area, visceral fat area, and muscle area, and x represents the conversion parameter .

In this way, accurate abdominal fat and muscle area can be obtained, and the accuracy of actual fat and muscle area can be improved.

Optionally, the image segmentation method further includes:

Obtain scanning layer thickness information from the abdominal CT image data, and multiply the actual area of the subcutaneous fat, visceral fat, and muscle by the scanning layer thickness to obtain the actual volume of the subcutaneous fat, visceral fat, and muscle.

In this embodiment, the Series information of the abdominal CT image data in the DICOM format includes serial number, inspection modality, image location, inspection description and description, image orientation, image location, layer thickness, layer-to-layer spacing , Actual relative position and body position, etc. Therefore, the scanning layer thickness information can be obtained from the CT image data in the DICOM format. The actual area of the subcutaneous fat area, visceral fat area, and muscle area is multiplied by the scanning layer thickness to obtain the actual volume of the subcutaneous fat, visceral fat, and muscle.

Optionally, the image segmentation method further includes:

The predicted segmentation label and the real label corresponding to the gold standard image are respectively input into the discriminant network model, and the discriminant scores of the predicted segmentation result image and the gold standard image are obtained respectively, and the predicted segmentation is determined according to the discriminant scores Based on the gap between the result image and the gold standard image, parameter adjustment is performed on the generation network model based on the gap, so as to optimize the generation network model.

In this way, the generation network model can be optimized by adjusting the parameters of the generation network model, so as to improve the effect of abdominal image segmentation.

In this embodiment, the gold standard image is a segmentation result manually annotated by a person, and is used to compare with the result of network estimation to evaluate the performance of the generated network model. The gold standard image uses different colors to represent subcutaneous fat, muscle, bone, visceral fat, internal organs, and background. Please refer to Figure 7. Figure 7 is a gold standard image representing subcutaneous fat, muscle, bone, visceral fat, internal organs, and background areas in different grayscale colors.

Please refer to FIG. 7, which is a schematic diagram of the architecture of the discriminant network model. The discriminant network model includes 6 convolutional layers. The first convolutional layer 802 includes a 3*3 convolutional layer and a nonlinear ReLu activation function; the second convolutional layer 803 includes a 3*3 convolutional layer and a batch standardized algorithm model. , Non-linear ReLu activation function; the third convolutional layer 804 includes 3*3 convolutional layer, batch standardized algorithm model, nonlinear ReLu activation function; the fourth convolutional layer 805 includes 3*3 convolutional layer, batch standardized algorithm model , Non-linear ReLu activation function; The fifth convolutional layer 806 includes 3*3 convolutional layers, batch standardized algorithm models, and nonlinear ReLu activation functions; The sixth convolutional layer 807 includes global average pooling, 1*1 convolutional layer . 801 represents the predicted segmentation label of 512*512*6 dimensions or the real label corresponding to the gold standard image.

In this embodiment, the predicted segmentation label of 512*512*6 dimensions and the real label corresponding to the gold standard image are input into the discriminant network model, and a convolution operation with a size of 3 and a step size of 2 is used for down-sampling. The number of downsampling corresponds to the number of downsampling of the encoder in the generative network model. A total of 5 downsampling is obtained to obtain a 16*16*256 feature map. Finally, the global average pooling and 1*1 convolution kernel are used to obtain gold The discriminant scores of the standard image and the predicted segmented image.

In this embodiment, the optimization of the KL divergence (Kullback Leibler divergence) between the predicted label result image and the gold standard image is adjusted to the optimization of the bulldozer distance (Earth Mover distance), and the bulldozer distance can be always Guide the optimization of the generative network model without being troubled by the disappearance of the gradient.

In this embodiment, gradient penalty is used to accelerate the convergence of the training process of the generating network model and the discriminant network model. The gradient penalty of the zero center is easier to converge to the center point, so the gradient penalty of the zero center is used.

In this embodiment, the generating network model and the discriminant network model each have a corresponding loss function.

The loss function of the generated network model is as follows:

in,

λ=0.001,

The loss function of the discriminant network model is as follows:

Where c=0, λ=10, and p _inter (I ^inter ) is a derivative distribution obtained by interpolation between the true sample distribution and the false sample distribution.

The following is a Chinese description of the loss function in English: Loss: loss; Orig: original image; Dice: dice coefficient; Gen: generating network model; I: image; Mask; mask; D: discriminating network model; G: generating network Model; p_g: false sample distribution; p_train: true sample distribution; P_inter: derived distribution obtained by interpolation between true sample distribution and false sample distribution; C: center, and C equals 0 to zero center.

The generative network model and the discriminant network model reduce the values of these two loss functions through continuous learning to achieve the goal of optimization.

The image segmentation method proposed in this application inputs the JPG format abdominal image into a generative network model constructed based on the Vnet network model; generates 6-channel predicted segmentation labels through the generative network model; according to the 6-channel prediction The segmentation label obtains the predicted segmentation result image, where the predicted segmentation result image includes subcutaneous fat image, muscle image, bone image, visceral fat image, internal organ image, and background image. In this way, it is possible to obtain more accurate abdominal muscle images and fat images without manual labeling, reduce the time for segmentation of abdominal muscle images and fat images, and improve the segmentation effect of abdominal muscle images and fat images.

Referring to FIG. 8, it is an image segmentation device 100 proposed in the present application.

The image segmentation device 100 described in this application can be installed in a computer device. According to the implemented functions, the image segmentation device may include a conversion module 101, a processing module 102, a generation module 103, and an acquisition module 104. The module described in the present invention can also be called a unit, which refers to a series of computer program segments that can be executed by the processor of a computer device and can complete fixed functions, and are stored in the memory of the computer device.

In this embodiment, the functions of each module/unit are as follows:

The conversion module 101 is used to convert the abdominal CT image data in DICOM format into the abdominal image in JPG format.

In this embodiment, the specific window width and window level for the abdominal image can be set to a window width of 400 HU and a window level of 10 HU. It is understandable that the abdominal CT image data in the DICOM format contains protected health information (PHI) of the patient, such as name, gender, age, and other image-related information, such as captured and generated images Equipment information, some medical context related information, etc. The DICOM format abdominal CT image data carries a lot of information, which can be divided into the following four categories: (a) Patient information, (b) Examination and Study information, (c) Series information, (d) Image Image information. Patient information includes patient name, patient ID, patient gender, patient weight, etc. Study information includes: inspection number, inspection instance number, inspection date, inspection time, inspection location, inspection description, etc. Series information includes serial number, inspection mode, image location, inspection description and instructions, image orientation, image location, layer thickness, layer-to-layer spacing, actual relative position, and body position. Image information includes information such as the time the image was taken, pixel spacing, image code, and sampling rate on the image. According to the pixel spacing, the conversion parameters between the pixel points and the physical space area can be obtained, and the actual area of the physical space corresponding to the pixel area can be calculated according to the conversion parameters.

The processing module 102 is configured to construct a generation network model based on the Vnet network model, and input the abdominal image in JPG format into the generation network model.

Optionally, referring to FIG. 9, the processing module 102 includes:

A setting sub-module 1021 is used to set the convolution kernel in the encoding stage of the Vnet network model to a two-dimensional convolution kernel;

The replacement sub-module 1022 is used to replace the deconvolution in the decoding stage of the Vnet network model with bilinear interpolation to obtain a modified Vnet network model;

The access sub-module 1023 is used to access the channel attention CA module in the modified Vnet network model to obtain the generated network model, wherein the CA module is used to obtain the modified Vnet network model Semantic information of the high-level feature map generated in the encoding stage and the decoding stage, and selecting pixel information belonging to the high-level feature map from the low-level feature map according to the semantic information;

The generating module 103 is configured to generate 6-channel predicted segmentation labels through the generation network model, where the 6-channel predicted segmentation labels include subcutaneous fat, muscle, bone, visceral fat, internal organs, and background predicted segmentation labels .

Optionally, referring to FIG. 10, the generating module 103 includes:

The first obtaining sub-module 1031 is configured to obtain the feature map of each coding layer through the coding stage of the generated network model;

The second obtaining sub-module 1032 is configured to obtain the feature map of each decoding layer through the decoding stage of the generation network model;

The first processing sub-module 1033 is used for channelizing and activating the high-level features of the h*w*2c dimension of the next layer of the adjacent coding layer in the coding phase through the CA module in the coding phase to obtain different The first weight result of the channel, the first weight result of the different channels is multiplied by the low-level features of the upper 2h*2w*c dimension of the adjacent coding layer to obtain the first feature map of the 2h*2w*c dimension ；

The second processing sub-module 1034 is used to perform channelization operation and activation operation on the advanced features of the upper 2h*2w*c dimension of the adjacent decoding layer in the decoding stage through the CA module in the decoding stage to obtain different The second weight result of the channel; the second weight result of the different channel is multiplied by the low-level features of the next layer of the adjacent coding layer with 2h*2w*c dimensions to obtain a second feature map of 2h*2w*c dimensions ；

The third processing sub-module 1035 is configured to obtain all the features according to the feature maps obtained in each layer of the encoding stage, the feature maps obtained in each layer of the decoding stage, the first feature map, and the second feature map. The 6-channel prediction segmentation label.

In this embodiment, the feature map of each coding layer is obtained through the coding stage of the generating network model. In the coding stage, the feature map obtained by the next coding layer of the adjacent coding layer is higher than the feature map obtained by the previous coding layer. To be advanced. The high-level features acquired by the next layer of the adjacent coding layer in the coding stage of the generative network model are high-level features of h*w*2c dimensions, where h represents the height of the graph, w represents the width of the graph, and 2c represents the number of features . The low-level features obtained by the upper layer of the adjacent coding layer in the coding stage of the generating network model are low-level features of dimension 2h*2w*c, 2h represents the height of the graph, 2w represents the width of the graph, and c represents the number of features.

The first processing submodule 1033 is also used to pass the high-level features of the h*w*2c dimension of the next layer of the adjacent coding layer through the global average pooling, 1*1 convolution, and batch normalization of the CA module (Batch Normalization, BN) algorithm model and non-linear (Rectified Linear Units, ReLu) activation function to obtain a 1*1*c feature channel, where c represents the number of features; the 1*1*c feature channel is fully connected Layer and sigmoid activation function to obtain the first weight results of different channels.

The second processing submodule 1034 is also used to pass the h*w*2c-dimensional high-level features of the upper layer of the adjacent decoding layer through the global average pooling, 1*1 convolution, and BN algorithm of the CA module The model and the ReL activation function are used to obtain 1*1*c feature channels, where c represents the number of features; the 1*1*c feature channels are passed through the fully connected layer and the sigmoid activation function to obtain the second weight results of different channels.

Please refer to FIG. 4 again. The processing flow of the CA module mainly includes a channelized Channelization operation, an activation operation, and a weight assignment reweighting operation. In the encoding stage, the CA module is used to channelize the advanced features of the next layer of the adjacent encoding layer in the encoding stage, where the channelization operation includes: converting the advanced features of the next layer of the adjacent encoding layer Through the global average pooling of the CA module, the 1*1 convolution, the BN algorithm model, and the ReLu activation function, a feature channel of 1*1*c is obtained, and c represents the number of features; the 1*1*c The feature channel performs an activation operation, where the activation operation includes: passing the 1*1*c feature channel through a fully connected layer and a sigmoid activation function to obtain the weight results of different channels; and comparing the weight results of the different channels with The low-level features of the upper layer of adjacent coding layers are multiplied to obtain a first feature map, and the first feature map is a 2h*2w*c dimension feature map.

The acquisition module 104 is configured to obtain a predicted segmentation result image according to the 6-channel predicted segmentation label, where the predicted segmentation result image includes subcutaneous fat image, muscle image, bone image, visceral fat image, internal organ image, Background image.

Optionally, the image segmentation device 100 further includes:

The determining module is used to determine the number of pixels in the subcutaneous fat area, the visceral fat area, and the muscle area from the predicted segmentation result image, and determine the subcutaneous fat based on the determined number of pixels and the pre-acquired physical space conversion parameters , The actual area of visceral fat and muscle.

Optionally, the image segmentation device 100 further includes:

The calculation module is used to obtain scan layer thickness information from the abdominal CT image data, and multiply the actual area of the subcutaneous fat, visceral fat, and muscle by the scan layer thickness to obtain the actual area of the subcutaneous fat, visceral fat, and muscle. volume.

Optionally, the image segmentation device 100 further includes:

The optimization module is configured to input the predicted segmentation label and the real label corresponding to the gold standard image into the discriminant network model to obtain the discriminant scores of the predicted segmentation result image and the gold standard image, respectively, based on the discriminant scores Determine the gap between the predicted segmentation result image and the gold standard image, and adjust the parameters of the generation network model based on the gap to optimize the generation network model.

In this embodiment, the gold standard image is a segmentation result manually annotated by a person, and is used to compare with the result of network estimation to evaluate the performance of the generated network model. The gold standard image uses different colors to represent subcutaneous fat, muscle, bone, visceral fat, internal organs, and background. Please refer to 6. Figure 6 is a gold standard image representing subcutaneous fat, muscle, bone, visceral fat, internal organs, and background areas with different grayscale colors.

The loss function of the generated network model is as follows:

in,

λ=0.001,

The loss function of the discriminant network model is as follows:

The image segmentation device proposed in this application inputs the JPG format abdominal image into a generation network model constructed based on the Vnet network model; generates a 6-channel prediction segmentation label through the generation network model; according to the 6-channel prediction The segmentation label obtains the predicted segmentation result image, where the predicted segmentation result image includes subcutaneous fat image, muscle image, bone image, visceral fat image, internal organ image, and background image. In this way, it is possible to obtain more accurate abdominal muscle images and fat images without manual labeling, reduce the time for segmentation of abdominal muscle images and fat images, and improve the segmentation effect of abdominal muscle images and fat images.

As shown in FIG. 10, it is a schematic structural diagram of a computer device for implementing the image segmentation method of the present application.

The computer device 1 may include a processor 10, a memory 11, and a bus, and may also include a computer program stored in the memory 11 and running on the processor 10, such as abdomen CT image segmentation based on a Vnet network model. Procedure 12.

Wherein, the memory 11 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, mobile hard disk, multimedia card, card-type memory (such as SD or DX memory, etc.), magnetic memory, magnetic disk, CD etc. The memory 11 may be an internal storage unit of the computer device 1 in some embodiments, for example, a mobile hard disk of the computer device 1. In other embodiments, the memory 11 may also be an external storage device of the computer device 1, such as a plug-in mobile hard disk, a smart media card (SMC), and a secure digital (Secure Digital) equipped on the computer device 1. , SD) card, flash card (Flash Card), etc. Further, the memory 11 may also include both an internal storage unit of the computer device 1 and an external storage device. The memory 11 can not only be used to store application software and various data installed in the computer device 1, such as the code of the abdominal CT image segmentation program based on the Vnet network model, etc., but also can be used to temporarily store the output that has been output or will be output. data.

The processor 10 may be composed of integrated circuits in some embodiments, for example, may be composed of a single packaged integrated circuit, or may be composed of multiple integrated circuits with the same function or different functions, including one or more Combinations of central processing unit (CPU), microprocessor, digital processing chip, graphics processor, and various control chips, etc. The processor 10 is the control unit of the computer device, which uses various interfaces and lines to connect the various components of the entire computer device, and runs or executes programs or modules stored in the memory 11 (for example, based on The abdominal CT image segmentation program of the Vnet network model, etc.), and call the data stored in the memory 11 to execute various functions of the computer device 1 and process data.

The bus may be a peripheral component interconnect standard (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The bus can be divided into address bus, data bus, control bus and so on. The bus is configured to implement connection and communication between the memory 11 and at least one processor 10 and the like.

FIG. 10 only shows a computer device with components. Those skilled in the art can understand that the structure shown in FIG. 10 does not constitute a limitation on the computer device 1, and may include fewer or more components than shown in the figure. Components, or combinations of certain components, or different component arrangements.

For example, although not shown, the computer device 1 may also include a power source (such as a battery) for supplying power to various components. Preferably, the power source may be logically connected to the at least one processor 10 through a power management device, thereby controlling power The device implements functions such as charge management, discharge management, and power consumption management. The power supply may also include any components such as one or more DC or AC power supplies, recharging devices, power failure detection circuits, power converters or inverters, and power status indicators. The computer device 1 may also include various sensors, Bluetooth modules, Wi-Fi modules, etc., which will not be repeated here.

Further, the computer device 1 may also include a network interface. Optionally, the network interface may include a wired interface and/or a wireless interface (such as a WI-FI interface, a Bluetooth interface, etc.), which is usually used in the computer equipment 1 Establish a communication connection with other computer equipment.

Optionally, the computer device 1 may also include a user interface. The user interface may be a display (Display) and an input unit (such as a keyboard (Keyboard)). Optionally, the user interface may also be a standard wired interface or a wireless interface. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or the like. Among them, the display can also be called a display screen or a display unit as appropriate, and is used to display the information processed in the computer device 1 and to display a visualized user interface.

It should be understood that the embodiments are only for illustrative purposes, and are not limited by this structure in the scope of the patent application.

The abdominal CT image segmentation program 12 based on the Vnet network model stored in the memory 11 in the computer device 1 is a combination of multiple instructions. When running in the processor 10, it can realize:

Specifically, for the specific implementation method of the above-mentioned instructions by the processor 10, reference may be made to the description of the relevant steps in the embodiment corresponding to FIG. 1, which will not be repeated here. It should be emphasized that, in order to further ensure the privacy and security of the aforementioned DICOM format abdominal CT image data and the JPG format abdominal image, the DICOM format abdominal CT image data and the JPG format abdominal image can also be stored In a node of a blockchain.

Further, if the integrated module/unit of the computer device 1 is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer readable storage medium. The computer-readable storage medium may be non-volatile or volatile. The computer-readable medium may include: any entity or device capable of carrying the computer program code, a recording medium, a U disk, or a mobile hard disk , Floppy disks, compact discs.

In the several embodiments provided in this application, it should be understood that the disclosed equipment, device, and method may be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the modules is only a logical function division, and there may be other division methods in actual implementation.

The modules described as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the objectives of the solutions of the embodiments.

In addition, the functional modules in the various embodiments of the present application may be integrated into one processing unit, or each unit may exist alone physically, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware, or may be implemented in the form of hardware plus software functional modules.

For those skilled in the art, it is obvious that the present application is not limited to the details of the foregoing exemplary embodiments, and the present application can be implemented in other specific forms without departing from the spirit or basic characteristics of the present application.

The blockchain referred to in this application is a new application mode of computer technology such as distributed data storage, point-to-point transmission, consensus mechanism, and encryption algorithm. Blockchain, essentially a decentralized database, is a series of data blocks associated with cryptographic methods. Each data block contains a batch of network transaction information for verification. The validity of the information (anti-counterfeiting) and the generation of the next block.

In addition, it is obvious that the word "including" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices stated in the system claims can also be implemented by one unit or device through software or hardware. The second class words are used to indicate names, and do not indicate any specific order.

Claims

An image segmentation method, wherein the method includes the steps:

Convert abdominal CT image data in DICOM format to abdominal image in JPG format;

Constructing a generation network model based on the Vnet network model, and inputting the abdominal image in JPG format into the generation network model;

Generate 6-channel predicted segmentation labels through the generation network model, where the 6-channel predicted segmentation labels include subcutaneous fat, muscle, bone, visceral fat, internal organs, and background predicted segmentation labels;

The predicted segmentation result image is obtained according to the 6-channel predicted segmentation label, where the predicted segmentation result image includes subcutaneous fat image, muscle image, bone image, visceral fat image, internal organ image, and background image.
The image segmentation method according to claim 1, wherein the abdominal CT image data in DICOM format and the abdominal image in JPG format are stored in a blockchain, and the generating network model based on the Vnet network model includes the following step:

Setting the convolution kernel in the encoding stage of the Vnet network model to a two-dimensional convolution kernel;

Replacing the deconvolution in the decoding stage of the Vnet network model with bilinear interpolation to obtain a modified Vnet network model;

The channel attention CA module is connected to the modified Vnet network model to obtain the generative network model, where the CA module is used to obtain the advanced level generated during the encoding and decoding stages of the modified Vnet network. The semantic information of the feature map, and selecting pixel information belonging to the high-level feature map from the low-level feature map according to the semantic information;

Wherein, the high-level feature map and the low-level feature map are determined according to the sequence of obtaining feature maps in the encoding stage and the decoding stage. Among the adjacent encoding layers in the encoding stage, the feature map obtained by the next encoding layer is higher than that of the previous encoding. The feature map obtained by the layer is higher-level; in the adjacent coding layers of the decoding stage, the feature map obtained by the previous decryption layer is lower than the feature map obtained by the next decryption layer.
8. The image segmentation method according to claim 1, wherein said generating a 6-channel predicted segmentation label through said generating network model comprises the following steps:

Acquiring a feature map of each coding layer through the coding stage of the generating network model;

Acquiring a feature map of each decoding layer through the decoding stage of the generating network model;

In the encoding stage, the CA module is used to channelize and activate the high-level features of the h*w*2c dimension of the next layer of the adjacent encoding layer in the encoding stage to obtain the first weight results of different channels, and then The first weight results of the different channels are multiplied by the low-level features of the upper 2h*2w*c dimension of the adjacent coding layer to obtain the first feature map of the 2h*2w*c dimension;

In the decoding stage, the CA module is used to channelize and activate the advanced features of the upper 2h*2w*c dimension of the adjacent decoding layer in the decoding stage to obtain the second weight results of different channels; The second weight results of the different channels are multiplied by the low-level features of the next layer of 2h*2w*c dimensions of the adjacent coding layer to obtain a second feature map of 2h*2w*c dimensions;

According to the feature map obtained in each layer of the encoding stage, the feature map obtained in each layer of the decoding stage, the first feature map, and the second feature map, the 6-channel prediction segmentation label is obtained.
The image segmentation method according to any one of claims 1 to 3, wherein after the predicted segmentation result image is obtained according to the predicted segmentation label of the 6 channels, the method further comprises the following steps:

Determine the number of pixels in the subcutaneous fat region, visceral fat region, and muscle region from the predicted segmentation result image, and determine the subcutaneous fat, visceral fat, and muscle based on the determined number of pixels and pre-acquired physical space conversion parameters The actual area.
The image segmentation method of claim 4, wherein after the actual area of subcutaneous fat, visceral fat, and muscle is determined according to the determined number of pixels and the physical space conversion parameters obtained in advance, the method further It includes the following steps:

Obtain scanning layer thickness information from the abdominal CT image data, and multiply the actual area of the subcutaneous fat, visceral fat, and muscle by the scanning layer thickness to obtain the actual volume of the subcutaneous fat, visceral fat, and muscle.
The image segmentation method of claim 5, wherein the actual area of the subcutaneous fat area, visceral fat area, and muscle area is multiplied by the scanning layer thickness to obtain the actual area of the subcutaneous fat, visceral fat, and muscle. After volume, the method further includes the following steps:

The real labels corresponding to the predicted segmentation label and the gold standard image are respectively input into the discriminant network model to obtain the discriminant scores of the predicted segmentation result image and the gold standard image respectively, and the discriminant scores are used to determine the Predict the gap between the segmentation result image and the gold standard image, and adjust the parameters of the generation network model based on the gap to optimize the generation network model.
3. The image segmentation method of claim 2, wherein the information of the CT image data in the DICOM format includes the time when the image was taken, the pixel pitch, the image code, and the sampling rate on the image.
An image segmentation device, wherein the device includes:

Conversion module: used to convert abdominal CT image data in DICOM format into abdominal image in JPG format;

Processing module: used to construct a generation network model based on the Vnet network model, and input the abdominal image in JPG format into the generation network model;

Generating module: used to generate 6-channel predicted segmentation labels through the generation network model, where the 6-channel predicted segmentation labels include subcutaneous fat, muscle, bone, visceral fat, internal organs, and background predicted segmentation labels;

Obtaining module: used to obtain predicted segmentation result images according to the 6-channel predicted segmentation tags, where the predicted segmentation result images include subcutaneous fat images, muscle images, bone images, visceral fat images, internal organs images, and background images.
A computer device includes a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor implements the following steps when the processor executes the computer program:

Convert abdominal CT image data in DICOM format to abdominal image in JPG format;

Constructing a generation network model based on the Vnet network model, and inputting the abdominal image in JPG format into the generation network model;

Generate 6-channel predicted segmentation labels through the generation network model, where the 6-channel predicted segmentation labels include subcutaneous fat, muscle, bone, visceral fat, internal organs, and background predicted segmentation labels;

The predicted segmentation result image is obtained according to the 6-channel predicted segmentation label, where the predicted segmentation result image includes subcutaneous fat image, muscle image, bone image, visceral fat image, internal organ image, and background image.
The computer device according to claim 9, wherein the abdomen CT image data in DICOM format and the abdominal image in JPG format are stored in a blockchain, and the building and generating network model based on the Vnet network model comprises the following steps :

Setting the convolution kernel in the encoding stage of the Vnet network model as a two-dimensional convolution kernel;

Replacing the deconvolution in the decoding stage of the Vnet network model with bilinear interpolation to obtain a modified Vnet network model;

The channel attention CA module is connected to the modified Vnet network model to obtain the generative network model, where the CA module is used to obtain the advanced level generated during the encoding and decoding stages of the modified Vnet network. The semantic information of the feature map, and selecting pixel information belonging to the high-level feature map from the low-level feature map according to the semantic information;

Wherein, the high-level feature map and the low-level feature map are determined according to the sequence of obtaining feature maps in the encoding stage and the decoding stage. Among the adjacent encoding layers in the encoding stage, the feature map obtained by the next encoding layer is higher than that of the previous encoding. The feature map obtained by the layer is higher-level; in the adjacent coding layers of the decoding stage, the feature map obtained by the previous decryption layer is lower than the feature map obtained by the next decryption layer.
9. The computer device according to claim 9, wherein said generating a 6-channel predicted segmentation label through said generating network model comprises the following steps:

Acquiring a feature map of each coding layer through the coding stage of the generating network model;

Acquiring a feature map of each decoding layer through the decoding stage of the generating network model;

In the encoding stage, the CA module is used to channelize and activate the high-level features of the h*w*2c dimension of the next layer of the adjacent encoding layer in the encoding stage to obtain the first weight results of different channels, and then The first weight results of the different channels are multiplied by the low-level features of the upper 2h*2w*c dimension of the adjacent coding layer to obtain the first feature map of the 2h*2w*c dimension;

In the decoding stage, through the CA module, channelize and activate the high-level features of the upper 2h*2w*c dimension of the adjacent decoding layer in the decoding stage to obtain the second weight results of different channels; The second weight results of the different channels are multiplied by the low-level features of the 2h*2w*c dimension of the next layer of the adjacent coding layer to obtain the second feature map of the 2h*2w*c dimension;

According to the feature map obtained at each layer of the encoding stage, the feature map obtained at each layer of the decoding stage, the first feature map, and the second feature map, the 6-channel prediction segmentation label is obtained.
The computer device according to any one of claims 9 to 11, wherein after the predicted segmentation result image is obtained according to the predicted segmentation label of the 6 channels, the processor further implements the following steps when executing the computer program :

Determine the number of pixels in the subcutaneous fat region, visceral fat region, and muscle region from the predicted segmentation result image, and determine the subcutaneous fat, visceral fat, and muscle based on the determined number of pixels and pre-acquired physical space conversion parameters The actual area.
The computer device according to claim 12, wherein, after determining the actual areas of subcutaneous fat, visceral fat, and muscle according to the determined number of pixels and pre-acquired physical space conversion parameters, the method further comprises The following steps:

Obtain scanning layer thickness information from the abdominal CT image data, and multiply the actual area of the subcutaneous fat, visceral fat, and muscle by the scanning layer thickness to obtain the actual volume of the subcutaneous fat, visceral fat, and muscle.
The computer device of claim 13, wherein the actual area of the subcutaneous fat area, visceral fat area, and muscle area is multiplied by the scanning layer thickness to obtain the actual volume of the subcutaneous fat, visceral fat, and muscle After that, the processor further implements the following steps when executing the computer program:

The real labels corresponding to the predicted segmentation label and the gold standard image are respectively input into the discriminant network model to obtain the discriminant scores of the predicted segmentation result image and the gold standard image respectively, and the discriminant scores are used to determine the Predict the gap between the segmentation result image and the gold standard image, and adjust the parameters of the generation network model based on the gap to optimize the generation network model.
10. The computer device of claim 10, wherein the information of the CT image data in the DICOM format includes the time when the image was taken, the pixel pitch, the image code, and the sampling rate on the image.
A computer-readable storage medium having a computer program stored on the computer-readable storage medium, wherein, when the computer program is executed by a processor, the following steps are implemented:

Convert abdominal CT image data in DICOM format to abdominal image in JPG format;

Constructing a generation network model based on the Vnet network model, and inputting the abdominal image in JPG format into the generation network model;

Generate 6-channel predicted segmentation labels through the generation network model, where the 6-channel predicted segmentation labels include subcutaneous fat, muscle, bone, visceral fat, internal organs, and background predicted segmentation labels;

The predicted segmentation result image is obtained according to the 6-channel predicted segmentation label, where the predicted segmentation result image includes subcutaneous fat image, muscle image, bone image, visceral fat image, internal organ image, and background image.
16. The computer-readable storage medium according to claim 16, wherein after the predicted segmentation result image is obtained according to the predicted segmentation label of the 6 channels, the computer program further implements the following steps when being executed by the processor:

Determine the number of pixels in the subcutaneous fat region, visceral fat region, and muscle region from the predicted segmentation result image, and determine the subcutaneous fat, visceral fat, and muscle based on the determined number of pixels and pre-acquired physical space conversion parameters The actual area.
17. The computer-readable storage medium of claim 17, wherein after the actual areas of subcutaneous fat, visceral fat, and muscle are determined according to the determined number of pixels and pre-acquired physical space conversion parameters, the When the computer program is executed by the processor, the following steps are also implemented:

Obtain scanning layer thickness information from the abdominal CT image data, and multiply the actual area of the subcutaneous fat, visceral fat, and muscle by the scanning layer thickness to obtain the actual volume of the subcutaneous fat, visceral fat, and muscle.
The computer-readable storage medium of claim 18, wherein the actual area of the subcutaneous fat area, visceral fat area, and muscle area is multiplied by the scanning layer thickness to obtain the subcutaneous fat, visceral fat, and muscle area. After the actual volume, the method further includes the following steps:

The real labels corresponding to the predicted segmentation label and the gold standard image are respectively input into the discriminant network model to obtain the discriminant scores of the predicted segmentation result image and the gold standard image respectively, and the discriminant scores are used to determine the Predict the gap between the segmentation result image and the gold standard image, and adjust the parameters of the generation network model based on the gap to optimize the generation network model.
The computer-readable storage medium according to claim 16, wherein the abdominal CT image data in DICOM format and the abdominal image in JPG format are stored in a blockchain, and the network model is constructed based on the Vnet network model, It includes the following steps:

Setting the convolution kernel in the encoding stage of the Vnet network model as a two-dimensional convolution kernel;

Replacing the deconvolution in the decoding stage of the Vnet network model with bilinear interpolation to obtain a modified Vnet network model;

The channel attention CA module is connected to the modified Vnet network model to obtain the generative network model, where the CA module is used to obtain the advanced level generated during the encoding and decoding stages of the modified Vnet network. The semantic information of the feature map, and selecting pixel information belonging to the high-level feature map from the low-level feature map according to the semantic information;

Wherein, the high-level feature map and the low-level feature map are determined according to the sequence of obtaining feature maps in the encoding stage and the decoding stage. Among the adjacent encoding layers in the encoding stage, the feature map obtained by the next encoding layer is higher than that of the previous encoding. The feature map obtained by the layer is higher-level; in the adjacent coding layers of the decoding stage, the feature map obtained by the previous decryption layer is lower than the feature map obtained by the next decryption layer.