A liver and liver tumor data segmentation method and system

Abstract

The present invention discloses a method and system for segmenting liver and liver tumor data. Based on the Unet network model, a long-distance attention mechanism and a multi-feature fusion module are proposed, and a cross-layer attention mechanism is designed using the hierarchical structure of the network; the cross-layer attention mechanism and the multi-feature fusion module are embedded in the Unet network, and the ResCLA-MNet segmentation network model is constructed in combination with the residual structure; the LiTS liver and liver tumor public CT dataset is used for training, verification, and testing the segmentation performance of the network model; the 3DircaDb-01 liver and liver tumor CT dataset is used to test the generalization performance of the ResCLA-MNet network model to ensure that the network has a certain data migration applicability; the hospital abdominal liver and liver tumor CT dataset is collected, and the trained network model is used to practice on the dataset to verify the application effect of the network. Based on Unet, the present invention draws on the Unet network architecture, combines the advantages of the attention mechanism and multi-feature fusion, and improves the accuracy of the network model in segmenting the liver and liver tumor.

Description

Technical Field

[0001] The present invention belongs to the technical field of medical image processing, and in particular relates to a method and system for segmenting liver and liver tumor data. Background Art

[0002] In the past, segmentation of the liver and liver tumors relied on annotation by professional doctors, which required a certain level of experience and often led to misdiagnosis and missed diagnoses. With the help of computer-assisted technology, it is of great significance to use computers to automatically and accurately segment the target area.

[0003] First, manual labeling is time-consuming and labor-intensive. A patient's liver CT image may contain hundreds of slices. Labeling each slice consumes a lot of the doctor's energy and time. Computer-assisted segmentation can greatly improve the efficiency of segmentation.

[0004] Second, manual segmentation is somewhat subjective and highly dependent on the professional level of the doctor. The segmentation results of doctors of different levels are not the same. The use of computer-assisted automatic segmentation methods can avoid this problem.

[0005] Third, this automated and efficient segmentation can provide doctors with quantitative information such as the lesion's condition for subsequent diagnosis and treatment, providing important quantitative reference and decision-making basis for doctors' subsequent surgery and treatment plan formulation.

[0006] The automatic segmentation method of liver and liver tumor CT images based on attention mechanism and multi-feature fusion can efficiently identify and segment the liver and liver tumors in abdominal CT images, and has high practical application value. Summary of the Invention

[0007] The technical problem to be solved by the present invention is to address the deficiencies in the above-mentioned existing technologies and provide a method and system for segmenting liver and liver tumor data. The method utilizes the Unet benchmark network model, combines the residual structure, embeds a cross-layer attention mechanism and a multi-feature fusion module to improve the accuracy of the network model in segmenting the liver and liver tumors, and is used to solve the technical problem that liver and liver tumor data cannot be accurately segmented.

[0008] The present invention adopts the following technical solutions:

[0009] A method for segmenting liver and liver tumor data, comprising the following steps:

[0010] S1. Preprocess the LiTS and 3DircaDb-01 liver and liver tumor public CT datasets, as well as the liver and liver tumor CT dataset to be segmented, and divide the preprocessed LiTS liver and liver tumor dataset into training set, validation set, and test set;

[0011] S2. Build a ResCLA-MNet segmentation network model with a cross-layer attention mechanism and a multi-feature fusion module. Use the LiTS liver and liver tumor training set, validation set, and test set obtained in step S1 to train, validate, test, and perform ablation experiments on the ResCLA-MNet segmentation network model.

[0012] S3. Use the ResCLA-MNet segmentation network model trained in step S2 to perform a generalization test on the 3DircaDb-01 liver and liver tumor dataset obtained in step S1, and obtain the evaluation index results of the generalization test and the segmentation contour display diagram;

[0013] S4. Input the liver and liver tumor CT data set to be segmented after preprocessing in step S1 into the ResCLA-MNet segmentation network model trained and tested in steps S2 and S3, obtain the model segmentation evaluation index results and segmentation contour display diagram, and realize the segmentation of liver and liver tumor CT data.

[0014] Specifically, in step S1, windowing and histogram equalization are performed on the LiTS and 3DircaDb-01 public CT datasets of liver and liver tumors and the CT dataset of liver and liver tumors to be segmented, respectively. Then, flipping, rotation, and noise addition are performed to amplify the dataset after histogram equalization. Finally, the LiTS liver and liver tumor dataset is divided into training, validation, and test sets of liver and liver tumors. The 3DircaDb-01 liver and liver tumor dataset is obtained for generalization testing. The preprocessed liver and liver tumor dataset to be segmented is used for verification.

[0015] Furthermore, the windowed grayscale value F(i) is expressed as follows:

[0016]

[0017] Among them, f(i) represents the original grayscale value, W w Indicates the window width, W c Indicates the center of the window.

[0018] Specifically, step S2 is as follows:

[0019] S201. Construct a cross-layer attention mechanism and a multi-feature fusion module, and embed them together with the residual structure into the Unet network to obtain the ResCLA-MNet network model.

[0020] S202. Remove modules from the ResCLA-MNet network model obtained in step S201 to obtain ResUnet, Res-Mnet, ResCLA-Net, and ResCLA-MNet segmentation networks.

[0021] S203. Use the LiTS liver and liver tumor training set and validation set obtained in step S1 to train and validate the ResUnet, Res-Mnet, ResCLA-Net, and ResCLA-MNet segmentation networks obtained in step S202, respectively. Use the LiTS liver and liver tumor test set obtained in step S1 to test the trained ResUnet, Res-Mnet, ResCLA-Net, and ResCLA-MNet segmentation networks, respectively. Use the Dice coefficient, Jaccard coefficient, and Precision indicators to evaluate the test results, and obtain the network-segmented liver and liver tumor contour renderings.

[0022] Furthermore, step S201 is specifically as follows:

[0023] S2011. For the cross-layer attention mechanism module, first build a long-distance attention mechanism, use strip pooling to compress and splice the horizontal and vertical features of the CT image, extract features through convolution operation, split the spliced ​​information, and then restore the number of information channels through convolution. Use Sigmoid operation to obtain attention weights, and use matrix multiplication to obtain attention weights of the same size as the original image. The weights are assigned to the original image to complete the assignment of long-distance attention weights;

[0024] S2012. After completing the assignment of long-distance attention weights, add a squeeze and incentive attention mechanism and improve the mechanism so that the output attention weight size can be transferred across layers. The long-distance attention mechanism and the squeeze and incentive attention mechanism are collectively called the cross-layer attention mechanism module;

[0025] S2013. For the multi-feature fusion module, the long-distance attention mechanism (LDA) is added to the skip connection of Unet, and the depth-wise separable convolution is used to pass the information in the upper-layer skip connection to the lower-layer skip connection. This part of the module is called the multi-feature fusion module.

[0026] S2014. Based on the Unet segmentation network model, a cross-layer attention mechanism module is added to the downsampling process of Unet, a multi-feature fusion module is added to the skip link, and a residual structure is added to the downsampling convolution process to construct the ResCLA-MNet segmentation network model.

[0027] Furthermore, in step S202, the optimizer of the ResUnet, Res-MNet, ResCLA-Net, and ResCLA-Mnet segmentation networks is set to the Adam optimizer, the activation function is the ReLU activation function and the Sigmoid activation function, the initial learning rate is 0.0001, the learning decay rate is set to 0.9, the number of iterations is 100, the input image size is 512×512, the batch size is set to 4, and the loss function is the cross entropy loss function:

[0028]

[0029] in, Indicates the probability that the predicted pixel y belongs to the target class, where y is the sample label;

[0030] The Dice loss function is:

[0031]

[0032] Among them, X represents the true sample label and Y represents the segmentation image predicted by the model.

[0033] Furthermore, in step S203, the Dice coefficient index Dice(X,Y), the Jaccard similarity coefficient index Jaccard(X,Y) and the accuracy rate Precesion(X,Y) are calculated as follows:

[0034]

[0035]

[0036]

[0037] Among them, TP represents the number of samples correctly classified in the positive samples; FN represents the number of samples incorrectly classified in the positive samples; FP represents the number of samples incorrectly classified in the negative samples.

[0038] Specifically, in step S3, the Dice coefficient, Jaccard similarity coefficient, and accuracy are used as evaluation indicators. The parameters of the ResUnet, Res-Mnet, ResCLA-Net, and ResCLA-MNet segmentation networks obtained in step S2 are saved and tested using the 3DircaDb-01 liver and liver tumor datasets to obtain evaluation indicators and segmentation contour display diagrams.

[0039] Specifically, step S4 is as follows:

[0040] S401, performing histogram equalization and noise processing on the segmented liver and liver tumor CT data to obtain an unnoised liver and liver tumor CT image and a noisy liver and liver tumor CT image;

[0041] S402, using the ResCLA-MNet segmentation network model trained in step S2 to perform liver and liver tumor segmentation on the original data without noise, the average Dice value for liver segmentation is 0.814, and the average Dice value for liver tumor segmentation is 0.662;

[0042] S403. Use the ResCLA-MNet segmentation network model trained in step S2 to perform liver and liver tumor segmentation on the noisy data. The average Dice value for liver segmentation is 0.774, and the average Dice value for liver tumor segmentation is 0.608.

[0043] In a second aspect, an embodiment of the present invention provides a liver and liver tumor data segmentation system, comprising:

[0044] The data module preprocesses the LiTS and 3DircaDb-01 liver and liver tumor public CT datasets, as well as the liver and liver tumor CT dataset to be segmented, and divides the preprocessed LiTS liver and liver tumor dataset into training, validation, and test sets.

[0045] The network module builds a ResCLA-MNet segmentation network model with a cross-layer attention mechanism and a multi-feature fusion module. The LiTS liver and liver tumor training, validation, and test sets obtained from the data module are used to train, validate, test, and perform ablation experiments on the ResCLA-MNet segmentation network model.

[0046] The test module uses the ResCLA-MNet segmentation network model trained in the network module to perform a generalization test on the 3DircaDb-01 liver and liver tumor dataset obtained in the data module, obtaining the evaluation index results of the generalization test and the segmentation contour display diagram;

[0047] The segmentation module inputs the liver and liver tumor CT data set to be segmented after preprocessing by the data module into the ResCLA-MNet segmentation network model trained and tested by the network module and the test module, obtains the evaluation index results of the model segmentation and the segmentation contour display diagram, and realizes the segmentation of liver and liver tumor CT data.

[0048] Compared with the prior art, the present invention has at least the following beneficial effects:

[0049] The liver and liver tumor data segmentation method uses the Unet network as the base network to segment the target organs; uses a cross-layer attention mechanism to assign different weights to image data so that the network can pay more attention to the target organs; and fuses multiple features of image data to combine image features at different levels of the liver and liver tumors, fully utilizing them and improving the accuracy of liver and liver tumor segmentation.

[0050] Furthermore, the initial data is subjected to pre-processing operations such as windowing, histogram equalization, and data enhancement to ensure the discrimination and contrast of the CT image and expand the data. The HU value range of the original CT image is large, and the displayed image contrast is very poor. The different organs therein cannot be effectively observed by the naked eye. At the same time, the large HU value range is not conducive to subsequent segmentation experiments. In order to avoid the above situation, the present invention uses an image pre-processing method, using windowing operations. According to the characteristics of the liver organ, the HU value in the image is adjusted by adjusting the two parameters of the bit window and bit width, so that each organ in the CT image has a certain degree of discrimination; using histogram equalization, the histogram of the image is converted into an approximate distribution to enhance the image contrast and accelerate the convergence speed of network training; and using image enhancement methods to expand the liver and liver tumor data, increasing the data volume.

[0051] Furthermore, the use of the Unet network with a residual structure as a baseline segmentation network ensures the lower limit of the network segmentation results, which plays an important role in subsequent network design. The Unet network is widely used in fields such as medical image segmentation and target detection. It consists of three parts: an encoder, a decoder, and a skip connection. The encoder structure includes convolution and pooling operations. It extracts image feature information through 3×3 convolution and uses maximum pooling to increase the receptive field, allowing the network to learn more semantic information. The decoding part restores the underlying features to the original image size through upsampling and convolution operations. At the same time, it restores low-resolution images containing high-level semantic features to high-resolution images while retaining the semantic features. The skip connection connects the encoding and decoding parts, realizing the fusion of shallow and deep features, allowing the network to more fully utilize contextual information. The residual structure can prevent the problem of gradient explosion or gradient vanishing during backpropagation, while accelerating the training speed of the network. Compared with traditional manual segmentation methods, the efficiency and segmentation accuracy of the Unet network are significantly improved.

[0052] Furthermore, adding the attention mechanism to the segmentation network can make the segmentation network pay more attention to the target areas of the liver and liver tumors, and effectively improve the segmentation effect of the liver and liver tumors while obtaining features that are more relevant to the target. In traditional Unet, the deep features of the liver and liver tumors in CT image data are extracted through the encoding structure, but the irregularity of the target organ boundaries and the size and shape affect the extraction of features. Simple convolution and downsampling will lose the detailed features of the target, thereby affecting the overall segmentation effect. The attention mechanism gives corresponding weights to the target to be segmented and the background, so that the network pays more attention to the spatial information and channel information of the target to be segmented during segmentation. The liver and liver tumor areas with larger weights will be more easily perceived by the segmentation network and accurately segmented. In this way, even when the target boundaries are irregular and the size and shape are complex, the network can achieve good segmentation effects.

[0053] Furthermore, during the encoding phase, information from the upper layer is passed to the lower layer, making feature fusion more uniform and efficient. Each layer of the Unet network only receives information from the downsampled convolutional layers, without exchanging any other information. This approach is relatively simple. During the encoding process, the feature information extracted by the attention mechanism is passed across layers to the next layer. This not only strengthens information exchange between layers but also facilitates the subsequent feature fusion process of jump connections, making the information in feature fusion more balanced and efficient, thereby achieving higher segmentation accuracy.

[0054] Furthermore, constructing a multi-feature fusion module and integrating it into the Unet's skip connection process can alleviate the semantic gap and feature information imbalance, thereby improving the accuracy of liver and liver tumor segmentation. The original skip connection structure is simple, simply concatenating feature information from the encoding stage and the decoding stage, which can lead to a certain semantic gap problem. Through the multi-feature fusion module, the skip connections of adjacent layers are connected. Through depthwise separable convolution, the spatial attention feature information obtained by the upper-layer skip connection is fused with the spatial attention feature information obtained by the lower-layer skip connection. This adds more target-related information to the feature fusion stage, bridging the semantic gap. Subsequently, channel attention information is obtained and the obtained feature information is organized through channel attention and 1×1 convolution, finally achieving feature fusion in the skip connection. Through the multi-feature fusion module of the upper and lower layers, the information of features from different layers is combined, allowing the network to obtain richer feature information related to the target to be segmented, thereby improving the accuracy of liver and liver tumor segmentation.

[0055] It can be understood that the beneficial effects of the second aspect mentioned above can be found in the relevant description of the first aspect mentioned above, and will not be repeated here.

[0056] In summary, the present invention uses the Unet segmentation network model as the benchmark framework, adds a residual structure optimization encoding process to the Unet, and prevents the gradient explosion; utilizes the attention mechanism to make the network pay more attention to the spatial and channel information of the target area, and improves the network's boundary segmentation accuracy for targets of different sizes and shapes; incorporates the cross-layer idea into the segmentation network model, strengthens the information exchange between network layers, and makes feature fusion more balanced and efficient; adds a multi-feature fusion module to make up for the semantic gap problem of jump connections.

[0057] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 It is the overall flow chart of the present invention;

[0059] Figure 2 Schematic diagram of the process of establishing a data set in the present invention;

[0060] Figure 3 Preprocessing diagram for the dataset;

[0061] Figure 4 Construct a schematic for the long-range attention mechanism;

[0062] Figure 5 Improved schematics for cross-layer attention mechanisms and squeeze and incentive attention mechanisms;

[0063] Figure 6 Schematic diagram of multi-feature fusion module;

[0064] Figure 7 is a schematic diagram of the residual structure;

[0065] Figure 8 Schematic diagram of the ResCLA-MNet network model structure;

[0066] Figure 9 is the loss function curve of the liver training process;

[0067] Figure 10 This is the loss function curve of the liver tumor training process;

[0068] Figure 11 This is a graph showing the segmentation contours of the liver and liver tumor on the LiTS dataset;

[0069] Figure 12 This is the experimental flow chart on the 3DircaDb-01 dataset;

[0070] Figure 13 This is a graph showing the segmentation contours of the liver and liver tumor on the 3DircaDb-01 dataset;

[0071] Figure 14 This is a display of original CT data of patients from the Second Affiliated Hospital of Xi'an Jiaotong University;

[0072] Figure 15 This is a diagram showing hospital image preprocessing data;

[0073] Figure 16 This is a diagram showing the segmentation results of the original liver data;

[0074] Figure 17 This is a diagram showing the segmentation results of the original liver tumor data;

[0075] Figure 18 This is a graph showing the liver data segmentation results after adding noise;

[0076] Figure 19 This is a diagram showing the segmentation results of liver tumor data after adding noise. DETAILED DESCRIPTION

[0077] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0078] In the description of the present invention, it is to be understood that the terms “include” and “comprise” indicate the presence of the described features, wholes, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or collections thereof.

[0079] It should also be understood that the terms used in the present specification are only for the purpose of describing particular embodiments and are not intended to limit the present invention. As used in the present specification and the appended claims, the singular forms "a", "an", and "the" are intended to include the plural forms unless the context clearly indicates otherwise.

[0080] It should be further understood that the term "and / or" as used in the present specification and the appended claims refers to and includes any and all possible combinations of one or more of the associated listed items. For example, A and / or B may represent: A exists alone, A and B exist simultaneously, and B exists alone. In addition, the character " / " in the present invention generally indicates that the associated objects are in an "or" relationship.

[0081] It should be understood that although the terms "first," "second," and "third" may be used to describe preset ranges in embodiments of the present invention, these preset ranges should not be limited to these terms. These terms are merely used to distinguish one preset range from another. For example, without departing from the scope of embodiments of the present invention, the first preset range may also be referred to as the second preset range, and similarly, the second preset range may also be referred to as the first preset range.

[0082] The word "if," as used herein, may be interpreted as "at the time of" or "when" or "in response to determining" or "in response to detecting," depending on the context. Similarly, the phrases "if it is determined" or "if (stated condition or event) is detected" may be interpreted as "when it is determined" or "in response to the determination" or "when detecting (stated condition or event)" or "in response to detecting (stated condition or event)," depending on the context.

[0083] The accompanying drawings illustrate various schematic diagrams of structures according to embodiments disclosed herein. These figures are not drawn to scale; for clarity, some details are exaggerated and some details may be omitted. The shapes of the various regions and layers shown in the figures, as well as their relative sizes and positional relationships, are merely exemplary and may deviate in practice due to manufacturing tolerances or technical limitations. Those skilled in the art may design regions / layers with different shapes, sizes, and relative positions as needed.

[0084] The present invention provides a method for liver and liver tumor data segmentation. Based on the Unet network model, a long-range attention mechanism and a multi-feature fusion module are proposed, and a cross-layer attention mechanism is designed using the network's hierarchical structure. The cross-layer attention mechanism and the multi-feature fusion module are embedded in the Unet network model and combined with the residual structure to construct the ResCLA-MNet segmentation network model. The LiTS liver and liver tumor public CT dataset is used for training, verification, and testing the segmentation performance of the network model. The 3DircaDb-01 liver and liver tumor public CT dataset is used to test the generalization performance of the ResCLA-MNet network model to ensure that the network model has certain data migration applicability. A hospital abdominal liver and liver tumor CT dataset is collected and the trained network model is used to practice on the hospital dataset to verify the application effect of the network. Based on the Unet network model, the present invention draws on the Unet network architecture and combines the advantages of the attention mechanism and multi-feature fusion to improve the accuracy of the network model in liver and liver tumor segmentation.

[0085] See also Figure 1The present invention provides a method for segmenting liver and liver tumor data. The method is based on a Unet network model with a residual structure and embeds an attention mechanism into the network, so that the network can pay more attention to the spatial detail information and channel information of the target to be segmented, making the liver and liver tumor segmentation more accurate; cross-layer connection is realized in the encoding stage, which strengthens the information interaction between network layers and makes feature fusion more balanced and efficient; a multi-feature fusion module is added to the skip connection, so that information can be transmitted and fused across layers in the skip connection, alleviating the semantic gap problem and improving the accuracy of liver and liver tumor segmentation; the method comprises the following steps:

[0086] S1. Preprocess the LiTS and 3DircaDb-01 liver and liver tumor public CT datasets, as well as the hospital's liver and liver tumor CT dataset to be segmented, and divide the preprocessed LiTS liver and liver tumor dataset into training set, validation set, and test set;

[0087] S101. Dataset acquisition

[0088] This paper obtained two public datasets, LiTS and 3DircaDb-01, for the construction of a liver tumor CT image dataset. The LiTS dataset contains CT images of the liver and liver tumors of 131 patients. The image plane resolution ranges from 0.5mm to 1.0mm, the z-axis resolution ranges from 0.45mm to 6.0mm, the number of slices on the z-axis ranges from 42 to 1026, and the number of tumors per patient ranges from 0 to 75. The 3DircaDb-01 dataset contains a total of 20 patients, of which 15 patients have liver tumors, with the number of tumors ranging from 1 to 46. The image resolution of both datasets is 512×512.

[0089] S102. Data preprocessing and division

[0090] Based on the characteristics of the liver in abdominal CT images, windowing, histogram equalization, and image enhancement were used as data preprocessing methods. The resulting LiTS liver training and validation set consisted of 17,583 images, a liver test set of 2,567 images, and a liver tumor training and validation set of 7,935 images, and a liver tumor test set of 1,156 images. Furthermore, the 3DircaDb-01 liver dataset, with 2,074 images, and a liver tumor dataset, with 523 images, was also obtained.

[0091] See also Figure 2 , the implementation process is as follows:

[0092] S1021. Extract CT image slices containing the liver and liver tumor, and their corresponding doctor-annotated results, from the patient's data file based on the doctor's annotation results. The number of slices is the same as described in step S102.

[0093] S1022. Windowing technology is used to process the CT image slices. The calculation process of the windowing technology is as follows:

[0094]

[0095] Among them, F(i) represents the grayscale value after windowing, f(i) represents the original grayscale value, and W w Indicates the window width, W c Indicates the window center, the window width and window center are set to 250 and 45 respectively.

[0096] S1023, applying histogram equalization to the windowed image, setting the contrast limit to 2.0, and setting the block size to 8×8;

[0097] S1024, enhance the image after histogram equalization, including random rotation, flipping and noise addition. The preprocessing display is as follows Figure 3 shown.

[0098] The noise added is random Gaussian noise with a mean in the interval [0, 1] and a variance in the interval [0, 3]. The probability distribution of the added noise is as follows:

[0099]

[0100] Here, μ represents the mean and σ represents the variance.

[0101] S2. Build a ResCLA-MNet segmentation network model with a cross-layer attention mechanism and a multi-feature fusion module. Use the LiTS liver and liver tumor training set, validation set, and test set obtained in step S1 to train, validate, test, and perform ablation experiments on the ResCLA-MNet segmentation network model.

[0102] See also Figures 4 to 11 , the specific steps are as follows:

[0103] S201. Construct a cross-layer attention mechanism and a multi-feature fusion module, and combine them with the residual structure and embed them into the Unet network model to obtain the ResCLA-MNet segmentation network model.

[0104] The module construction process is as follows:

[0105] S2011. For the cross-layer attention mechanism module, we first build a long-distance attention mechanism. We use strip pooling to compress and splice the horizontal and vertical features of the CT image. After extracting features through convolution, we split the spliced ​​information and restore the number of information channels through convolution. We use Sigmoid operation to obtain attention weights, and use matrix multiplication to obtain attention weights of the same size as the original image. We assign the weights to the original image to complete the assignment of long-distance attention weights. This part is as follows: Figure 4 As shown;

[0106] S2012. After completing the assignment of long-distance attention weights, a squeeze and incentive attention mechanism is added and improved so that the output attention weight size can be transferred across layers. The long-distance attention mechanism and the squeeze and incentive attention mechanism are collectively referred to as the cross-layer attention mechanism. The improvement method of the squeeze and incentive attention mechanism is similar to the cross-layer attention mechanism (CLA). Figure 5 As shown;

[0107] S2013, for the multi-feature fusion module, the long-distance attention mechanism (LDA) is added to the jump connection of Unet, and the depth-separable convolution is used to pass the information in the upper jump connection to the jump connection of the lower layer. This part of the module is called the multi-feature fusion module. The implementation method is as follows Figure 6 As shown;

[0108] S2014, based on the Unet segmentation network model, added a cross-layer attention mechanism to the Unet downsampling process, added a multi-feature fusion module to the skip link, and added a residual structure to the downsampling convolution process to construct the ResCLA-MNet segmentation network model. The residual structure is as follows Figure 7 As shown, the overall structure of the network is as follows Figure 8 shown.

[0109] S202. Remove modules from the ResCLA-MNet network model obtained in step S201 to obtain ResUnet, Res-Mnet, and ResCLA-Net segmentation networks; use the LiTS liver and liver tumor training and validation sets obtained in step S1 to train and validate the ResUnet, Res-Mnet, ResCLA-Net, and ResCLA-MNet segmentation networks obtained in step S202, respectively;

[0110] The parameters are set as follows: the optimizer is Adam optimizer, the activation function is ReLU activation function and Sigmoid activation function, the initial learning rate is 0.0001, the learning decay rate is set to 0.9, the number of iterations is 100, the input image size is 512×512, the batch size is set to 4, and the loss function is cross entropy loss function and Dice loss function.

[0111] The calculation formula of the ReLU activation function is as follows:

[0112] ReLU(x)=max(x,0)

[0113] Where x represents a pixel in the image. The ReLU activation function resets all values ​​less than 0 to zero to achieve unilateral suppression.

[0114] The calculation formula of the Sigmoid activation function is as follows:

[0115]

[0116] Where x represents a single pixel in the image. The Sigmoid activation function maps pixels to a value between (0, 1) to obtain the probability of pixel classification. A larger value indicates a greater likelihood that the pixel belongs to the target region.

[0117] The calculation formula of the cross entropy loss function is as follows:

[0118]

[0119] in, Represents the probability that the predicted pixel y belongs to the target class, y is the sample label, when , it means that the predicted pixel is the same as the label, and the value of the cross entropy loss function is 0.

[0120] The calculation formula of Dice loss function is as follows:

[0121]

[0122] Where X represents the true sample label and Y represents the segmented image predicted by the model. When the predicted image is the same as the sample label, the value of the Dice loss function is 0.

[0123] The liver training loss function curve of the ResCLA-MNet segmentation network model is as follows Figure 9 As shown, the liver tumor training loss function curve is as follows Figure 10 shown.

[0124] S203. Use the LiTS liver and liver tumor test set obtained in step S1 to test the trained ResUnet, Res-Mnet, ResCLA-Net, and ResCLA-MNet segmentation network models respectively; use the three evaluation indicators of Dice coefficient, Jaccard similarity coefficient, and Precision accuracy to evaluate the model test results, and obtain the contour renderings of the liver and liver tumor segmented by the network model.

[0125] The evaluation index calculation formula is as follows:

[0126]

[0127]

[0128]

[0129] Among them, TP represents the number of samples correctly classified in the positive samples; FN represents the number of samples incorrectly classified in the positive samples; FP represents the number of samples incorrectly classified in the negative samples.

[0130] The standard deviation is calculated as follows:

[0131]

[0132] Where n represents the number of samples.

[0133] The final segmentation index results obtained by the ResCLA-MNet segmentation network model on the LiTS liver and liver tumor test sets are shown in Tables 1 and 2.

[0134] Table 1 LiTS liver test set segmentation experiment results (mean ± standard deviation)

[0135]

[0136]

[0137] Table 2 LiTS liver tumor test set segmentation experimental results (mean ± standard deviation)

[0138]

[0139] The segmentation effect of the ResCLA-MNet segmentation network model on the LiTS liver and liver tumor test set is as follows Figure 11 shown.

[0140] S3. Use the ResCLA-MNet segmentation network model trained in step S2 to perform a generalization test on the 3DircaDb-01 liver and liver tumor dataset obtained in step S1, and obtain the evaluation index results of the generalization test and the segmentation contour effect diagram.

[0141] See also Figure 12 , the specific steps are as follows:

[0142] S301. The four segmentation network model parameters obtained in step S2 are saved and tested directly on the 3DircaDb-01 liver and liver tumor datasets without secondary training, and evaluation index results and contour renderings of the liver and liver tumor segmented by the network model are obtained.

[0143] S302: The evaluation index adopts the Dice coefficient, Jaccard similarity coefficient and Precision accuracy described in step S203, and the obtained segmentation index results are shown in Table 3 and Table 4.

[0144] Table 3 3DircaDb-01 liver dataset segmentation experimental results (mean ± standard deviation)

[0145]

[0146] Table 4 3DircaDb-01 liver tumor dataset segmentation experimental results (mean ± standard deviation)

[0147]

[0148] The segmentation effect of the ResCLA-MNet segmentation network model on the 3DircaDb-01 liver and liver tumor dataset is shown in the figure. Figure 13 shown.

[0149] S4. Input the hospital's liver and liver tumor CT data set to be segmented after preprocessing in step S1 into the ResCLA-MNet segmentation network model trained and tested in steps S2 and S3, obtain the model segmentation evaluation index results and segmentation contour map, and realize the segmentation of the hospital's liver and liver tumor CT data.

[0150] The 16 patient CT image data collected from the Second Affiliated Hospital of Xi'an Jiaotong University were preprocessed and denoised to obtain the liver and liver tumor dataset to be segmented. After denoising, a total of 32 CT image data were obtained, such as Figure 14 As shown, the ResCLA-MNet network model trained in step S2 is used to segment it and obtain the segmentation evaluation index and segmentation contour map.

[0151] The specific steps are as follows:

[0152] S401, 16 patient CT image data collected from the Second Affiliated Hospital of Xi'an Jiaotong University (such as Figure 14 ) were histogram equalized and noise was added. After noise addition, a total of 32 CT image data were obtained, of which 16 were liver and liver tumor images without noise, and 16 were liver and liver tumor images with noise. The pre-processed images are shown in Figure 15 As shown;

[0153] S402: Use the ResCLA-MNet segmentation network model trained and tested in steps S2 and S3 to segment the liver and liver tumors on the original data without noise. The average Dice value of liver segmentation is 0.814, and the average Dice value of liver tumor segmentation is 0.662. The liver segmentation contour effect is shown in the figure below. Figure 16As shown, the liver tumor segmentation contour effect is as follows Figure 17 As shown;

[0154] S403. The ResCLA-MNet segmentation network model trained and tested in steps S2 and S3 is used to segment the liver and liver tumor on the noisy data. The average Dice value for liver segmentation is 0.774, and the average Dice value for liver tumor segmentation is 0.608. The liver segmentation contour effect is shown in the figure below. Figure 18 As shown, the liver tumor segmentation contour effect is as follows Figure 19 shown.

[0155] In another embodiment of the present invention, a liver and liver tumor data segmentation system is provided. The system can be used to implement the above-mentioned liver and liver tumor data segmentation method. Specifically, the liver and liver tumor data segmentation system includes a data module, a network module, a test module, and a segmentation module.

[0156] The data module preprocesses the LiTS and 3DircaDb-01 liver and liver tumor public CT datasets, as well as the liver and liver tumor CT dataset to be segmented, and divides the preprocessed LiTS liver and liver tumor dataset into training, validation, and test sets.

[0157] The network module builds a ResCLA-MNet segmentation network model with a cross-layer attention mechanism and a multi-feature fusion module. The LiTS liver and liver tumor training, validation, and test sets obtained from the data module are used to train, validate, test, and perform ablation experiments on the ResCLA-MNet segmentation network model.

[0158] The test module uses the ResCLA-MNet segmentation network model trained in the network module to perform a generalization test on the 3DircaDb-01 liver and liver tumor dataset obtained in the data module, obtaining the evaluation index results of the generalization test and the segmentation contour display diagram;

[0159] The segmentation module inputs the liver and liver tumor CT data set to be segmented after preprocessing by the data module into the ResCLA-MNet segmentation network model trained and tested by the network module and the test module, obtains the evaluation index results of the model segmentation and the segmentation contour display diagram, and realizes the segmentation of liver and liver tumor CT data.

[0160] In another embodiment of the present invention, a terminal device is provided, which includes a processor and a memory, wherein the memory is used to store a computer program, the computer program includes program instructions, and the processor is used to execute the program instructions stored in the computer storage medium. The processor may be a central processing unit (CPU), or may be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, which is suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions to implement the corresponding method flow or corresponding function; the processor described in the embodiment of the present invention can be used for the operation of the liver and liver tumor data segmentation method, including:

[0161] The LiTS and 3DircaDb-01 public CT datasets of liver and liver tumors, as well as the CT dataset of liver and liver tumors to be segmented, were preprocessed respectively, and the preprocessed LiTS liver and liver tumor dataset was divided into training set, validation set and test set; a ResCLA-MNet segmentation network model with a cross-layer attention mechanism and a multi-feature fusion module was constructed, and the ResCLA-MNet segmentation network model was trained, validated, tested and ablated using the LiTS liver and liver tumor training set, validation set and test set; the trained ResCLA-MNet segmentation network model was used to perform a generalization test on the 3DircaDb-01 liver and liver tumor dataset, and the evaluation index results of the generalization test and the segmentation contour display diagram were obtained; the preprocessed CT dataset of liver and liver tumors to be segmented was input into the trained and tested ResCLA-MNet segmentation network model, and the evaluation index results of the model segmentation and the segmentation contour display diagram were obtained, thereby realizing the segmentation of liver and liver tumor CT data.

[0162] In another embodiment of the present invention, the present invention further provides a storage medium, specifically a computer-readable storage medium (Memory), which is a memory device in a terminal device for storing programs and data. It can be understood that the computer-readable storage medium here can include both the built-in storage medium in the terminal device and, of course, the extended storage medium supported by the terminal device. The computer-readable storage medium provides a storage space, which stores the operating system of the terminal. In addition, one or more instructions suitable for being loaded and executed by the processor are also stored in the storage space. These instructions can be one or more computer programs (including program codes). It should be noted that the computer-readable storage medium here can be a high-speed RAM memory or a non-volatile memory (Non-Volatile Memory), such as at least one disk memory.

[0163] The processor may load and execute one or more instructions stored in a computer-readable storage medium to implement the corresponding steps of the method for segmenting liver and liver tumor data in the above embodiment. The processor may load and execute the following steps:

[0164] The LiTS and 3DircaDb-01 public CT datasets of liver and liver tumors, as well as the CT dataset of liver and liver tumors to be segmented, were preprocessed respectively, and the preprocessed LiTS liver and liver tumor dataset was divided into training set, validation set and test set; a ResCLA-MNet segmentation network model with a cross-layer attention mechanism and a multi-feature fusion module was constructed, and the ResCLA-MNet segmentation network model was trained, validated, tested and ablated using the LiTS liver and liver tumor training set, validation set and test set; the trained ResCLA-MNet segmentation network model was used to perform a generalization test on the 3DircaDb-01 liver and liver tumor dataset, and the evaluation index results of the generalization test and the segmentation contour display diagram were obtained; the preprocessed CT dataset of liver and liver tumors to be segmented was input into the trained and tested ResCLA-MNet segmentation network model, and the evaluation index results of the model segmentation and the segmentation contour display diagram were obtained, thereby realizing the segmentation of liver and liver tumor CT data.

[0165] In summary, the present invention provides a method and system for liver and liver tumor data segmentation. First, the liver and liver tumor CT data are preprocessed; secondly, the cross-layer attention mechanism and multi-feature fusion module are used in combination with the residual structure to construct the ResCLA-MNet segmentation network model. Combining the advantages of the attention mechanism and the multi-feature fusion module, the network model's accuracy in liver and liver tumor segmentation is improved; finally, the network model is applied in real-world scenarios to verify the accuracy and feasibility of the present invention.

[0166] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0167] In the above embodiments, the description of each embodiment has its own focus. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant description of other embodiments.

[0168] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed in the present invention can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present invention.

[0169] In the embodiments provided by the present invention, it should be understood that the disclosed devices / terminals and methods can be implemented in other ways. For example, the device / terminal embodiments described above are merely illustrative. For example, the division of the modules or units is merely a logical functional division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be through some interface, indirect coupling or communication connection of devices or units, and can be electrical, mechanical, or other forms.

[0170] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0171] In addition, the functional units in the various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0172] If the integrated module / unit 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. Based on this understanding, the present invention implements all or part of the process in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by a processor, it can implement the steps of the above-mentioned various method embodiments. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium, etc. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electric carrier signals and telecommunication signals.

[0173] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the steps in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0174] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0175] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0176] The above content is only for explaining the technical idea of ​​the present invention and cannot be used to limit the protection scope of the present invention. Any changes made on the basis of the technical solution in accordance with the technical idea proposed by the present invention shall fall within the protection scope of the claims of the present invention.

Claims

1. A method for segmenting liver and liver tumor data, characterized in that: The following steps are involved: S1. Preprocess the LiTS and 3DircaDb-01 liver and liver tumor public CT datasets, as well as the liver and liver tumor CT dataset to be segmented, and divide the preprocessed LiTS liver and liver tumor dataset into training set, validation set, and test set; S2. Build a ResCLA-MNet segmentation network model with a cross-layer attention mechanism and a multi-feature fusion module. Use the LiTS liver and liver tumor training set, validation set, and test set obtained in step S1 to train, validate, test, and perform ablation experiments on the ResCLA-MNet segmentation network model. Specifically: S201. Construct a cross-layer attention mechanism and a multi-feature fusion module, and embed them together with the residual structure into the Unet network to obtain the ResCLA-MNet network model, specifically: S2011. For the cross-layer attention mechanism module, first build a long-distance attention mechanism, use strip pooling to compress and splice the horizontal and vertical features of the CT image, extract features through convolution operation, split the spliced ​​information, and then restore the number of information channels through convolution. Use Sigmoid operation to obtain attention weights, and use matrix multiplication to obtain attention weights of the same size as the original image. The weights are assigned to the original image to complete the assignment of long-distance attention weights; S2012. After completing the assignment of long-distance attention weights, add a squeeze and incentive attention mechanism and improve the mechanism so that the output attention weight size can be transferred across layers. The long-distance attention mechanism and the squeeze and incentive attention mechanism are collectively called the cross-layer attention mechanism module; S2013. For the multi-feature fusion module, the long-distance attention mechanism (LDA) is added to the skip connection of Unet, and the depth-wise separable convolution is used to pass the information in the upper-layer skip connection to the lower-layer skip connection. This part of the module is called the multi-feature fusion module. S2014, based on the Unet segmentation network model, added a cross-layer attention mechanism module to the Unet downsampling process, added a multi-feature fusion module to the skip link, and added a residual structure to the downsampling convolution process to construct the ResCLA-MNet segmentation network model; S202. Remove modules from the ResCLA-MNet network model obtained in step S201 to obtain ResUnet, Res-Mnet, ResCLA-Net, and ResCLA-MNet segmentation networks. S203. Using the LiTS liver and liver tumor training and validation sets obtained in step S1, the ResUnet, Res-Mnet, ResCLA-Net, and ResCLA-MNet segmentation networks obtained in step S202 are trained and validated, respectively. Furthermore, using the LiTS liver and liver tumor test sets obtained in step S1, the trained ResUnet, Res-Mnet, ResCLA-Net, and ResCLA-MNet segmentation networks are tested, respectively. The test results are evaluated using the Dice coefficient, Jaccard coefficient, and Precision metrics, and contour images of the liver and liver tumor segmented by the network are obtained. S3. Use the ResCLA-MNet segmentation network model trained in step S2 to perform a generalization test on the 3DircaDb-01 liver and liver tumor dataset obtained in step S1, and obtain the evaluation index results of the generalization test and the segmentation contour display diagram; S4. Input the liver and liver tumor CT data set to be segmented after preprocessing in step S1 into the ResCLA-MNet segmentation network model trained and tested in steps S2 and S3, obtain the model segmentation evaluation index results and segmentation contour display diagram, and realize the segmentation of liver and liver tumor CT data.

2. The method for segmenting liver and liver tumor data according to claim 1, wherein: In step S1, windowing and histogram equalization are performed on the LiTS and 3DircaDb-01 public CT datasets of liver and liver tumors, as well as the CT dataset of liver and liver tumors to be segmented. The histogram-equalized datasets are then amplified by flipping, rotating, and adding noise. Finally, the LiTS liver and liver tumor dataset is partitioned into training, validation, and test sets for the liver and liver tumors, yielding the 3DircaDb-01 liver and liver tumor dataset for generalization testing. The preprocessed liver and liver tumor datasets to be segmented are used for testing.

3. The method for segmenting liver and liver tumor data according to claim 2, wherein: Grayscale value after windowing It is expressed as follows: in, represents the original grayscale value, Indicates the window width, Indicates the center of the window.

4. The method for segmenting liver and liver tumor data according to claim 1, wherein: In step S202, the optimizer of the ResUnet, Res-MNet, ResCLA-Net, and ResCLA-MNet segmentation networks is set to the Adam optimizer, the activation function is the ReLU activation function and the Sigmoid activation function, the initial learning rate is 0.0001, the learning decay rate is set to 0.9, the number of iterations is 100, the input image size is 512×512, the batch size is set to 4, and the loss function is the cross entropy loss function: in, Represents predicted pixels The probability of belonging to the target class, is the sample label; The Dice loss function is: in, represents the true sample label, Represents the segmentation image predicted by the model.

5. The method for segmenting liver and liver tumor data according to claim 1, wherein: In step S203, the Dice coefficient index , Jaccard similarity coefficient index and accuracy The calculation is as follows: in, Indicates the number of samples that are correctly classified in the positive samples; Indicates the number of samples that are misclassified in the positive samples; Indicates the number of samples that are misclassified in negative samples.

6. The method for segmenting liver and liver tumor data according to claim 1, wherein: In step S3, the Dice coefficient, Jaccard similarity coefficient, and accuracy are used as evaluation indicators. The parameters of the ResUnet, Res-Mnet, ResCLA-Net, and ResCLA-MNet segmentation networks obtained in step S2 are saved and tested using the 3DircaDb-01 liver and liver tumor datasets to obtain evaluation indicators and segmentation contour display diagrams.

7. The method for segmenting liver and liver tumor data according to claim 1, wherein: Step S4 is specifically as follows: S401, performing histogram equalization and noise processing on the segmented liver and liver tumor CT data to obtain an unnoised liver and liver tumor CT image and a noisy liver and liver tumor CT image; S402, using the ResCLA-MNet segmentation network model trained in step S2 to perform liver and liver tumor segmentation on the original data without noise, the average Dice value for liver segmentation is 0.814, and the average Dice value for liver tumor segmentation is 0.662; S403. Use the ResCLA-MNet segmentation network model trained in step S2 to perform liver and liver tumor segmentation on the noisy data. The average Dice value for liver segmentation is 0.774, and the average Dice value for liver tumor segmentation is 0.

608.

8. A liver and liver tumor data segmentation system, characterized by: include: The data module preprocesses the LiTS and 3DircaDb-01 liver and liver tumor public CT datasets, as well as the liver and liver tumor CT dataset to be segmented, and divides the preprocessed LiTS liver and liver tumor dataset into training, validation, and test sets. In the network module, a ResCLA-MNet segmentation network model with a cross-layer attention mechanism and a multi-feature fusion module is constructed. The LiTS liver and liver tumor training, validation, and test sets obtained from the data module are used to train, validate, test, and perform ablation experiments on the ResCLA-MNet segmentation network model. Specifically: Construct a cross-layer attention mechanism and a multi-feature fusion module, and embed them together with the residual structure into the Unet network to obtain the ResCLA-MNet network model; The modules in the obtained ResCLA-MNet network model are removed to obtain ResUnet, Res-Mnet, ResCLA-Net and ResCLA-MNet segmentation networks, specifically: For the cross-layer attention mechanism module, a long-distance attention mechanism is first constructed. The horizontal and vertical features of the CT image are compressed and spliced ​​using strip pooling. The spliced ​​information is split after extracting features through convolution operation, and the number of information channels is restored through convolution. The attention weight is obtained using Sigmoid operation, and the attention weight with the same size as the original image is obtained by matrix multiplication. The weight is assigned to the original image to complete the assignment of long-distance attention weight. After the assignment of long-distance attention weight is completed, the squeezing and excitation attention mechanism is added, and the mechanism is improved so that the output attention weight size can be transferred across layers, which is called The long-distance attention mechanism and the squeeze and excitation attention mechanism are collectively referred to as the cross-layer attention mechanism module. For the multi-feature fusion module, the long-distance attention mechanism (LDA) is added to the skip connection of Unet, and depthwise separable convolution is used to pass information from the upper-layer skip connection to the lower-layer skip connection. This module is called the multi-feature fusion module. Based on the Unet segmentation network model, the cross-layer attention mechanism module is added to the Unet downsampling process, the multi-feature fusion module is added to the skip connection, and the residual structure is added to the downsampling convolution process to construct the ResCLA-MNet segmentation network model. The obtained LiTS liver and liver tumor training and validation sets were used to train and validate the obtained ResUnet, Res-Mnet, ResCLA-Net, and ResCLA-MNet segmentation networks, respectively. The obtained LiTS liver and liver tumor test sets were used to test the trained ResUnet, Res-Mnet, ResCLA-Net, and ResCLA-MNet segmentation networks, respectively. The Dice coefficient, Jaccard coefficient, and Precision index were used to evaluate the test results, and the network-segmented liver and liver tumor contour renderings were obtained. The test module uses the ResCLA-MNet segmentation network model trained in the network module to perform a generalization test on the 3DircaDb-01 liver and liver tumor dataset obtained in the data module, obtaining the evaluation index results of the generalization test and the segmentation contour display diagram; The segmentation module inputs the liver and liver tumor CT data set to be segmented after preprocessing by the data module into the ResCLA-MNet segmentation network model trained and tested by the network module and the test module, obtains the evaluation index results of the model segmentation and the segmentation contour display diagram, and realizes the segmentation of liver and liver tumor CT data.

Images