A liver CT image-based lesion segmentation device

Abstract

The application belongs to the technical field of CT image processing devices, and particularly relates to a lesion segmentation device based on liver CT images, a preprocessing module pre-processes liver CT images; a segmentation model establishing module establishes a CT image lesion area segmentation model, and then a training module is used to train the CT image lesion area segmentation model; finally, a segmentation module is used to segment the liver and the lesion area of the pre-processed liver CT image by using the trained CT image lesion area segmentation model. The liver CT image is pre-processed to adjust the CT image to a proper size and contrast, 3D convolution is used instead of 2D convolution in the training process of the CT image lesion area segmentation model to retain the connection between slices, [3*3] deep separable convolution is used to extract features, and [1*1] convolution is used to fuse features and adjust the number of channels, thereby effectively improving the accuracy of the segmentation of the lesion area in the CT image and improving the work efficiency of doctors.

Description

Technical Field

[0001] This invention belongs to the technical field of CT image processing devices, and particularly relates to a lesion segmentation device based on liver CT images. Background Technology

[0002] Primary liver cancer (hereinafter referred to as liver cancer) is currently the fourth most common malignant tumor in my country, seriously threatening the lives and health of the Chinese people. Hepatocellular carcinoma is the most common pathological type, frequently occurring in individuals with chronic viral hepatitis and cirrhosis. Patients with these conditions are often categorized as high-risk groups for liver cancer. Early-stage liver cancer patients can benefit from various radical treatments, primarily surgical resection. Therefore, early detection and treatment are crucial for improving the prognosis of liver cancer patients. Currently, medical imaging examinations have become a routine assessment method in the medical management of high-risk liver cancer populations, aiding in the early detection of liver cancer. The core of medical imaging is the image itself; from image generation to usage, every step influences the application of medical imaging in the assessment of high-risk liver cancer populations.

[0003] Medical imaging examinations are the process of image generation. With the advancement of technology, the functions of CT and MR equipment have been greatly improved in liver cancer imaging. For example, there is spectral CT that can identify substances, various functional MR imaging sequences that can quantify tissue function, and the use of liver and gallbladder cell-specific contrast agents. However, this has also led to increased complexity and personalization in operation, which places higher demands on the operators of the scanning equipment—radiologists: the setting of sequence parameters, the selection of scanning phases, the use of special contrast agents, and image post-processing methods all affect image quality.

[0004] Currently, after a patient has a liver CT scan, doctors visually identify lesions on the liver CT image. However, since most patients' liver CT images show mostly healthy areas with only a small portion being lesion areas, and given the large number of patients, visual identification by doctors would require identifying the entire area of ​​the liver CT image, which would consume a lot of time and effort, resulting in very low work efficiency.

[0005] Therefore, how to segment lesion areas based on existing liver CT images is a technical problem that urgently needs to be solved. Summary of the Invention

[0006] The purpose of this invention is to provide a lesion segmentation device based on liver CT images, in order to solve the technical problem that the current process of doctors identifying lesions through liver CT images is relatively inefficient.

[0007] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0008] A lesion segmentation device based on liver CT images includes a preprocessing module, a segmentation model establishment module, a training module, and a segmentation module; the preprocessing module is connected to the segmentation module, the segmentation model establishment module is connected to the training module, and the training module is connected to the segmentation module.

[0009] The preprocessing module is used to preprocess the CT images to be segmented;

[0010] The segmentation model establishment module is used to establish a segmentation model of lesion regions in CT images;

[0011] The training module is used to train the CT image lesion region segmentation model using a large number of CT images with labeled liver lesions.

[0012] The segmentation module is used to segment the liver and lesion regions of the preprocessed CT image to be segmented using a trained CT image lesion region segmentation model.

[0013] The segmentation module is based on the DeeplabV3+ semantic segmentation model structure and uses the Xception neural network as the backbone network.

[0014] The Xception neural network is configured with three parts: Entry flow, Middle flow, and Exit flow.

[0015] In the Entry flow, depthwise separable convolution is used to obtain shallow information of the image, and depthwise separable convolution with a stride of 2 is used to replace the pooling layer in the traditional network.

[0016] Use more convolutions in the middle flow to extract image features;

[0017] The final output feature map in the Exit flow has 2048 channels and a stride of 16.

[0018] The segmentation module adopts an Encoder-Decoder structure. In the Decoder, the feature map is subjected to trilinear interpolation and then concatenated with the feature map with the same resolution in the Encoder through a [1*1] convolution to adjust the channels. Finally, after a [3*3] convolution, the feature map is upsampled to obtain the final segmentation result.

[0019] Preferably, the preprocessing module resamples the liver CT image to a 5 mm voxel spacing, sets the network input size to [128*128*100] mm, adjusts the CT image to the liver window width and window level, and adjusts the number of channels to conform to the network input format.

[0020] Preferably, the preprocessing module sets the width of the liver window to 150 HU and the window position to 30 HU.

[0021] Preferably, the input CT image with labeled liver lesions is first processed by the Xception neural network to extract deep features, and then the shallow features are combined with the deep features by the decoder.

[0022] Preferably, when training the CT image lesion region segmentation model, FocalLoss is used as the loss function. First, the true value is one-hot encoded, and then the FocalLoss loss function is used to calculate the loss between the predicted value and the true value.

[0023] The preferred formula for the FocalLoss loss function is as follows:

[0024] L=-∑y i *logpi

[0025] Where y represents the true value and pi represents the predicted value, the FocalLoss loss is then calculated, and the FocalLoss loss formula is as follows:

[0026] FL = -(1-pt) γ *celoss

[0027] in:

[0028] pt=x*yi+(1-x)*yi

[0029] Where x represents the predicted value, y represents the true value, γ is a constant, and celoss is the cross-entropy loss.

[0030] Preferably, the liver CT images are first preprocessed using a preprocessing module, resampling them to a 5 mm voxel spacing, setting the network input size to [128*128*100] mm, adjusting the CT images to the liver window width and level, and adjusting the number of channels to conform to the network input format; then, a CT image lesion region segmentation model is established using a segmentation model building module; a specified number of labeled liver lesion CT images are acquired, and the CT image lesion region segmentation model is trained using the specified number of labeled liver lesion CT images through a training module; then, the segmentation module uses the trained CT image lesion region segmentation model to segment the liver and lesion regions in the preprocessed CT images to be segmented; and outputs the liver and lesion region segmentation results.

[0031] Preferably, when the training module trains the CT image lesion region segmentation model, it first uses depthwise separable convolution to obtain shallow information of the image in the Entryflow, and uses depthwise separable convolution with a stride of 2 to replace the pooling layer in the traditional network; then, it uses multiple convolutions in the Middleflow to obtain liver image features in the CT image, and finally outputs a feature map with 2048 channels and a stride of 16 in the Exitflow; in the Decoder, it uses trilinear interpolation on the feature map and combines it with the feature map with the same resolution as the Encoder, then adjusts the channels through [1*1] convolution and performs feature stitching, and finally performs upsampling after a [3*3] convolution to obtain the final segmentation result of the CT image lesion region.

[0032] The beneficial effects of this invention include:

[0033] The present invention provides a lesion segmentation device based on liver CT images. A preprocessing module preprocesses the liver CT images; a segmentation model building module establishes a lesion region segmentation model for the CT images, which is then trained by a training module; finally, the segmentation module uses the trained lesion region segmentation model to segment the liver and lesion regions in the preprocessed liver CT images. By preprocessing the liver CT images to adjust their size and contrast, and by using 3D convolution instead of 2D convolution during the training of the lesion region segmentation model to preserve the connections between slices, a [3*3] depthwise separable convolution is used for feature extraction, and a [1*1] convolution is used for feature fusion and channel number adjustment. This effectively improves the accuracy of lesion region segmentation in CT images and increases the efficiency of physicians. Attached Figure Description

[0034] Figure 1 This is a schematic diagram of the lesion segmentation device based on liver CT images of the present invention.

[0035] Figure 2 This is a schematic diagram of the CT image lesion region segmentation model of the present invention.

[0036] Figure 3 This is a schematic diagram of the structure of the present invention, which uses a CT image lesion region segmentation model for segmentation. Detailed Implementation

[0037] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely represents selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0038] This application will present various aspects, embodiments, or features relating to systems that may include multiple devices, components, modules, etc. It should be understood and appreciated that individual systems may include additional devices, components, modules, etc., and / or may not include all the devices, components, modules, etc. discussed in conjunction with the accompanying drawings. Furthermore, combinations of these approaches are also possible.

[0039] Furthermore, in the embodiments of this application, words such as "exemplarily" and "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design that is described as an "example" in this application should not be construed as being better or more advantageous than other embodiments or designs. Rather, the use of the word "example" is intended to present the concept in a specific manner.

[0040] In the embodiments of this application, the terms "information," "signal," "message," "channel," and "singaling" may sometimes be used interchangeably. It should be noted that, without emphasizing their distinction, they all convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing their distinction, they all convey the same meaning.

[0041] The following is in conjunction with the appendix Figures 1-3 The present invention will be further described in detail below:

[0042] See appendix Figures 1-3 As shown, a lesion segmentation device based on liver CT images includes a preprocessing module, a segmentation model establishment module, a training module, and a segmentation module; the preprocessing module is connected to the segmentation module, the segmentation model establishment module is connected to the training module, and the training module is connected to the segmentation module.

[0043] The preprocessing module is used to preprocess the CT images to be segmented;

[0044] The segmentation model establishment module is used to establish a segmentation model of lesion regions in CT images;

[0045] The training module is used to train the CT image lesion region segmentation model using a large number of CT images with labeled liver lesions.

[0046] The segmentation module is used to segment the liver and lesion regions of the preprocessed CT image to be segmented using a trained CT image lesion region segmentation model.

[0047] The segmentation module is based on the DeeplabV3+ semantic segmentation model structure and uses the Xception neural network as the backbone network.

[0048] The Xception neural network is configured with three parts: Entry flow, Middle flow, and Exit flow.

[0049] In the Entry flow, depthwise separable convolution is used to obtain shallow information of the image, and depthwise separable convolution with a stride of 2 is used to replace the pooling layer in the traditional network.

[0050] Use more convolutions in the middle flow to extract image features;

[0051] The final output feature map in the Exit flow has 2048 channels and a stride of 16.

[0052] The segmentation module adopts an Encoder-Decoder structure. In the Decoder, the feature map is subjected to trilinear interpolation and then concatenated with the feature map with the same resolution in the Encoder through a [1*1] convolution to adjust the channels. Finally, after a [3*3] convolution, the feature map is upsampled to obtain the final segmentation result.

[0053] Because liver CT images contain a large amount of information, the hardware requirements for deep learning are relatively high. Therefore, before segmenting liver CT images through the segmentation module, the size or contrast of the input CT images needs to be adjusted through the preprocessing module to reduce the hardware requirements during the deep learning process.

[0054] When training the CT image lesion region segmentation model, the training module first provides a large number of CT images with labeled liver lesions, ensuring that each CT image containing a lesion has a corresponding label. Then, a suitable CT image lesion region segmentation model is designed, and the training module then uses a large number of labeled liver lesion CT images to train this model.

[0055] Because the trained CT image lesion region segmentation model has the ability to predict or classify unknown liver CT images, the segmentation module can use this model to segment the liver and lesion regions in the preprocessed liver CT images, generating labels for both. Doctors can then directly identify the segmented liver CT images, effectively improving their work efficiency.

[0056] In the above scheme, the preprocessing module resamples the liver CT image to a 5 mm voxel spacing, sets the network input size to [128*128*100] mm, adjusts the CT image to the liver window width and window level, and adjusts the number of channels to conform to the network input format. The preprocessing module sets the liver window width to 150 HU, the window level to 30 HU, and adjusts the number of channels to conform to the network input format.

[0057] The CT image lesion region segmentation model in step S2 uses the Xception neural network as its backbone network, which contains 65 convolutional layers. This neural network structure is a linear stack of deep separable convolutional layers with residual connections, making the structure easy to define and modify.

[0058] The Xception neural network is configured with three parts: Entry flow, Middle flow, and Exit flow.

[0059] In the Entry flow, depthwise separable convolution is used to obtain shallow information of the image, and depthwise separable convolution with a stride of 2 is used to replace the pooling layer in the traditional network to adapt to different resolutions.

[0060] Use more convolutions in the middle flow to extract image features;

[0061] The final output feature map in the Exit flow has 2048 channels and a stride of 16.

[0062] In the above scheme, see Figure 2 and Figure 3The CT image lesion region segmentation model uses an Encoder-Decoder structure. In the Decoder, the feature map is interpolated using trilinear interpolation and then concatenated with a feature map of the same resolution in the Encoder via a [1*1] convolution to adjust the channels. Finally, after a [3*3] convolution, upsampling is performed to obtain the final prediction result. The network uses 3D convolutions instead of 2D convolutions to preserve the connections between slices. A [3*3] depthwise separable convolution is used for feature extraction, and a [1*1] convolution is used for feature fusion and channel number adjustment.

[0063] The input CT images with labeled liver lesions are first processed by Xception to extract deep features, and then the shallow features are combined with the deep features by the decoder to enhance the robustness of the model.

[0064] When training a CT image lesion region segmentation model, a loss function needs to be defined to measure the difference between the actual output and the target output in order to evaluate the accuracy of the segmentation results. In the model training process of this application, FocalLoss is used as the loss function. First, the ground truth values ​​are one-hot encoded, and then the FocalLoss loss function is used to calculate the loss between the predicted value and the ground truth value.

[0065] The parameters of the CT image lesion region segmentation model are optimized based on the FocalLoss loss function. During the optimization process, algorithms such as gradient descent can be used to iteratively adjust the parameters of the CT image lesion region segmentation model. The model is continuously optimized so that its error on the training dataset gradually decreases and converges to the minimum value.

[0066] The FocalLoss loss function used in the above scheme is formulated as follows:

[0067] L=-Σy i *logpi;

[0068] Where y represents the true value and pi represents the predicted value, the FocalLoss loss is then calculated, and the FocalLoss loss formula is as follows:

[0069] FL = -(1-py) γ *celoss;

[0070] in:

[0071] pt = x * yi + (1 - x) * yi;

[0072] Where x represents the predicted value, y represents the true value, γ is a constant, and celoss is the cross-entropy loss.

[0073] Furthermore, the Adam optimizer is used to adjust model parameters, including hyperparameters, to minimize the FocalLoss loss function, effectively attempting to find the optimal parameter combination that produces the most accurate predictions. The Adam optimizer's rated learning rate is set to 1e-5, and the batch size is set to 1. The model converges after 2000 epochs. Here, the Adam optimizer updates network parameters based on gradient information to minimize the FocalLoss loss function. For example, it adaptively adjusts the model's learning rate based on historical gradient information, using a larger learning rate in the early stages of model training for rapid convergence, and a smaller learning rate in the later stages to accurately find the minimum of the loss function. The Adam optimizer also adjusts the momentum parameter to balance the influence of the previous and current gradients on parameter updates, thus avoiding premature trapping in local minima. The Adam optimizer normalizes the model parameters, ensuring that updates to each model parameter have a similar magnitude, thereby improving training performance.

[0074] The formulas involved in updating the model parameters of the Adam optimizer are as follows:

[0075] m t =β1m t-1 +(1-β1)gt;

[0076]

[0077]

[0078]

[0079]

[0080] Where gt is the parameter gradient of the model, and β1 and β2 are the decay coefficients of the two exponentially weighted averages. and It is the moving average after gradient bias correction, θt+1 is the updated parameter, and η is the learning rate, used to avoid division by zero.

[0081] This lesion segmentation device based on liver CT images first acquires a large number of liver CT images during the lesion segmentation process. Then, a preprocessing module preprocesses the liver CT images, resampling them to a 5 mm voxel spacing, setting the network input size to [128*128*100] mm, adjusting the CT images to the liver window width and level, and adjusting the number of channels to conform to the network input format. Next, a segmentation model building module is used to establish a CT image lesion region segmentation model. Then, a specified number of labeled liver lesion CT images are acquired, and a training module trains the CT image lesion region segmentation model using these labeled liver lesion CT images.

[0082] When training the CT image lesion region segmentation model, the training module first uses depthwise separable convolution to obtain shallow information of the image in the Entry flow, and uses depthwise separable convolution with a stride of 2 to replace the pooling layer in the traditional network; then, multiple convolutions are used in the Middle flow to obtain liver image features in the CT image, and the final output feature map in the Exit flow has 2048 channels and a stride of 16; in the Decoder, trilinear interpolation is used on the feature map and it is matched with the feature map with the same resolution as the Encoder, and then the channels are adjusted by [1*1] convolution before feature stitching is performed. Finally, after a [3*3] convolution layer, upsampling is performed to obtain the final segmentation result of the CT image lesion region.

[0083] Then, the segmentation module uses the trained CT image lesion region segmentation model to segment the liver and lesion regions in the preprocessed CT image to be segmented; and outputs the segmentation results of the liver and lesion regions.

[0084] In summary, the present invention provides a lesion segmentation device based on liver CT images. The device preprocesses the liver CT images using a preprocessing module; establishes a lesion region segmentation model for the CT images using a segmentation model building module; trains the lesion region segmentation model using a training module; and finally, uses the trained lesion region segmentation model to segment the liver and lesion regions in the preprocessed liver CT images using a segmentation module. By preprocessing the liver CT images to adjust their size and contrast to appropriate levels, and by using 3D convolution instead of 2D convolution to preserve the relationships between slices during the training of the lesion region segmentation model, the device uses [3*3] depthwise separable convolution for feature extraction and [1*1] convolution for feature fusion and channel number adjustment. This effectively improves the accuracy of lesion region segmentation in CT images and increases the efficiency of physicians.

[0085] The embodiments described above merely illustrate specific implementation methods of this application, and while the descriptions are detailed and specific, they should not be construed as limiting the scope of protection of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the technical solution of this application, and these modifications and improvements all fall within the scope of protection of this application.

Claims

1. A liver-CT image-based lesion segmentation apparatus, characterized by comprising: The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device.

2. The lesion segmentation apparatus based on liver CT images according to claim 1, wherein The application relates to a CT image liver and lesion region segmentation method and device. 3.The lesion segmentation device based on liver CT images of claim 2, wherein, The application relates to a CT image liver and lesion region segmentation method and device.

4. The lesion segmentation apparatus based on liver CT images according to claim 1, wherein The application relates to a CT image liver and lesion region segmentation method and device.

5. The lesion segmentation apparatus based on liver CT images according to claim 4, wherein, The application relates to a CT image liver and lesion region segmentation method and device.

6. The lesion segmentation apparatus based on liver CT images according to claim 5, wherein, The application relates to a CT image liver and lesion region segmentation method and device. L = -∑y i *logpi; The application relates to a CT image liver and lesion region segmentation method and device. FL = -(1 - pt) γ *celoss; The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device. The application relates to a CT image liver and lesion region segmentation method and device.

7. The lesion segmentation apparatus based on liver CT images according to claim 1, wherein The liver CT image is preprocessed by a preprocessing module, the liver CT image is resampled to a 5mm voxel interval, the network input size is set to [128*128*100]mm, the CT image is adjusted to a liver window width and window level, and the number of channels is adjusted to conform to the network input format; a CT image lesion region segmentation model is established by a segmentation model establishment module; a specified number of CT images with labeled liver lesions are obtained, the CT image lesion region segmentation model is trained by a training module on the specified number of CT images with labeled liver lesions; then a segmentation module uses the trained CT image lesion region segmentation model to segment the liver and lesion regions of the preprocessed CT image to be segmented; and the liver and lesion region segmentation result is output. 8.The lesion segmentation device based on liver CT images of claim 7, wherein, When the training module trains the CT image lesion region segmentation model, first, the shallow information of the image is obtained by using a depth separable convolution in the Entry flow, and a depth separable convolution with a step of 2 is used instead of a pooling layer in a traditional network; then in the Middle flow, a plurality of convolutions are used to obtain the liver image features in the CT image, and in the Exit flow, a feature map with a channel number of 2048 and a stride of 16 is finally output; in the Decoder, the feature map is used for trilinear interpolation, and is combined with the feature map with the same resolution in the Encoder, then after adjusting the channels by [1*1] convolution, feature splicing is performed, and finally after a layer of [3*3] convolution, up-sampling is performed to obtain the final segmentation result of the CT image lesion region.

Images