Method, system and device for automatic segmentation of clinical target area based on contour prior guidance

Abstract

The present invention provides a method, system and device for automatic segmentation of clinical target area based on contour prior guidance, including: acquiring patient image data; registering the scanned image before treatment with the scanned image during treatment, selecting a set registration method to generate a registration mapping according to the characteristics of the input image, applying the mapping to the CTV contour before treatment, generating the registered CTV contour, and using it as a priori guidance for the model; inputting the scanned image during treatment and the registered CTV contour into a dual-branch network model for feature extraction respectively, using the contour information before treatment in the entire segmentation process to assist the segmentation network in accurately identifying and adapting to the changes in the CTV boundary; training through rich medical imaging data, adjusting and optimizing model parameters; obtaining the automatic segmentation results of the clinical target area through a trained deep learning segmentation model. The present invention improves the work efficiency of doctors and promotes the automation development of adaptive radiotherapy processes.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image segmentation, and in particular to a method and system for automatic segmentation of a clinical target area based on contour prior guidance. Background Art

[0002] Radiotherapy (RT) is an effective means to improve the survival rate of patients with various types of cancer, among which accurate segmentation of the clinical target volume (CTV) is the key to ensure the success of radiotherapy. During the treatment process, since the tumor and surrounding organs may change significantly, it is necessary to re-outline the target area and optimize the dose distribution based on the latest imaging information. Adaptive radiotherapy (ART) uses various imaging devices to monitor tumors and normal organs in real time before and during treatment, and updates the target area contour according to tumor changes to improve the treatment effect and reduce damage to healthy tissues. However, the implementation of ART requires a lot of resources, especially the frequent outlining of CTV contours.

[0003] In recent years, deep learning automatic segmentation algorithms have been widely used in the field of radiotherapy, but they have limitations in clinical application, especially in the accuracy of CTV segmentation. CTV re-segmentation is different from initial segmentation. It can use previous contour information, and there is a correlation between the latest contour and the initial contour. However, existing technologies do not consider the particularity of CTV delineation in adaptive radiotherapy, but simply apply deep learning algorithms directly to the CTV segmentation task. This approach is not reasonable, resulting in the segmentation model often being of low accuracy. Summary of the invention

[0004] To solve the above problems, the present invention aims to propose a method and system for automatic segmentation of clinical target area based on contour prior guidance, which is a new method that can comprehensively consider the characteristics of CTV delineation in ART, make full use of the contour information before treatment, and improve the accuracy of CTV segmentation.

[0005] To achieve the above object, the technical solution of the present invention is achieved as follows:

[0006] A method for automatic segmentation of a clinical target area based on contour prior guidance comprises the following steps:

[0007] S1: Acquire patient image data: scanned image and corresponding CTV contour before treatment; scanned image and corresponding CTV contour during treatment;

[0008] S2: The pre-treatment scan image is registered with the scan image during treatment. According to the characteristics of the input image, a set registration method is selected to generate a registration map, and then this map is applied to the pre-treatment CTV contour to generate a registered CTV contour, which is used as a priori guidance for the deep learning segmentation model;

[0009] S3: The scanned images during the treatment and the registered CTV contours are input into the dual-branch network model under the deep learning segmentation model for feature extraction. The contour information before treatment is used in the entire segmentation process to assist the segmentation network in accurately identifying and adapting to the changes in the CTV boundaries.

[0010] S4: Train with rich medical imaging data to adjust and optimize deep learning segmentation model parameters;

[0011] S5: Obtain automatic segmentation results of clinical target areas through the trained deep learning segmentation model.

[0012] Furthermore, in step S3, the dual-branch network model includes two branches, namely a contour prior sub-network and a CTV segmentation sub-network. In the encoding stage, the features of the two branches are integrated through the attention fusion module and transmitted to the next layer of the CTV segmentation sub-network; in the decoding stage, the features are transmitted to the subsequent layers of the CTV segmentation sub-network through channel splicing; the output features of the decoding layer are collected and fused through the CTV sub-segmentation network to generate an accurate CTV segmentation map.

[0013] Furthermore, the contour prior sub-network is responsible for extracting contour features before treatment, and the CTV segmentation sub-network is responsible for extracting image features during treatment.

[0014] Furthermore, the contour prior subnetwork is a simplified U-Net architecture, including five encoding layers and four decoding layers; two 3x3 convolutional blocks connected in series are used to extract the rough position and shape information of CTV, and the generated features are The CTV segmentation subnetwork is composed of a series of U-shaped feature extractors, which are nested in a U-shaped network and use multiple combined downsampling modules to gradually reduce the size of the feature map and expand the receptive field, extract multi-scale features, and generate features as follows: .

[0015] Furthermore, the combined downsampling module adopts a simulated feature pyramid structure, and by integrating different downsampling methods, captures and retains multi-level information in the image on a single feature map, avoiding information loss caused by multiple downsampling in the U-shaped feature extractor, specifically:

[0016]

[0017]

[0018]

[0019]

[0020] The input is , the output is , , and represents a 3x3 convolution operation with a stride of 2, average pooling, and maximum pooling; the symbol represents the addition of elements, represents the batch normalization operation, Represents the ReLU activation function.

[0021] Furthermore, in the encoding stage, the attention fusion module enables the CTV segmentation sub-network to pay more attention to the potential target area through multi-scale convolution and attention weight mechanism; specifically, the following steps are included:

[0022] Step 1: Perform multi-scale convolution on the features of the contour prior sub-network and generate after channel splicing , to accommodate the changes in size and shape of different CTV profiles;

[0023]

[0024] in, Respectively , , The convolution operation, Represents a channel splicing operation;

[0025] Step 2: Generate attention weights through the sigmoid function to weight the features extracted by the CTV segmentation subnetwork so that it pays more attention to the potential CTV area;

[0026]

[0027]

[0028] in, is the generated attention weight, symbol represents element-wise multiplication, is the weighted feature;

[0029] Step 3: Concatenate the weighted features with the original features, and finally obtain the output features through a 1x1 convolution operation , ensuring that the model can utilize both fine-grained weighted features and comprehensive original features to obtain a richer feature representation;

[0030] )).

[0031] Furthermore, in the decoding stage, in order to simplify the processing flow, the features of the two branches are fused by channel splicing to obtain And passed to the next layer of the CTV segmentation sub-network;

[0032] .

[0033] In order to achieve the above object, the present invention also provides a clinical target area automatic segmentation system based on contour prior guidance, comprising the following modules:

[0034] Input processing module: used to obtain patient image data: scanned images before treatment and corresponding CTV contours; scanned images during treatment and corresponding CTV contours;

[0035] Contour mapping module: used to register the scanned image before treatment with the scanned image during treatment. According to the characteristics of the input image, an appropriate registration method is selected to generate a registration map, which is then applied to the pre-treatment CTV contour to generate the registered CTV contour, which is used as a priori guidance for the deep learning segmentation model;

[0036] Model building module: used to input the scanned image during treatment of the input processing module and the registered CTV contour in the contour mapping module into the dual-branch network model under the deep learning segmentation model for feature extraction respectively, and use the contour information before treatment in the whole segmentation process to assist the segmentation network to accurately identify and adapt to the changes of CTV boundaries;

[0037] Training and tuning module: used to train, adjust and optimize deep learning segmentation model parameters through rich medical imaging data;

[0038] Automatic segmentation module: used to obtain automatic segmentation results of clinical target areas through the trained deep learning segmentation model.

[0039] In order to achieve the above-mentioned purpose, the present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the automatic clinical target area segmentation method based on contour prior guidance as described above is implemented.

[0040] Beneficial effects: The present invention can automatically predict the segmentation results of various medical images based on the images during the patient's treatment and the contour data before the treatment; the images of different periods can be roughly aligned using the registration technology to construct contour prior information; then a two-branch fusion network is constructed to learn contour features and image features respectively; a deep learning model based on a convolutional neural network is established and trained using a wealth of clinical cases, and the model can automatically predict the segmentation results of CTV. For new patients, only the contour before treatment and the images during treatment need to be input, and the present invention can quickly and accurately predict the segmentation results personalized for the patient, assisting professional physicians in formulating new radiotherapy plans. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The accompanying drawings constituting a part of the present invention are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the accompanying drawings:

[0042] Figure 1 It is a flow chart of the method for automatic segmentation of clinical target area based on contour prior guidance according to an embodiment of the present invention;

[0043] Figure 2 A schematic diagram of the architecture of a deep learning segmentation model of a clinical target area automatic segmentation method based on contour prior guidance according to an embodiment of the present invention;

[0044] Figure 3 It is a schematic diagram of the architecture of a U-shaped feature extractor in the method for automatic segmentation of a clinical target area based on contour prior guidance according to an embodiment of the present invention;

[0045] Figure 4 It is a schematic diagram of the architecture of the combined downsampling module in the automatic segmentation method of clinical target area based on contour prior guidance according to an embodiment of the present invention;

[0046] Figure 5 It is a schematic diagram of the architecture of the attention fusion module in the method for automatic segmentation of clinical target areas based on contour prior guidance according to an embodiment of the present invention;

[0047] Figure 6 It is a schematic diagram of the CTV segmentation result of the clinical target area automatic segmentation method based on contour prior guidance described in an embodiment of the present invention for a cervical cancer CT image;

[0048] Figure 7 It is a schematic diagram of the structure of the automatic clinical target area segmentation system based on contour prior guidance described in an embodiment of the present invention. DETAILED DESCRIPTION

[0049] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0050] Example 1

[0051] See also Figure 1-6 : A method for automatic segmentation of clinical target area based on contour prior guidance, comprising the following steps:

[0052] S1: Acquire patient image data: scanned image and corresponding CTV contour before treatment; scanned image and corresponding CTV contour during treatment;

[0053] S2: The pre-treatment scan image is registered with the scan image during treatment. According to the characteristics of the input image, a set registration method is selected to generate a registration map, and then this map is applied to the pre-treatment CTV contour to generate a registered CTV contour, which is used as a priori guidance for the deep learning segmentation model;

[0054] S3: The scanned images during the treatment and the registered CTV contours are input into the dual-branch network model under the deep learning segmentation model for feature extraction. The contour information before treatment is used in the entire segmentation process to assist the segmentation network in accurately identifying and adapting to the changes in the CTV boundaries.

[0055] S4: Train with rich medical imaging data to adjust and optimize deep learning segmentation model parameters;

[0056] S5: Obtain automatic segmentation results of clinical target areas through the trained deep learning segmentation model.

[0057] This embodiment can automatically predict the segmentation results of various medical images based on the images during the patient's treatment and the contour data before treatment; the images of different periods can be roughly aligned using registration technology to construct contour prior information; then a two-branch fusion network is constructed to learn contour features and image features respectively; a deep learning model based on a convolutional neural network is established and trained using a wealth of clinical cases, and the model can automatically predict the segmentation results of CTV. For new patients, only the contour before treatment and the images during treatment need to be input, and this embodiment can quickly and accurately predict the segmentation results personalized for the patient, assisting professional physicians in formulating new radiotherapy plans.

[0058] In a specific example, in step S3, the dual-branch network model includes two branches, namely a contour prior sub-network and a CTV segmentation sub-network. In the encoding stage, the features of the two branches are integrated through the attention fusion module and passed to the next layer of the CTV segmentation sub-network; in the decoding stage, the features are passed to the subsequent layers of the CTV segmentation sub-network through channel splicing; the output features of the decoding layer are collected and fused through the CTV sub-segmentation network to generate an accurate CTV segmentation map.

[0059] This embodiment fuses the two through a unique attention fusion module in the encoding stage and through simple channel splicing in the decoding stage, which not only ensures the fusion effect but also reduces the computational complexity.

[0060] In a specific example, the contour prior sub-network is responsible for extracting contour features before treatment, and the CTV segmentation sub-network is responsible for extracting image features during treatment.

[0061] This embodiment can select a suitable registration method to convert the contour image features before treatment to the image during treatment, ensuring that these contours conform to the current anatomical structure and ensuring that the patient's previous specific information can be effectively utilized.

[0062] In a specific example, the contour prior subnetwork is a simplified U-Net architecture, including five encoding layers and four decoding layers; two 3x3 convolutional blocks connected in series are used to extract the rough position and shape information of the CTV, and the generated features are The CTV segmentation subnetwork is composed of a series of U-shaped feature extractors, which are nested in a U-shaped network and use multiple combined downsampling modules to gradually reduce the size of the feature map and expand the receptive field, extract multi-scale features, and generate features as follows: .

[0063] In a specific example, the combined downsampling module adopts a simulated feature pyramid structure, but it does not obtain multi-scale feature maps through different levels in the convolutional network, but captures and retains multi-level information in the image on a single feature map by integrating different downsampling methods, avoiding information loss caused by multiple downsampling in the U-shaped feature extractor, specifically:

[0064]

[0065]

[0066]

[0067]

[0068] The input is , the output is , , and represents a 3x3 convolution operation with a stride of 2, average pooling, and maximum pooling; the symbol represents the addition of elements, represents the batch normalization operation, Represents the ReLU activation function.

[0069] In a specific example, in the encoding stage, the attention fusion module uses multi-scale convolution and attention weight mechanism to make the CTV segmentation sub-network pay more attention to the potential target area, thereby reducing unnecessary search range and more effectively utilizing prior information; specifically, the following steps are included:

[0070] Step 1: Perform multi-scale convolution on the features of the contour prior sub-network and generate after channel splicing , to accommodate the changes in size and shape of different CTV profiles;

[0071]

[0072] in, Respectively , , The convolution operation, Represents a channel splicing operation;

[0073] Step 2: Generate attention weights through the sigmoid function to weight the features extracted by the CTV segmentation sub-network, so that it pays more attention to the potential CTV area, thereby reducing unnecessary search range and improving segmentation efficiency;

[0074]

[0075]

[0076] in, is the generated attention weight, symbol represents element-wise multiplication, is the weighted feature;

[0077] Step 3: Concatenate the weighted features with the original features, and finally obtain the output features through a 1x1 convolution operation , ensuring that the model can utilize both fine weighted features and comprehensive original features to obtain a richer feature representation;

[0078] )).

[0079] In a specific example, in the decoding stage, in order to simplify the processing flow, the features of the two branches are fused by channel splicing to obtain And pass it to the next layer of the CTV segmentation sub-network, so that the network can use the extracted features for subsequent segmentation tasks more quickly, reducing computational complexity and potential overfitting risks;

[0080] .

[0081] In a specific example, the deep learning segmentation model collects the output of each decoding layer of the CTV segmentation subnetwork, upsamples it to the same resolution as the input image, merges the channels of all feature maps, applies 1x1 convolution and sigmoid function, and obtains the final CTV segmentation result.

[0082] In a specific example, the deep learning segmentation model uses a training set to train the model and uses a test set to evaluate its performance. During the training phase, the SGD optimizer is used, the batch size is set to 24, a total of 150 rounds of training are performed, the initial learning rate is set to 0.005, and a dynamic learning rate adjustment strategy is adopted. The learning rate lr is updated according to the following formula:

[0083]

[0084] in, represents the initial learning rate, is the current iteration number, To allow for the total number of iterations, gradually reducing the learning rate during training helps the model converge to the optimal solution better in the later stages.

[0085] It should be noted that the training set of this embodiment can use the clinical target segmentation dataset of cervical cancer patients, and each case includes: (1) CT scan before treatment and the corresponding CTV outline; (2) CT scan during treatment and the corresponding CTV outline. The data used are all carefully delineated by radiologists based on CT scan images. These annotations are used as true labels in our experiments to train and verify the performance of the model. For data set division, the training set and test set are divided in a ratio of 4:1. The image is cropped to 224×224 pixels, the pixel value is limited to the range of -200 to 300 to highlight the key area, and linear normalization is performed to scale the pixel value to 0 to 1 to optimize model training. In addition, the training data undergoes data enhancement operations such as flipping and translation, while the test data is not subjected to data enhancement processing.

[0086] In a specific implementation, when processing medical images, this embodiment only needs to input the image to be processed, and the CTV segmentation subnetwork of the deep learning segmentation model will automatically identify and segment the CTV area according to the image content. Figure 6 , the method of this embodiment is used to segment the cervical cancer CT image. The results in the figure show that the segmentation result of Example 1 is basically consistent with the true label.

[0087] In summary, the model of this embodiment adopts a dual-branch network structure design, which cleverly introduces contour prior information into the deep learning model, providing clear guidance for the model; by using a U-shaped feature extractor in the CTV segmentation subnetwork, multi-scale features can be extracted, the feature expression ability of the network is enhanced, and it is helpful to identify complex CTV morphology. The constructed attention fusion module helps to optimize the feature fusion process, further enhance the recognition ability of CTV boundaries, and significantly improve the accuracy of CTV segmentation. This embodiment uses the previous specific data of each patient, which is more in line with the actual clinical delineation process. It not only improves the work efficiency of doctors, but also provides patients with more personalized segmentation results and formulates more accurate treatment plans. It has important clinical application value and practical significance.

[0088] Example 2

[0089] To achieve the above purpose, see Figure 7 :This embodiment also provides a clinical target area automatic segmentation system based on contour prior guidance, including the following modules:

[0090] Input processing module: used to obtain patient image data: scanned images before treatment and corresponding CTV contours; scanned images during treatment and corresponding CTV contours;

[0091] Contour mapping module: used to register the scanned image before treatment with the scanned image during treatment. According to the characteristics of the input image, an appropriate registration method is selected to generate a registration map, which is then applied to the pre-treatment CTV contour to generate the registered CTV contour, which is used as a priori guidance for the deep learning segmentation model;

[0092] Model building module: used to input the scanned image during treatment of the input processing module and the registered CTV contour in the contour mapping module into the dual-branch network model under the deep learning segmentation model for feature extraction respectively, and use the contour information before treatment in the whole segmentation process to assist the segmentation network to accurately identify and adapt to the changes of CTV boundaries;

[0093] Training and tuning module: used to train, adjust and optimize deep learning segmentation model parameters through rich medical imaging data;

[0094] Automatic segmentation module: used to obtain automatic segmentation results of clinical target areas through the trained deep learning segmentation model.

[0095] The automatic clinical target area segmentation system based on contour prior guidance of this embodiment has the same advantages as the automatic clinical target area segmentation method based on contour prior guidance described above over the prior art, which will not be described in detail here.

[0096] Example 3

[0097] This embodiment also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, and the method of embodiment 1 is implemented when the processor executes the computer program.

[0098] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for automatic segmentation of clinical target volume based on contour prior guidance, characterized in that: The following steps are involved: S1: Acquire patient image data: scanned image and corresponding CTV contour before treatment; scanned image and corresponding CTV contour during treatment; S2: The pre-treatment scan image is registered with the scan image during treatment. According to the characteristics of the input image, a set registration method is selected to generate a registration map, and then this map is applied to the pre-treatment CTV contour to generate a registered CTV contour, which is used as a priori guidance for the deep learning segmentation model; S3: The scanned images during the above treatment and the registered CTV contour are input into the dual-branch network model under the deep learning segmentation model for feature extraction respectively. The contour information before treatment is used in the whole segmentation process to assist the segmentation network to accurately identify and adapt to the changes in the CTV boundary. In step S3, the dual-branch network model includes two branches: a contour prior sub-network and a CTV segmentation sub-network. In the encoding stage, the features of the two branches are integrated through the attention fusion module and transmitted to the next layer of the CTV segmentation sub-network; in the decoding stage, the features are transmitted to the subsequent layers of the CTV segmentation sub-network through channel splicing; the output features of the decoding layer are collected and fused through the CTV sub-segmentation network to generate an accurate CTV segmentation map; S4: Train, adjust and optimize deep learning segmentation model parameters through medical imaging data; S5: The automatic segmentation results of the clinical target area are obtained through the trained deep learning segmentation model. The deep learning segmentation model collects the output of each decoding layer of the CTV segmentation subnetwork, upsamples it to the same resolution as the input image, merges the channels of all feature maps, applies 1x1 convolution and sigmoid function, and obtains the final CTV segmentation result.

2. The method for automatic segmentation of clinical target volume based on contour prior guidance according to claim 1, characterized in that: The contour prior sub-network is responsible for extracting contour features before treatment, and the CTV segmentation sub-network is responsible for extracting image features during treatment.

3. The method for automatic segmentation of clinical target volume based on contour prior guidance according to claim 1, characterized in that: The contour prior subnetwork is a simplified U-Net architecture, including five encoding layers and four decoding layers. Two 3x3 convolutional blocks in series are used to extract the rough position and shape information of CTV. The generated features are The CTV segmentation subnetwork is composed of a series of U-shaped feature extractors, which are nested in a U-shaped network and use multiple combined downsampling modules to gradually reduce the size of the feature map and expand the receptive field, extract multi-scale features, and generate features as follows: .

4. The method for automatic segmentation of clinical target volume based on contour prior guidance according to claim 3, characterized in that: The combined downsampling module adopts a simulated feature pyramid structure and captures and retains multi-level information in the image on a single feature map by integrating different downsampling methods, thereby avoiding information loss caused by multiple downsampling in the U-shaped feature extractor. Specifically: The input is , the output is , , and represents a 3x3 convolution operation with a stride of 2, average pooling, and maximum pooling; the symbol represents the addition of elements, represents the batch normalization operation, Represents the ReLU activation function.

5. The method for automatic segmentation of clinical target volume based on contour prior guidance according to claim 1, characterized in that: In the encoding stage, the attention fusion module uses multi-scale convolution and attention weight mechanism to make the CTV segmentation sub-network pay more attention to the potential target area; specifically, the following steps are included: Step 1: Perform multi-scale convolution on the features of the contour prior sub-network and generate after channel splicing , to accommodate the changes in size and shape of different CTV profiles; in, Respectively , , The convolution operation, Represents a channel splicing operation; Step 2: Generate attention weights through the sigmoid function to weight the features extracted by the CTV segmentation sub-network so that it pays more attention to the potential CTV area; in, for Activation function, is the generated attention weight, symbol represents element-wise multiplication, is the weighted feature; Step 3: Concatenate the weighted features with the original features, and finally obtain the output features through a 1x1 convolution operation , ensuring that the model can utilize both fine-grained weighted features and comprehensive original features to obtain a richer feature representation; ))。 6. The method for automatic segmentation of clinical target volume based on contour prior guidance according to claim 5, characterized in that: In the decoding stage, in order to simplify the processing flow, the features of the two branches are fused by channel splicing to obtain And passed to the next layer of the CTV segmentation sub-network; 。 7. A clinical target volume automatic segmentation system based on contour prior guidance, characterized in that: Includes the following modules: Input processing module: used to obtain patient image data: scanned images before treatment and corresponding CTV contours; scanned images during treatment and corresponding CTV contours; Contour mapping module: used to register the scanned image before treatment with the scanned image during treatment. According to the characteristics of the input image, an appropriate registration method is selected to generate a registration map, which is then applied to the pre-treatment CTV contour to generate the registered CTV contour, which is used as a priori guidance for the deep learning segmentation model; Model construction module: used to input the scanned image during treatment of the input processing module and the registered CTV contour in the contour mapping module into the dual-branch network model under the deep learning segmentation model for feature extraction respectively, and use the contour information before treatment in the whole segmentation process to assist the segmentation network to accurately identify and adapt to the changes in the CTV boundary. The dual-branch network model includes two branches: a contour prior sub-network and a CTV segmentation sub-network. In the encoding stage, the features of the two branches are integrated through the attention fusion module and transmitted to the next layer of the CTV segmentation sub-network; in the decoding stage, the features are transmitted to the subsequent layers of the CTV segmentation sub-network through channel splicing; the output features of the decoding layer are collected and fused through the CTV sub-segmentation network to generate an accurate CTV segmentation map; Training and tuning module: used to train, adjust and optimize deep learning segmentation model parameters through rich medical imaging data; Automatic segmentation module: used to obtain the automatic segmentation results of the clinical target area through the trained deep learning segmentation model. The deep learning segmentation model collects the output of each decoding layer of the CTV segmentation subnetwork, upsamples it to the same resolution as the input image, merges the channels of all feature maps, applies 1x1 convolution and sigmoid function, and obtains the final CTV segmentation result.

8. A computer device, characterized in that: The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the method for automatic segmentation of a clinical target area based on contour prior guidance as described in any one of claims 1 to 6 is implemented.

Images