Method and system for automatically sketching high-risk clinical target regions and endangered organs in cervical cancer brachytherapy based on deep learning

Abstract

The invention provides a deep learning-based automatic sketching method and system for high-risk clinical target regions and endangered organs in cervical cancer brachytherapy, and is applied to the technical field of data processing. According to the application, CT image information and MR image information of a cervical cancer patient are processed based on a positioning network, and 3D position information and form information of a source applicator and image feature sequence information are generated; performing classification processing on the CT image information of the cervical cancer patient based on a positioning network, and generating a priori knowledge label; processing the priori knowledge label based on the segmentation network, and generating a source applicator positioning target factor; the source applicator positioning target factor, the 3D position information and form information of the source applicator and the image feature sequence information are processed based on the segmentation network, and high-risk clinical target region and organ segmentation parameter information is generated; and processing the high-risk clinical target region and organ segmentation parameter information to generate an automatic sketching result of the high-risk clinical target region and the organ at risk for cervical cancer close-range treatment.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and particularly relates to an automatic delineation method and system for high-risk clinical target areas and organs at risk in cervical cancer brachytherapy based on deep learning. Background Art

[0002] In the field of cervical cancer treatment, brachytherapy is an important means, and the positioning of the applicator and the delineation of high-risk clinical target areas and organs at risk are extremely crucial. However, there are obvious defects in the current technology. In the positioning and delineation process, it mainly relies on manual operation by doctors. This method is not only inefficient, but also due to the differences in doctors' experience and professional levels, the accuracy of the delineation results varies greatly, with large individual differences.

[0003] In terms of image reconstruction, there are interference factors affecting the image quality, and the resolution ability for soft tissues is also poor, increasing the difficulty for accurate positioning and delineation. Existing automatic segmentation algorithms are also not ideal, with the accuracy not meeting clinical requirements, requiring a large amount of manual intervention, and consuming a large amount of hardware resources during operation, being restricted in practical applications. Generally speaking, there is an urgent need for a new technical solution to solve these problems, so as to improve the accuracy and efficiency of delineating high-risk clinical target areas and organs at risk in cervical cancer brachytherapy, reduce the workload of doctors, and reduce the impact of individual differences.

[0004] It should be noted that the information disclosed in the above background art section is only used to enhance the understanding of the background of the present disclosure, and therefore includes information that does not constitute the prior art known to those of ordinary skill in the art. Summary of the Invention

[0005] The purpose of the present application is to provide an automatic delineation method and system for high-risk clinical target areas and organs at risk in cervical cancer brachytherapy based on deep learning, which at least to a certain extent overcomes the problems existing in the prior art. By using the 3D U-Net network structure, the segmentation process is divided into two parts: a positioning network and a segmentation network. The positioning network extracts features through a region classifier, multi-layer surfaces, and pooling layers, and after being processed by a fully connected layer, it realizes the 3D positioning and reconstruction of the applicator, and generates an image feature sequence and a prior knowledge label. Then, the segmentation network generates a positioning target factor for the applicator based on the prior knowledge label, and then combines the 3D position, shape, and image feature sequence information of the applicator, and after being processed by feature extraction, fusion, and a decoder, it generates high-risk clinical target area and organ segmentation parameter information. Finally, the segmentation parameter information is subjected to format conversion and visual rendering, and with the help of a preset image post-processing algorithm and an edge detection algorithm, an automatic delineation result of high-risk clinical target areas and organs at risk in cervical cancer brachytherapy is generated. It realizes full automation from image input to segmentation result, reduces manual intervention, and improves the segmentation accuracy and efficiency.

[0006] Other features and advantages of the present application will become apparent from the following detailed description, or be learned in part from the practice of the present invention.

[0007] According to one aspect of the present application, there is provided an automatic delineation method for high-risk clinical target regions and organs at risk in cervical cancer brachytherapy based on deep learning, including: obtaining CT image information and MR image information of a cervical cancer patient, high-risk clinical target region and organ at risk delineation requirement information, and an automatic cervical cancer delineation model, wherein the automatic cervical cancer delineation model includes a localization network and a segmentation network; processing the CT image information and MR image information of the cervical cancer patient based on the localization network to generate 3D position information and morphological information of the applicator, as well as image feature sequence information; classifying and processing the CT image information of the cervical cancer patient based on the localization network to generate prior knowledge labels; processing the prior knowledge labels based on the segmentation network to generate applicator localization target factors; processing the applicator localization target factors, 3D position information and morphological information of the applicator, and image feature sequence information based on the segmentation network to generate high-risk clinical target region and organ segmentation parameter information; processing the high-risk clinical target region and organ segmentation parameter information to generate an automatic delineation result of the high-risk clinical target region and organs at risk in cervical cancer brachytherapy.

[0008] Another aspect of the present application is an automatic delineation device for high-risk clinical target regions and organs at risk in cervical cancer brachytherapy based on deep learning, characterized by including: an acquisition module for obtaining CT image information and MR image information of a cervical cancer patient, high-risk clinical target region and organ at risk delineation requirement information, and an automatic cervical cancer delineation model, wherein the automatic cervical cancer delineation model includes a localization network and a segmentation network; a processing module for processing the CT image information and MR image information of the cervical cancer patient based on the localization network to generate 3D position information and morphological information of the applicator, as well as image feature sequence information; classifying and processing the CT image information of the cervical cancer patient based on the localization network to generate prior knowledge labels; processing the prior knowledge labels based on the segmentation network to generate applicator localization target factors; processing the applicator localization target factors, 3D position information and morphological information of the applicator, and image feature sequence information based on the segmentation network to generate high-risk clinical target region and organ segmentation parameter information; processing the high-risk clinical target region and organ segmentation parameter information to generate an automatic delineation result of the high-risk clinical target region and organs at risk in cervical cancer brachytherapy.

[0009] According to yet another aspect of the present application, there is provided a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a second processor, it implements the above-mentioned automatic delineation method for high-risk clinical target regions and organs at risk in cervical cancer brachytherapy based on deep learning.

[0010] A method and system for automatically delineating high-risk clinical target regions and organs at risk in cervical cancer brachytherapy based on deep learning. The server uses a 3D U-Net network structure to divide the segmentation process into a localization network and a segmentation network. The localization network extracts features through a region classifier, multiple curved surfaces, and pooling layers, and after being processed by a fully connected layer, it realizes 3D localization and reconstruction of the applicator, and generates an image feature sequence and a priori knowledge label. Then, the segmentation network generates an applicator localization target factor based on the a priori knowledge label, and then combines the 3D position, shape, and image feature sequence information of the applicator, and through feature extraction, fusion, and decoder processing, generates high-risk clinical target region and organ segmentation parameter information. Finally, the segmentation parameter information is subjected to format conversion and visualization drawing, and with the help of a preset image post-processing algorithm and edge detection algorithm, an automatic delineation result of the high-risk clinical target region and organs at risk in cervical cancer brachytherapy is generated. It realizes full automation from image input to segmentation result, reduces manual intervention, improves segmentation accuracy and efficiency, and has important clinical application value.

[0011] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and do not limit the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 A flowchart showing a method for automatically delineating high-risk clinical target regions and organs at risk in cervical cancer brachytherapy based on deep learning provided by an embodiment of the present application;

[0013] Figure 2 A schematic structural diagram showing an apparatus for automatically delineating high-risk clinical target regions and organs at risk in cervical cancer brachytherapy based on deep learning provided by an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0014] The following describes the preferred embodiments of the present invention with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only for the purpose of illustrating and explaining the present invention, and are not used to limit the present invention.

[0015] The following combines Figure 1 to describe a method for automatically delineating high-risk clinical target regions and organs at risk in cervical cancer brachytherapy based on deep learning according to an exemplary embodiment of the present application. In one embodiment, the present application also proposes a method and system for automatically delineating high-risk clinical target regions and organs at risk in cervical cancer brachytherapy based on deep learning. Figure 1 Schematically shows a flowchart of a method for automatically delineating high-risk clinical target regions and organs at risk in cervical cancer brachytherapy based on an embodiment of the present application. As Figure 1 shown, the method is applied to a server and includes:

[0016] S101. Obtain the CT image information, MR image information, high-risk clinical target volume and organ-at-risk delineation requirement information, and the cervical cancer automatic delineation model of the cervical cancer patient.

[0017] In one implementation, in the radiotherapy department of a large general hospital, a patient diagnosed with cervical cancer was admitted. To formulate an accurate brachytherapy plan, the medical team first arranged for the patient to undergo CT and MR scans. The CT scan used a multi-slice spiral CT device, and the scanning range was from the symphysis pubis to the iliac crest level of the patient. The scanning parameters were set as tube voltage 120 kV, tube current 250 mA, and slice thickness 1 mm. This could obtain images with high density resolution, clearly showing structures such as bones and metal applicators, providing an important basis for determining the position and shape of the applicator subsequently. At the same time, an MR scan was performed using a 3.0T magnetic resonance device, applying a T2-weighted imaging sequence with scanning parameters of repetition time (TR) 4000 ms, echo time (TE) 100 ms, and slice thickness also 1 mm. MR images could clearly distinguish soft tissues, and had excellent display effects on the boundaries between cervical tumors and surrounding normal tissues such as the uterus, vagina, rectum, and bladder, helping to accurately define the scope of high-risk clinical target volumes and organs at risk. Through these two scanning methods, the CT image information and MR image information of the patient's pelvic region were obtained, and these image data were stored in the hospital's Picture Archiving and Communication System (PACS) in digital form, waiting for subsequent processing.

[0018] The high-risk clinical target volume and organ-at-risk delineation requirement information was determined based on the patient's specific condition, tumor stage, physical condition, and clinical treatment standards. The medical team would comprehensively analyze the patient's medical record data, including information such as the pathological diagnosis report, the size, location, and infiltration degree of the tumor. If the patient was in stage II of cervical cancer and the tumor invaded part of the parametrium but did not reach the pelvic wall, the high-risk clinical target volume (HR-CTV) included the cervix, uterine body, part of the vagina, and the invaded parametrium. The organs at risk mainly included the bladder, rectum, sigmoid colon, and small intestine, etc. The doctor would clarify the specific requirements for delineating the high-risk clinical target volume and organs at risk based on this information, such as the boundary range of the high-risk clinical target volume and the areas that needed to be protected for organs at risk. This information was recorded in the form of an electronic medical record and associated with the patient's image information, providing clinical guidance for subsequent automatic delineation.

[0019] In another implementation, obtaining an automatic cervical cancer contouring model includes: obtaining a preset automatic contouring model, a training sample set, and data training-related information. The preset automatic contouring model includes a localization network and a segmentation network. The training sample set includes CT image information and MR image information of cervical cancer patients with different stages and body types. The data training-related information is used to represent the contouring requirement information of the high-risk clinical target volume and organs at risk for cervical cancer patients with different stages and body types. First, a preset automatic contouring model framework is constructed. The localization network of this preset model adopts a 3D-CNN (three-dimensional convolutional neural network) architecture, which has strong image feature extraction capabilities. The model includes a localization network and a segmentation network. The localization network aims to accurately identify key information such as the applicator position from the patient's images, and the segmentation network is responsible for accurately dividing the high-risk clinical target volume and organs at risk based on the localization results and other relevant information.

[0020] The hospital utilized its rich clinical data resources to collect CT image information and MR image information of 300 cervical cancer patients with different stages and body types. These patients covered cervical cancer cases in the early, middle, and late stages, and had different body types, including thin, normal, and obese. For example, patient A was an early-stage cervical cancer patient with a thin body type; patient B was in the middle stage with a normal body type; and patient C was a late-stage patient with an obese body type. These diverse data could provide extensive learning materials for the model, enabling it to adapt to various clinical situations. According to clinical experience and medical standards, detailed contouring requirement information of the high-risk clinical target volume and organs at risk was formulated for each patient. For early-stage patients, the scope of the high-risk clinical target volume was relatively limited, possibly only including the cervix and a small surrounding area; for mid- and late-stage patients, the high-risk clinical target volume would expand as the tumor spread. Organs at risk such as the bladder and rectum also had different contouring requirements according to their relative positions to the tumor and the degree of possible radiation exposure. These information were organized into structured data tables and associated with the corresponding CT and MR image information one by one to form data training-related information.

[0021] Based on a preset classification ratio, the training sample set is processed to generate a training set and a validation set. The training sample set includes CT image information and MR image information of cervical cancer patients with different stages and body types. The training sample set is processed according to a preset classification ratio of 7:3. Randomly select 210 cases from the 300 patient data as the training set, and the remaining 90 cases as the validation set. The training set is used for training the model to let the model learn the relationship between image features and the high-risk clinical target volume and organs at risk; the validation set is used to evaluate the performance of the model on unseen data. During the division process, ensure that both the training set and the validation set contain patient data with different stages and body types to ensure the generalization ability of the model in different situations.

[0022] Data augmentation, random rotation, and flipping operations are performed on the training set to generate preprocessed CT image information and MR image information. To further enrich the data diversity of the training set and improve the robustness of the model, data augmentation operations are carried out on the training set. For each CT image and MR image in the training set, random rotation and flipping operations are performed. A CT image is randomly rotated by a certain angle (such as a random angle between 5° and 15°), or horizontally or vertically flipped. After such processing, the original 210 cases of training data are expanded in quantity, and the model can learn more image features at different angles and directions, enhancing its adaptability to various image changes.

[0023] Based on the training set and relevant information on data training, the preset automatic contouring model is processed to generate an automatic contouring prediction feature vector. Among them, the automatic contouring prediction feature vector includes the segmentation prediction target information for high-risk clinical target volumes and organs at risk. During the training process of the model, the data is processed layer by layer through a localization network and a segmentation network. The localization network first extracts features from CT and MR images to identify the potential features of the applicator, high-risk clinical target volumes, and organs at risk. The segmentation network then attempts to predict the segmentation results of high-risk clinical target volumes and organs at risk based on these features, combined with the contouring requirement information of high-risk clinical target volumes and organs at risk, to generate an automatic contouring prediction feature vector. This vector contains the segmentation prediction target information of the model for different regions, such as the preliminary judgment of the position, size, and shape of a certain tumor region. Based on the preprocessed CT image information, MR image information, and the automatic contouring prediction feature vector, the preset automatic contouring model is processed to generate a trained automatic contouring model. The model will continuously adjust its own parameters through the backpropagation algorithm according to the difference between the prediction result and the actual contouring requirement information of high-risk clinical target volumes and organs at risk to optimize the performance of the model. After multiple rounds of iterative training, the model gradually learns to more accurately identify and segment high-risk clinical target volumes and organs at risk, forming a trained automatic contouring model.

[0024] The trained automatic contouring model is processed based on the validation set to generate validation results. Based on the validation results, the trained automatic contouring model is further processed to generate an automatic cervical cancer contouring model. The validation set is input into the trained automatic contouring model, and the model processes the data of each patient in the validation set to generate corresponding automatic contouring results, which are the validation results. The predicted high-risk clinical target volume and organ-at-risk contours of the model are compared with the standard results manually contoured by doctors to evaluate indicators such as the accuracy, precision, and recall rate of the model on the validation set. According to the validation results, the trained automatic contouring model is evaluated and adjusted. If the performance of the model on the validation set meets the expected standards, such as the accuracy rate being higher than 90%, and the precision and recall rate also meeting the clinical requirements, then this model is determined as the final automatic cervical cancer contouring model. If the model performs poorly, the reasons are further analyzed, the training parameters are adjusted or the model structure is improved, and training and validation are carried out again until the model performance meets the requirements. The finally obtained automatic cervical cancer contouring model will be applied to actual clinical work to provide accurate high-risk clinical target volume and organ-at-risk contouring support for the treatment of cervical cancer patients. After multiple rounds of training and validation, the model is optimized according to the validation results, and finally an automatic cervical cancer contouring model is obtained. This model is stored in the hospital server, waiting to be called for automatic contouring processing of the image data of new patients.

[0025] S102. Process the CT image information and MR image information of a cervical cancer patient based on the positioning network to generate the 3D position information and morphological information of the applicator, as well as the image feature sequence information.

[0026] In one implementation, the region classifier of the positioning network and the multi-layer surface and pooling layers are used to perform feature extraction processing on the CT image information and MR image information of a cervical cancer patient to generate target features related to the applicator and the image feature sequence information. A cervical cancer patient is admitted to the department. Before radiotherapy, the patient undergoes CT and MR scans to obtain the CT image information and MR image information of the patient. The region classifier of the positioning network starts to work. It acts like an intelligent filter that can identify specific regions related to the treatment of cervical cancer in the CT and MR images, especially the information related to the possible regions of the applicator. For example, it will focus on the pelvic region where the applicator may appear in the image and initially distinguish the approximate range where the applicator is located.

[0027] Next, the multi-layer surface and pooling layers start to deeply process these preliminarily screened regions. The surface layer analyzes the spatial structure in the image. It can transform and extract complex shape features such as the bending and folding of the applicator in the image. The pooling layer is responsible for screening and integrating these features, removing some unimportant detailed information while retaining key features. It can highlight the important features of the applicator while reducing the data volume. For example, through the processing of the pooling layer, the key shape and position features of the applicator are highlighted, while the features of some subtle noises or irrelevant tissues are weakened. After this series of operations, the target feature and image feature sequence information related to the applicator are generated. These target features contain key information such as the material, shape, and position of the applicator, and the image feature sequence information records the feature changes related to the applicator at different levels and scales of the image, laying a foundation for accurately determining the position of the applicator subsequently.

[0028] The fully connected layer of the positioning network processes the target features related to the applicator to generate the position information of the applicator in the CT image and / or MR image. The fully connected layer of the positioning network receives the target features related to the applicator generated previously. The neurons of the fully connected layer are interconnected with all neurons of the previous layer, which enables it to make full use of the relationships between the target features. By performing complex calculations and analyses on these target features, the fully connected layer generates the position information of the applicator in the CT image and MR image.

[0029] Since CT images and MR images each have different advantages, CT images show high-density substances such as metals clearly, while MR images have stronger resolution ability for soft tissues. The positioning network will refer to the information of both CT images and MR images to determine the position of the applicator. For example, CT images can clearly show the general outline and position of the applicator, but the boundary between the applicator and the surrounding soft tissues may not be shown clearly enough; while although MR images show the soft tissue boundary well, the imaging of the applicator in them may be relatively blurred. The fully connected layer will integrate the advantages of these two types of images and accurately calculate the accurate position information of the applicator in the CT image and MR image through the analysis of the target features, providing accurate data support for subsequent 3D reconstruction.

[0030] Based on the positioning network, the position information of the applicator in CT images and / or MR images is segmented, outlined, and 3D reconstructed to generate the 3D position information and morphological information of the applicator. After obtaining the position information of the applicator in CT images and MR images based on the positioning network, the subsequent segmentation, outlining, and 3D reconstruction processes are carried out. It is clearly stated here that the operations are performed based on the position information in both CT images and MR images simultaneously. The positioning network accurately outlines the contour of the applicator in the images according to the position information of the applicator in CT images and MR images, and segments the applicator from the surrounding tissues and organs. For example, by comparing the high-density characteristics of the applicator in CT images with the differences in soft tissues in MR images, the boundary of the applicator can be determined more precisely.

[0031] Then, 3D reconstruction is performed using the results of these segmentations and outlines. The positioning network integrates the two-dimensional position information of the applicator in CT images and MR images and constructs a three-dimensional model of the applicator through specific algorithms. In this process, the position changes of the applicator at different image levels and its spatial relationship with the surrounding tissues are fully considered, and finally the 3D position information and morphological information of the applicator are generated. These 3D information can intuitively display the accurate position and shape of the applicator in the patient's body, and the placement position and dose distribution of the radiation source can be more accurately planned based on this information, thereby improving the radiotherapy effect and reducing damage to normal tissues.

[0032] S103, classify the CT image information of cervical cancer patients based on the positioning network to generate prior knowledge labels.

[0033] In one implementation, the CT images of cervical cancer patients are classified based on the positioning network to generate different anatomical structure information and tissue type information in the CT images. The convolutional layers in the positioning network are like a group of detectors with different "fields of view", which slide on the image to capture features at various scales. The convolutional layers with smaller convolutional kernels can capture fine features such as blood vessel textures, while the convolutional layers with larger convolutional kernels focus on macroscopic features such as organ contours. The pooling layer downsamples the feature maps output by the convolutional layer, reducing the data volume while retaining key features and improving processing efficiency. The fully connected layer integrates the feature vectors after convolution and pooling to determine which anatomical structure or tissue type each image region belongs to. After this series of complex operations, the positioning network can identify various anatomical structures and tissue types in the patient's CT images, such as distinguishing different structures like the uterus, bladder, rectum, cervical tumor, and bones. And these information are marked on the image in a specific coding manner to provide a basis for subsequent processing.

[0034] Obtain target features related to the applicator, feature information related to high-risk clinical targets and organs at risk, where the target features related to the applicator are used to characterize the location and morphological information of the applicator in the CT image, and the feature information related to high-risk clinical targets and organs at risk is used to characterize the characteristic attributes of the high-risk clinical targets and organs at risk. In the previous process of feature extraction and processing of CT images, the localization network has acquired a large amount of feature information. Among them, the target features related to the applicator include key information such as the material, shape, and position of the applicator. Since the applicator is made of metal, it will show unique high-density features on the CT image. The localization network can identify and extract these features to determine the location and approximate shape of the applicator in the image. For the feature information related to high-risk clinical targets and organs at risk, the localization network will focus on the tumor area and the surrounding organs that need to be protected. For high-risk clinical targets of tumors, their boundary features will be extracted, such as grayscale changes and texture features in the transition area between tumors and normal tissues; for organs at risk, such as the bladder, their shape, position, and contrast with surrounding tissues will be focused on. After these features are extracted, they are stored and represented in the form of vectors to facilitate subsequent integrated processing with anatomical structure and tissue type information.

[0035] Based on the target features related to the applicator, the feature information related to the high-risk clinical target area and the endangered organ, the different anatomical structure information and tissue type information in the CT image are processed to generate the classification results. After obtaining various feature information, the localization network will fuse the target features related to the applicator, the feature information related to the high-risk clinical target area and the endangered organ with the different anatomical structure information and tissue type information in the previously generated CT image. Specifically, the localization network will analyze the positional relationship between the applicator and the surrounding anatomical structure, determine whether the applicator is located near the high-risk clinical target area, and the distance to the endangered organ. For a specific area in the CT image, if it has the characteristics of a high-risk clinical target area of ​​the tumor and is close to the applicator, then this area will be given a higher weight in the classification results and may be classified as a key area closely related to radiotherapy; if a certain area belongs to an endangered organ and is close to the applicator, it will be marked as an area that needs special attention and protection. In this way, the localization network comprehensively evaluates and classifies each area in the CT image and generates a preliminary classification result, which roughly divides the importance and potential risks of different areas during radiotherapy.

[0036] Process the classification results to generate prior knowledge labels. The localization network will refine and optimize the classification results, removing some unreasonable classifications caused by image noise or feature extraction errors. Using the method of threshold processing, exclude some regions with feature intensities below a specific threshold from the key regions; or use a clustering algorithm to merge adjacent regions with similar features into a whole. After these processes, prior knowledge labels will be generated according to the finally determined classification results. These labels are marked on the CT images in a specific format, such as using different colors or numerical values to represent different region categories. Red represents high-risk regions, which may be the high-risk clinical target regions of tumors close to the applicator; blue represents the regions of organs at risk that need to be protected; green represents relatively safe normal tissue regions. These prior knowledge labels will provide important reference information for subsequent applicator localization, segmentation of high-risk clinical target regions and organs at risk, helping the segmentation network to more accurately identify and outline the boundaries of each region, and improving the accuracy and efficiency of radiotherapy plan formulation.

[0037] S104, process the prior knowledge labels based on the segmentation network to generate the applicator localization target factor.

[0038] In one implementation, feature extraction is performed on the prior knowledge labels, high-risk clinical target regions, and information on the delineation requirements of organs at risk based on the segmentation network. The relevant features of the applicator, high-risk clinical target regions, and organs at risk are extracted from different angles and scales through the waveform layer and the pooling layer respectively. Among them, the relevant features of the applicator include the position and morphological features of the applicator, and the material and density features of the applicator. The relevant features of the high-risk clinical target regions include the boundary and range features of the high-risk clinical target regions, and the tissue characteristic features of the high-risk clinical target regions. The relevant features of the organs at risk include the position and morphological features of the organs at risk, and the relative position features of the organs at risk and the high-risk clinical target regions. The waveform layer can be regarded as a set of special filters, which have different shapes and frequency responses and can "scan" the input prior knowledge labels, high-risk clinical target regions, and information on the delineation requirements of organs at risk from different angles.

[0039] For the extraction of the position and morphological features of the applicator, the waveform layer analyzes according to the imaging characteristics of the applicator in CT or MR images. The applicator presents specific shapes and position distributions in the images. The waveform layer performs specific convolution operations, just like matching with "templates" of different shapes on the images. If the applicator is slender, the convolution kernel similar in shape to it in the waveform layer will produce a strong response to it, thus capturing the shape information of the applicator. By adjusting parameters such as the size and stride of the convolution kernel, the applicator can be observed at different scales. A large convolution kernel can obtain the overall position and approximate morphology of the applicator, while a small convolution kernel can focus on the detailed features of the applicator, such as the surface texture, etc. For the material and density features of the applicator, since its material has unique gray-scale or signal intensity manifestations on the images, the waveform layer can extract these features through convolution operations sensitive to the changes in the gray-scale values of the images. The applicator made of metal appears as a high-density area in the CT image, and the waveform layer can detect the boundaries and internal gray-scale changes of this high-density area, thereby obtaining the features related to the material and density.

[0040] For the extraction of the boundary and scope features of the high-risk clinical target area, the waveform layer utilizes its sensitivity to the edges and textures of the images. There are differences in gray-scale or signal intensity between the high-risk clinical target area and the surrounding normal tissues. The convolution kernel of the waveform layer will produce large output values at these differences, thus outlining the boundary of the high-risk clinical target area. By using convolution kernels of different scales, the scope of the high-risk clinical target area can be determined at both the macroscopic and microscopic levels. A larger-scale convolution kernel can determine the approximate scope of the high-risk clinical target area, while a smaller-scale convolution kernel can refine the boundary and capture the subtle changes at the edge of the high-risk clinical target area. For the tissue characteristic features of the high-risk clinical target area, such as information on the density and metabolic activity of the tissue, the waveform layer can extract them by analyzing the gray-scale distribution patterns in different regions of the image and the contrast with the surrounding tissues. The metabolic activities of tumor tissues and normal tissues are different, which are manifested as different gray-scale or signal features on the images. The waveform layer can identify these feature differences, thereby obtaining the tissue characteristic information of the high-risk clinical target area.

[0041] When extracting the position and morphological features of organs at risk, the waveform layer operates based on the shape characteristics of the organs at risk and their position distribution in the image. The bladder presents a specific shape and position in the image. The waveform layer captures its contour information through a convolution kernel that matches the shape of the bladder, and at the same time uses convolution kernels of different scales to determine its size and position. For the relative position features of the organs at risk and the high-risk clinical target areas, the waveform layer comprehensively analyzes the position information of the organs at risk and the high-risk clinical target areas. By comparing their coordinate positions in the image and the relationships such as the distance and angle between them, the relative position features are extracted. By converting the position information of the organs at risk and the high-risk clinical target areas into vector representations, the waveform layer can extract feature vectors that reflect their relative position relationships through specific operations.

[0042] The pooling layer follows the waveform layer closely and further processes the features extracted by the waveform layer. The main role of the pooling layer is to reduce the data volume while retaining key features, improving the computational efficiency and the robustness of the model. For the features of the applicator, high-risk clinical target areas, and organs at risk, the pooling layer is implemented through operations such as max pooling or average pooling. Max pooling selects the maximum value within a local area as the output, which can highlight the most significant features. In the feature map of the applicator, max pooling can retain the most representative shape or position features of the applicator, ignoring some subtle changes, thus reducing the data volume while maintaining key information. Average pooling, on the other hand, calculates the average value within the local area as the output, which can smooth the features and reduce the influence of noise. Through the processing of the pooling layer, the relevant features of the applicator, high-risk clinical target areas, and organs at risk extracted from different angles and scales are further screened and integrated, preparing for subsequent feature fusion.

[0043] Feature fusion processing is performed on the applicator-related features, high-risk clinical target area-related features, and organ-at-risk-related features to generate an applicator localization target factor, where the applicator localization target factor is used to characterize the relationship between applicator localization and the delineation of high-risk clinical target areas and organs at risk. After the processing of the waveform layer and the pooling layer, the relevant features of the applicator, high-risk clinical target areas, and organs at risk are obtained. Next, these features need to be fused to generate the applicator localization target factor. The process of feature fusion is like piecing together different puzzle pieces to form a more complete and representative "image" to accurately characterize the relationship between applicator localization and the delineation of high-risk clinical target areas and organs at risk.

[0044] In this process, first, the features related to the applicator, the features related to the high-risk clinical target volume, and the features related to the organs at risk are combined. For example, the feature vectors corresponding to the position and morphological features, the material and density features of the applicator are arranged and combined with the boundary and extent features, the tissue property feature vectors of the high-risk clinical target volume, and the position and morphological features of the organs at risk, and the relative position feature vectors with respect to the high-risk clinical target volume. Then, it is processed through a specific fusion algorithm. The fusion method can adopt the weighted summation method, and different weights are assigned to each feature vector according to the importance of each feature in determining the applicator positioning and the delineation relationship between the high-risk clinical target volume and the organs at risk. For those features that have a greater impact on the applicator positioning and the delineation relationship, higher weights are assigned; while for those with a smaller impact, lower weights are given. For example, if in the current case, the boundary and extent features of the high-risk clinical target volume are crucial for determining the accurate positioning of the applicator, then the weight of this part of the feature vector will be relatively high during the fusion process.

[0045] Suppose the feature vector of the position and morphological features of the applicator is A, the feature vector of the material and density features is B, the feature vector of the boundary and extent features of the high-risk clinical target volume is C, the feature vector of the tissue property features is D, the feature vector of the position and morphological features of the organs at risk is E, and the feature vector of the relative position with respect to the high-risk clinical target volume is F. After weight assignment, assume the weight of A is ω1, the weight of B is ω2, the weight of C is ω3, the weight of D is ω4, the weight of E is ω5, and the weight of F is ω6. Then, the fused feature vector G can be expressed as: G = ω1×A + ω2×B + ω3×C + ω4×D + ω5×E + ω6×F.

[0046] In addition to weighted summation, a more complex neural network structure such as a multi-layer perceptron (MLP) can also be used for fusion. The MLP can learn more complex non-linear relationships between features and further improve the fusion effect. These combined feature vectors are input into the MLP, and after being processed by multiple layers of neurons, a new feature vector is output, and this vector is the applicator positioning target factor.

[0047] The applicator positioning target factor synthesizes multi-faceted information of the applicator, the high-risk clinical target volume, and the organs at risk. It can clearly reflect the positional relationship, spatial layout, and mutual influence between the applicator and the high-risk clinical target volume and the organs at risk. For example, through the applicator positioning target factor, it can be known whether the applicator is accurately located near the high-risk clinical target volume and what the safe distance is from the organs at risk, thus providing a key basis for accurately determining the position of the applicator and delineating the high-risk clinical target volume and the organs at risk subsequently, and helping doctors formulate more accurate and safer radiotherapy plans.

[0048] S105. Process the applicator positioning target factors, the 3D position information and morphological information of the applicator, and the image feature sequence information based on a segmentation network to generate high-risk clinical target volume and organ segmentation parameter information.

[0049] In one implementation, perform feature extraction and feature fusion processing on the applicator positioning target factors, the 3D position information and morphological information of the applicator, and the image feature sequence information based on a segmentation network to generate an automatic contouring prediction feature vector. The segmentation network is like an intelligent analysis system that receives these "clues" including the applicator positioning target factors, the 3D position information and morphological information of the applicator, and the image feature sequence information. The applicator positioning target factors contain the relationship information between the applicator and the high-risk clinical target volume and the organs at risk. The 3D position information and morphological information clarify the specific position and shape of the applicator in the patient's body. The image feature sequence information covers various feature details extracted from the images. The segmentation network performs feature extraction and feature fusion processing on this information.

[0050] During feature extraction, the convolutional layers in the network will act like "detectors" to carefully analyze the input information. For the material features of the applicator, the convolutional layers can capture the unique manifestations of the applicator material in the image through sensitive operations on the changes in image gray values, such as the gray value difference between the metal applicator and the surrounding tissues. For the high-risk clinical target volume and the organs at risk, the convolutional layers will identify the unique features of these regions based on their boundaries, textures, etc. Just like when identifying the boundary of the high-risk clinical target volume, the convolutional layers can keenly sense the changes in gray values or signal intensities between the high-risk clinical target volume and the normal tissues.

[0051] Then, feature fusion is carried out, which is like piecing together different puzzle pieces to form a more complete "puzzle". The segmentation network will integrate the features of the applicator, the high-risk clinical target volume, and the organs at risk that are extracted. For example, combine the position features of the applicator with the boundary features of the high-risk clinical target volume to judge the relative position relationship between the applicator and the high-risk clinical target volume; fuse the morphological features of the organ at risk with its relative position features with the high-risk clinical target volume to comprehensively understand the spatial layout of the organ at risk and the high-risk clinical target volume. Through this fusion, a new automatic contouring prediction feature vector is generated. This vector is different from the automatic contouring prediction feature vector generated during the model training stage. It is generated based on the specific images and relevant information of the current patient, is more targeted, and can reflect the tumor and organ characteristics of this patient individual.

[0052] The decoder based on the segmentation network processes the automatically contoured prediction feature vectors through upsampling and convolution to generate the segmentation parameter information of high-risk clinical target regions and organs. After obtaining the automatically contoured prediction feature vectors for the patient, the decoder of the segmentation network starts to work. The decoder can be regarded as a "restoration and refinement" tool, which processes the automatically contoured prediction feature vectors through upsampling and convolution. The upsampling operation is like enlarging a reduced picture. It expands the low-resolution information in the automatically contoured prediction feature vectors to a resolution close to that of the original image, restoring the size information of the image. In this process, some detail information may be lost, so convolution operations are needed to further refine it.

[0053] The convolution operation will deeply analyze and process the expanded features. It will scan these features again and, based on the previously learned knowledge, more precisely define the boundaries of high-risk clinical target regions and organs at risk, and make more accurate judgments about their shapes and positions. After multiple processes of upsampling and convolution, the segmentation parameter information of high-risk clinical target regions and organs is finally generated. These parameter information includes detailed information such as the positions, shapes, and sizes of high-risk clinical target regions and organs at risk, such as the boundary coordinates of high-risk clinical target regions and the contour descriptions of organs at risk. These parameter information are the key basis for subsequent automatic contouring. Doctors can accurately outline the contours of high-risk clinical target regions and organs at risk on the patient's images based on these parameter information, providing important support for formulating precise radiotherapy plans.

[0054] S106, process the segmentation parameter information of high-risk clinical target regions and organs to generate the automatic contouring results of high-risk clinical target regions and organs at risk for cervical cancer brachytherapy.

[0055] In one implementation, format conversion and visualization rendering processing are performed on the segmentation parameter information of high-risk clinical target regions and organs to generate tumor contour image information and organ-at-risk image information, where the tumor contour image information and organ-at-risk image information are highlighted with lines of different colors. After obtaining the segmentation parameter information of high-risk clinical target regions and organs, the system first performs format conversion. Since the segmentation parameter information initially exists in the form of a digital matrix recognizable by a computer and cannot be directly understood intuitively by doctors, it needs to be converted into a format suitable for image display. This is like translating a complex string of passwords into words that people can understand. For example, converting the digital information describing the positions, shapes, etc. of high-risk clinical target regions and organs at risk into data conforming to common image formats (such as DICOM format).

[0056] After the format conversion is completed, the system performs visual rendering. Based on this converted information, the system will draw the tumor contour image information and the organ-at-risk image information on the image. For easy distinction, the tumor contour is highlighted with red lines, while the organs-at-risk are highlighted with blue lines. In the case of this patient, when the system receives the segmentation parameter information of the high-risk clinical target volume, it will accurately outline the possible area of the tumor with red lines on the corresponding CT or MR image, clearly showing the general shape and scope of the tumor; for organs-at-risk such as the bladder and rectum, their contours are depicted with blue lines. In this way, doctors can immediately distinguish the positions of the tumor and the organs-at-risk on the image.

[0057] Based on the preset image post-processing algorithm and edge detection algorithm, the tumor contour image information and the organ-at-risk image information are processed to generate the automatic delineation results of the high-risk clinical target volume and the organs-at-risk for cervical cancer brachytherapy. The preset image post-processing algorithm is mainly used to optimize the image quality and improve the accuracy and reliability of the automatic delineation results. The median filtering algorithm can be adopted as the preset image post-processing algorithm. This algorithm processes each pixel point in the image to remove the noise interference in the image. In the image of this patient, due to reasons such as device acquisition or transmission, there may be some isolated noise points, which may affect the accurate judgment of the boundaries of the tumor and the organs-at-risk. The median filtering algorithm takes each pixel point as the center and selects a specific-sized neighborhood (such as a 3×3 or 5×5 pixel matrix), sorts the pixel values in the neighborhood, and takes the median value as the new value of this pixel point. In this way, the image can be effectively smoothed, the noise removed, and the contours of the tumor and the organs-at-risk become clearer.

[0058] The morphological processing algorithm, including erosion and dilation operations, can also be adopted. For the tumor contour image, the erosion operation can remove some small protrusions that may be caused by noise or image artifacts, making the contour smoother and more accurate. The dilation operation can expand some areas that may be narrowed or discontinuous due to noise and other reasons to ensure the complete representation of the tumor. When processing the tumor contour image of this patient, the erosion operation is first performed to remove those small protrusions that should not exist, and then the dilation operation is performed to make the tumor contour more continuous and complete. For the organ-at-risk image, a similar method is also adopted to ensure the accuracy of its contour.

[0059] Edge detection algorithms play a crucial role in determining the precise boundaries of tumors and organs at risk. The Canny edge detection algorithm is a commonly used edge detection method. It first performs Gaussian filtering on the image to further smooth the image and reduce the impact of noise on edge detection. Then, it calculates the gradient intensity and direction of each pixel point in the image. In the image of this patient, there are differences in gray scale or signal intensity between the tumor and normal tissues, and between the organs at risk and the surrounding tissues. These differences will appear as relatively large gradient values in the gradient calculation. The Canny algorithm will refine the edges through non-maximum suppression based on this gradient information, only retaining those pixel points with significant gradient changes as edge points. Finally, through double-threshold processing and hysteresis tracking, the final edges are determined. In this way, the boundaries of the tumor and organs at risk can be accurately found, providing an accurate basis for automatic contouring.

[0060] After being processed by the preset image post-processing algorithm and the edge detection algorithm, the system integrates this processed image information to generate the automatic contouring results of the high-risk clinical target volume and organs at risk for cervical cancer brachytherapy. In this result, the boundaries of the high-risk clinical target volume of the tumor and the organs at risk are more precise and clear, and doctors can directly formulate radiotherapy plans based on these results. The system will present the automatic contouring results to doctors in an intuitive form. For example, in the electronic medical record system, it shows the contours of different tissues in a layered form, and doctors can clearly see the positional relationship between the tumor and the surrounding organs at risk, so as to more accurately plan the radiotherapy dose and irradiation range, improve the radiotherapy effect, and at the same time minimize the damage to normal tissues.

[0061] The server uses the 3D U-Net network structure to divide the segmentation process into two parts: the localization network and the segmentation network. It obtains the CT and MR image information, clinical contouring requirement information, and automatic contouring model of cervical cancer patients. The localization network extracts features through a region classifier, multi-layer surfaces, and pooling layers, and after being processed by a fully connected layer, it realizes the 3D localization and reconstruction of the applicator, and generates an image feature sequence and a priori knowledge label.

[0062] Next, the segmentation network generates the applicator localization target factor based on the a priori knowledge label, and then combines the 3D position, shape, and image feature sequence information of the applicator. After being processed by feature extraction, fusion, and a decoder, it generates the segmentation parameter information of the high-risk clinical target volume and organs. Finally, it performs format conversion and visualization drawing on the segmentation parameter information, and with the help of the preset image post-processing algorithm and the edge detection algorithm, it generates the automatic contouring results of the high-risk clinical target volume and organs at risk for cervical cancer brachytherapy.

[0063] In terms of model training, a preset model, a training sample set, and relevant information are obtained. The training set and the validation set are divided. After enhancing the training set, the model is trained. Through multiple rounds of training and validation optimization, the final automatic cervical cancer delineation model is obtained. This solution realizes full automation from image input to segmentation results, reduces manual intervention, improves segmentation accuracy and efficiency, and has important clinical application value.

[0064] In one implementation, as Figure 2 shown, the present application also provides an automatic delineation device for high-risk clinical target areas and organs at risk in brachytherapy for cervical cancer based on deep learning, including:

[0065] An acquisition module 201, configured to acquire CT image information, MR image information, high-risk clinical target area and organ-at-risk delineation requirement information, and an automatic cervical cancer delineation model of a cervical cancer patient, wherein the automatic cervical cancer delineation model includes a localization network and a segmentation network;

[0066] A processing module 202, configured to process the CT image information and MR image information of the cervical cancer patient based on the localization network to generate 3D position information and morphological information of the applicator, and image feature sequence information; classify and process the CT image information of the cervical cancer patient based on the localization network to generate prior knowledge labels; process the prior knowledge labels based on the segmentation network to generate applicator localization target factors; process the applicator localization target factors, 3D position information and morphological information of the applicator, and image feature sequence information based on the segmentation network to generate high-risk clinical target area and organ segmentation parameter information; process the high-risk clinical target area and organ segmentation parameter information to generate an automatic delineation result of the high-risk clinical target area and organs at risk in brachytherapy for cervical cancer.

[0067] Each embodiment in the present application is described in a related manner. The same or similar parts among the embodiments can be referred to each other. The key point of each embodiment is to illustrate the differences from other embodiments. In particular, for the embodiments of the method, electronic device, electronic equipment, and readable storage medium for evaluating the automatic delineation of high-risk clinical target areas and organs at risk in brachytherapy for cervical cancer based on deep learning, since they are basically similar to the embodiment of the method for automatic delineation of high-risk clinical target areas and organs at risk in brachytherapy for cervical cancer based on deep learning described above, the description is relatively simple, and the relevant parts can refer to the partial description of the embodiment of the method for automatic delineation of high-risk clinical target areas and organs at risk in brachytherapy for cervical cancer based on deep learning described above.

Claims

1. A method for automatically delineating high-risk clinical target areas and organs at risk in cervical cancer brachytherapy based on deep learning, characterized in that: include: Obtaining CT image information and MR image information of cervical cancer patients, high-risk clinical target areas and risk organ delineation demand information, and an automatic delineation model for cervical cancer, wherein the automatic delineation model for cervical cancer includes a positioning network and a segmentation network; Based on the positioning network, the CT image information and MR image information of the cervical cancer patient are processed to generate the 3D position information and morphological information of the applicator, as well as the image feature sequence information; Based on the positioning network, the CT image information of cervical cancer patients is classified and processed to generate prior knowledge labels; The prior knowledge labels are processed based on the segmentation network to generate the applicator positioning target factor; Based on the segmentation network, the applicator positioning target factors, the applicator's 3D position information and morphological information, and the image feature sequence information are processed to generate high-risk clinical target areas and organ segmentation parameter information; The high-risk clinical target area and organ segmentation parameter information are processed to generate automatic delineation results of high-risk clinical target areas and organs at risk for cervical cancer brachytherapy.

2. The method according to claim 1, characterized in that Get the cervical cancer automatic delineation model, including: Obtaining a preset automatic delineation model, a training sample set, and data training related information, wherein the preset automatic delineation model includes a positioning network and a segmentation network, the training sample set includes CT image information and MR image information of cervical cancer patients of different stages and different body types, and the data training related information is used to characterize high-risk clinical target areas and risk organ delineation requirements of cervical cancer patients of different stages and different body types; The training sample set is processed based on a preset classification ratio to generate a training set and a validation set, wherein the training sample set includes CT image information and MR image information of cervical cancer patients of different stages and different body shapes; Perform data augmentation, random rotation and flipping on the training set to generate preprocessed CT image information and MR image information; Processing the preset automatic delineation model based on the training set and data training related information to generate an automatic delineation prediction feature vector, wherein the automatic delineation prediction feature vector includes segmentation prediction target information for high-risk clinical target areas and organs at risk; Processing the preset automatic delineation model based on the preprocessed CT image information and MR image information and the automatic delineation prediction feature vector to generate a trained automatic delineation model; Process the trained automatic delineation model based on the validation set to generate validation results; The trained automatic delineation model is processed based on the verification results to generate an automatic delineation model for cervical cancer.

3. The method according to claim 1, characterized in that Based on the positioning network, the CT image information and MR image information of cervical cancer patients are processed to generate 3D position information and morphological information of the applicator, as well as image feature sequence information, including: The regional classifier based on the localization network and the multi-layer surface and pooling layers perform feature extraction processing on the CT image information and MR image information of cervical cancer patients to generate target features and image feature sequence information related to the applicator; The target features related to the applicator are processed based on the fully connected layer of the positioning network to generate the position information of the applicator in the CT image and / or the MR image; Based on the positioning network, the position information of the applicator in the CT image and / or MR image is segmented, outlined and 3D reconstructed to generate 3D position information and morphological information of the applicator.

4. The method according to claim 3, characterized in that Based on the positioning network, the CT images of cervical cancer patients are classified and processed to generate prior knowledge labels, including: Based on the positioning network, the CT images of cervical cancer patients are classified and processed to generate different anatomical structure information and tissue type information in the CT images; Acquire target features related to the applicator, feature information related to high-risk clinical target areas and organs at risk, wherein the target features related to the applicator are used to characterize the positioning and morphological information of the applicator in the CT image, and the feature information related to the high-risk clinical target areas and organs at risk are used to characterize the feature attributes of the high-risk clinical target areas and organs at risk; Based on the target features related to the applicator, the feature information related to the high-risk clinical target area and the risk organ, the different anatomical structure information and tissue type information in the CT image are processed to generate a classification result; The classification results are processed to generate prior knowledge labels.

5. The method according to claim 4, characterized in that Based on the segmentation network, the prior knowledge labels and the high-risk clinical target area and risk organ delineation requirements are processed to generate the applicator positioning target factors, including: Based on the segmentation network, feature extraction is performed on the prior knowledge labels and the high-risk clinical target area and risk organ delineation demand information. The waveform layer and the pooling layer are used to extract the applicator-related features, the high-risk clinical target area-related features and the risk organ-related features from different angles and scales. Among them, the applicator-related features include the location and morphological features of the applicator, the material and density features of the applicator, the high-risk clinical target area-related features include the boundary and range features of the high-risk clinical target area, and the tissue characteristic features of the high-risk clinical target area. The risk organ-related features include the location and morphological features of the risk organ, and the relative position features of the risk organ and the high-risk clinical target area. The applicator-related features, high-risk clinical target-related features and risk-organ-related features are fused to generate the applicator positioning target factor, where the applicator positioning target factor is used to characterize the relationship between the applicator positioning and the high-risk clinical target and risk-organ delineation.

6. The method according to claim 5, characterized in that Based on the segmentation network, the applicator positioning target factors, the applicator's 3D position information and morphological information, and the image feature sequence information are processed to generate high-risk clinical target areas and organ segmentation parameter information, including: Based on the segmentation network, feature extraction and feature fusion processing are performed on the applicator positioning target factor, the 3D position information and morphological information of the applicator, and the image feature sequence information to generate an automatic delineation prediction feature vector; The decoder based on the segmentation network processes the automatic delineation prediction feature vector through upsampling and convolution to generate high-risk clinical target areas and organ segmentation parameter information.

7. The method according to claim 1, characterized in that Process the high-risk clinical target and organ segmentation parameter information to generate automatic delineation results of high-risk clinical target and organs at risk for cervical cancer brachytherapy, including; Perform format conversion and visualization processing on high-risk clinical target area and organ segmentation parameter information to generate tumor contour image information and organ-at-risk image information, where the tumor contour image information and organ-at-risk image information are highlighted with lines of different colors respectively; The image information of tumor contours and organs at risk are processed to generate automatic delineation results of high-risk clinical target areas and organs at risk for cervical cancer brachytherapy.

8. A deep learning-based automatic delineation device for high-risk clinical target areas and organs at risk in cervical cancer brachytherapy, characterized in that: The device comprises: An acquisition module is used to acquire CT image information and MR image information of cervical cancer patients, high-risk clinical target areas and risk organ delineation demand information, and an automatic delineation model for cervical cancer, wherein the automatic delineation model for cervical cancer includes a positioning network and a segmentation network; A processing module is used to process the CT image information and MR image information of cervical cancer patients based on the positioning network to generate 3D position information and morphological information of the applicator, as well as image feature sequence information; classify the CT image information of cervical cancer patients based on the positioning network to generate prior knowledge labels; process the prior knowledge labels based on the segmentation network to generate the applicator positioning target factor; process the applicator positioning target factor, the applicator's 3D position information and morphological information, and image feature sequence information based on the segmentation network to generate high-risk clinical target area and organ segmentation parameter information; process the high-risk clinical target area and organ segmentation parameter information to generate the high-risk clinical target area and the automatic delineation result of the endangered organ for close-range treatment of cervical cancer.

9. An electronic device, characterized in that: include: a first processor; and a memory for storing executable instructions of the first processor; Wherein, the first processor is configured to execute the method for automatically delineating high-risk clinical target areas and organs at risk in cervical cancer brachytherapy based on deep learning as described in any one of claims 1 to 7 by executing the executable instructions.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the second processor, it implements the method for automatically delineating high-risk clinical target areas and organs at risk in cervical cancer brachytherapy based on deep learning as described in any one of claims 1 to 7.

Images