A training method for a CTV segmentation model after endometrial cancer surgery

Abstract

The present invention relates to a training method for a CTV segmentation model after endometrial cancer surgery, belonging to the technical field of target segmentation, and solves the problem of low segmentation accuracy in the prior art. The method includes the following steps: constructing a plurality of data centers, each data center including a multi-modal segmentation model and a multi-modal CTV sample set after endometrial cancer surgery; pre-training the multi-modal segmentation model based on the plurality of data centers to obtain a pre-trained CTV segmentation model after endometrial cancer surgery; screening style features based on the multi-modal CTV sample sets of the plurality of data centers; for a target data center, fine-tuning the pre-trained CTV segmentation model after endometrial cancer surgery based on the center style features to obtain a trained CTV segmentation model after endometrial cancer surgery. The efficient and accurate CTV segmentation after endometrial cancer surgery is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of target delineation, and in particular to a method for training a CTV segmentation model after endometrial cancer surgery. Background Art

[0002] The incidence and mortality of endometrial cancer (EC) have been increasing year by year. Volumetric modulated arc therapy (VMAT) has become one of the best radiotherapy methods for treating EC after surgery. Using CT or MR images to delineate the clinical target volume (CTV) is very important for the accuracy of dose distribution and reducing the acute radiation reaction of patients. Accurate target segmentation can ensure that treatment means such as radiotherapy act precisely on the tumor area, reduce the damage to surrounding normal tissues, improve the quality of life of patients and improve the prognosis.

[0003] Existing deep neural network models applied to medical image segmentation, such as nnUNet, are suitable for the segmentation of most organs. However, for small sample datasets and clinical target volumes with no obvious boundaries and great difficulty, the model performance is usually poor. If there is a dataset from other centers for the same task, it can be pre-trained on other centers and then fine-tuned on the small sample dataset through transfer learning.

[0004] However, the styles of different data centers are different. The prior art does not consider the styles of different data centers when training the CTV segmentation model after endometrial cancer surgery, so the accuracy of the trained model is low. Summary of the Invention

[0005] In view of the above analysis, embodiments of the present invention aim to provide a method for training a CTV segmentation model after endometrial cancer surgery to solve the problem of low existing segmentation accuracy.

[0006] On the one hand, embodiments of the present invention provide a method for training a CTV segmentation model after endometrial cancer surgery, including the following steps:

[0007] Construct a plurality of data centers, each data center including a multimodal segmentation model and a multimodal CTV sample set after endometrial cancer surgery;

[0008] Pre-train the multimodal segmentation model based on the plurality of data centers to obtain a pre-trained CTV segmentation model after endometrial cancer surgery;

[0009] Screen style features based on the multimodal CTV sample sets of multiple data centers;

[0010] For the target data center, fine-tune the pre-trained CTV segmentation model after endometrial cancer surgery based on the center style features to obtain a trained CTV segmentation model after endometrial cancer surgery.

[0011] Based on the further improvement of the above method, the samples of the multi-modal CTV dataset after endometrial cancer surgery include paired plain CT images and enhanced CT images of endometrial cancer; some samples also include the corresponding mask images of the plain CT images and enhanced CT images.

[0012] The multi-modal CTV sample set after endometrial cancer surgery based on multiple data centers screens style features in the following way:

[0013] For the multi-modal CTV sample set of each data center, extract the samples with corresponding mask images, take the intersection of the plain CT image and the enhanced CT image of the sample with the corresponding mask image respectively to obtain the plain CTV inner image and the enhanced CTV inner image; the plain CTV inner image and the enhanced CTV inner image are spliced to obtain the CTV inner image of the sample.

[0014] Calculate various radiomics features of each CTV inner image.

[0015] For each data center, based on each radiomics feature of each CTV inner image in this data center, construct each radiomics feature matrix of this data center.

[0016] Calculate the style representativeness of each radiomics feature based on each radiomics feature matrix of each data center.

[0017] Select a preset number of radiomics features with larger style representativeness as style features.

[0018] Based on the further improvement of the above method, the following formula is used to calculate the style representativeness of each radiomics feature:

[0019] ;

[0020] Where, represents the style representativeness of the th radiomics feature, N represents the number of data centers, represents the KL divergence, represents the th radiomics feature matrix of the i-th data center, represents the th radiomics feature matrix of the j-th data center.

[0021] Based on the further improvement of the above method, for the target data center, fine-tune the pre-trained CTV segmentation model after endometrial cancer surgery based on the center style features, including:

[0022] Take the intersection of the non-contrast CT image and the contrast-enhanced CT image of the training samples in the target data center with the corresponding mask images respectively to obtain the non-contrast CTV internal image and the contrast-enhanced CTV internal image; splice the non-contrast CTV internal image and the contrast-enhanced CTV internal image to obtain the true CTV internal image of the sample;

[0023] Input the non-contrast CT image and the contrast-enhanced CT image of the training sample into the pre-trained CTV segmentation model for endometrial cancer after surgery to obtain the first segmentation image and the second segmentation image; take the intersection of the non-contrast CT image and the first segmentation image to obtain the predicted non-contrast CTV internal image, take the intersection of the contrast-enhanced CT image and the second segmentation image to obtain the predicted contrast-enhanced CTV internal image, and splice the predicted non-contrast CTV internal image and the predicted contrast-enhanced CTV internal image to obtain the predicted CTV internal image of the sample;

[0024] Calculate the loss based on the true CTV internal image and the predicted CTV internal image of the sample, update the parameters of the pre-trained CTV segmentation model for endometrial cancer after surgery, and fine-tune the pre-trained CTV segmentation model for endometrial cancer after surgery.

[0025] Based on a further improvement of the above method, the sample set includes a support set and a query set; pre-train the multi-modal segmentation model based on the multiple data centers to obtain a pre-trained CTV segmentation model for endometrial cancer after surgery, including:

[0026] S21. Randomly select N data centers, and record the parameters of the current multi-modal segmentation model as the initial parameters;

[0027] S22. For each selected data center, train the local multi-modal segmentation model based on the local support set and update the parameters of the local multi-modal segmentation model;

[0028] S23. For each selected data center, calculate the loss on the local query set based on the updated local multi-modal segmentation model;

[0029] S24. Update the initial parameters based on the losses of each selected data center; update the parameters of the multi-modal segmentation model of each data center to the initial parameters;

[0030] S25. Determine whether the stop condition is reached. If so, stop the training to obtain the pre-trained CTV segmentation model for endometrial cancer after surgery; otherwise, return to step S21.

[0031] Based on a further improvement of the above method, use the following formula to update the initial parameters based on the losses of each selected data center:

[0032] ;

[0033] ;

[0034] Among them, represents the total loss of N selected data centers, represents the total loss of N selected data centers of the gradient, represents the loss of the i-th selected data center, represents the weight of the i-th selected data center, β represents the learning rate, represents the initial parameter.

[0035] Based on the further improvement of the above method, the following formula is used to calculate the loss:

[0036] ;

[0037] Among them, represents the multi-modal image segmentation loss, represents the multi-modal cross loss, represents the contrast loss.

[0038] Based on the further improvement of the above method, the following formula is used to calculate the multi-modal cross loss:

[0039] ;

[0040] Among them, represents the number of samples in the query set of the i-th data center that do not have corresponding mask images, represents the element value at the position in the first segmentation image of the k-th sample without a corresponding mask image, represents the element value at the position in the second segmentation image of the k-th sample without a corresponding mask image, m represents the number of elements in the first segmentation image.

[0041] Based on the further improvement of the above method, the contrast loss is calculated in the following manner:

[0042] For each sample in the query set that does not have a corresponding mask image, the first segmentation image and the second segmentation image of this sample are used as a positive sample pair; the first segmentation image of this sample and the second segmentation image of other samples in the query set that do not have a corresponding mask image are used as negative sample pairs; positive sample pairs and negative sample pairs corresponding to each sample in the query set that does not have a corresponding mask image are obtained;

[0043] The contrast loss is calculated based on the positive sample pairs and negative sample pairs corresponding to each sample in the query set that does not have a corresponding mask image.

[0044] Based on the further improvement of the above method, for each sample in the query set that has no corresponding mask image, the contrastive loss is calculated in the following manner for the positive sample pairs and negative sample pairs corresponding to the sample:

[0045] ;

[0046] Among them, represents the number of samples in the query set of the i-th data center that have no corresponding mask image, represents the first segmentation image of the k-th sample that has no corresponding mask image, represents the second segmentation image of the k-th sample that has no corresponding mask image, represents the second segmentation image of the j-th sample that has no corresponding mask image, τ represents the adjustment parameter, and dice(·,·) represents the DICE loss.

[0047] Compared with the prior art, the present invention constructs multiple data centers, pre-trains a multi-modal segmentation model based on the multiple data centers to obtain a pre-trained CTV segmentation model for endometrial cancer after surgery, screens style features through a multi-modal endometrial cancer post-operative CTV sample set based on the multiple data centers, and thus fine-tunes the pre-trained CTV segmentation model on the target data center based on the features that best represent different data centers, enabling the model to better adapt to the style of the target data center and improving the segmentation accuracy of the model on the target data center. Moreover, multiple data centers jointly train the segmentation model instead of mixing the data for training, so it does not require the sample distributions of the data centers to be consistent. By obtaining a pre-trained CTV segmentation model for endometrial cancer after surgery and then fine-tuning it on the target data set, a CTV segmentation model applicable to the target data set can be quickly trained, and the performance of the model will not be affected by the small amount of data in the target data set, thereby improving the segmentation performance of the model on the target data center. For the target data center, based on the trained CTV segmentation model for endometrial cancer after surgery, the multi-modal endometrial cancer image to be segmented is segmented, and the CTV segmentation result of the multi-modal endometrial cancer image to be segmented can be quickly and accurately obtained, improving the efficiency of CTV segmentation.

[0048] In the present invention, the above technical solutions can also be combined with each other to achieve more preferred combination schemes. Other features and advantages of the present invention will be described in the subsequent specification, and some advantages will be obvious from the specification or can be understood by implementing the present invention. The objectives and other advantages of the present invention can be realized and obtained from the content specifically pointed out in the specification and the drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] The accompanying drawings are only for the purpose of showing specific embodiments and are not considered to be a limitation of the present invention. Throughout the drawings, the same reference numerals represent the same components;

[0050] Figure 1 This is a flowchart of the training method for the CTV segmentation model after endometrial cancer surgery in an embodiment of the present invention. Specific embodiments

[0051] The following will specifically describe the preferred embodiments of the present invention in conjunction with the accompanying drawings. The accompanying drawings form a part of this application and are used together with the embodiments of the present invention to explain the principles of the present invention, rather than to limit the scope of the present invention.

[0052] A specific embodiment of the present invention discloses a training method for a CTV segmentation model after endometrial cancer surgery, as Figure 1 shown, including the following steps:

[0053] S1. Construct multiple data centers, each data center including a multi-modal segmentation model and a multi-modal CTV sample set after endometrial cancer surgery;

[0054] S2. Pre-train the multi-modal segmentation model based on the multiple data centers to obtain a pre-trained CTV segmentation model for the target area after endometrial cancer surgery;

[0055] S3. Screen style features based on the multi-modal CTV sample sets of multiple data centers;

[0056] S4. For the target data center, fine-tune the pre-trained CTV segmentation model after endometrial cancer surgery based on the center style features to obtain a trained CTV segmentation model after endometrial cancer surgery.

[0057] During implementation, each medical center collects multi-modal image data of endometrial cancer patients after surgery in the center to construct a multi-modal CTV sample set after endometrial cancer surgery. At the same time, each center is equipped with a multi-modal segmentation model with the same structure and initial parameters, thereby constructing a data center, that is, each data center has a local multi-modal segmentation model and a local sample set.

[0058] During implementation, the target data center is a dataset of a new data center. The data in the target dataset center is a small sample dataset. It is easy to overfit the model when training a small sample dataset, and the performance of the trained model is poor.

[0059] Compared with the prior art, the training method of the CTV segmentation model after endometrial cancer surgery provided in this embodiment constructs multiple data centers, pre-trains a multi-modal segmentation model based on the multiple data centers to obtain a pre-trained CTV segmentation model after endometrial cancer surgery, and screens style features based on the multi-modal CTV sample set after endometrial cancer surgery of the multiple data centers. Then, based on the features that best represent different data centers, the pre-trained CTV segmentation model is fine-tuned on the target data center, making the model more adaptable to the style of the target data center and improving the segmentation accuracy of the model on the target data center. Moreover, multiple data centers jointly train the segmentation model instead of mixing the data for training, so it does not require the sample distributions of each data center to be the same. By obtaining a pre-trained CTV segmentation model after endometrial cancer surgery and then fine-tuning it on the target data set, a CTV segmentation model suitable for the target data set can be quickly trained, and the performance of the model will not be affected by the small amount of data in the target data set, thus improving the segmentation performance of the model on the target data center. For the target data center, based on the trained CTV segmentation model after endometrial cancer surgery, the multi-modal endometrial cancer image to be segmented is segmented, and the CTV segmentation result of the multi-modal endometrial cancer image to be segmented can be quickly and accurately obtained, improving the efficiency of CTV segmentation.

[0060] Specifically, the postoperative CTV sample set of each data center includes a support set and a query set, that is, a part of the samples in the sample set are divided into the support set and a part of the samples are divided into the query set.

[0061] Specifically, pre-training the multi-modal segmentation model based on the multiple data centers to obtain a pre-trained CTV segmentation model after endometrial cancer surgery includes:

[0062] S21. Randomly select N data centers, and record the parameters of the current multi-modal segmentation model as the initial parameters;

[0063] S22. For each selected data center, train the local multi-modal segmentation model based on the local support set and update the parameters of the local multi-modal segmentation model;

[0064] S23. For each selected data center, calculate the loss on the local query set based on the updated local multi-modal segmentation model;

[0065] S24. Update the initial parameters based on the losses of each selected data center; update the parameters of the multi-modal segmentation model of each data center to the initial parameters;

[0066] S25. Determine whether the stop condition is reached. If so, stop training to obtain a pre-trained CTV segmentation model after endometrial cancer surgery. Otherwise, return to step S21.

[0067] During implementation, first randomly select N data centers for model training in the current round. The parameters of the current multi-modal segmentation model of each data center are the same, denoted as the initial parameter ф.

[0068] For each selected data center, first train the local multi-modal segmentation model based on the support set in the local dataset to update the parameters of the local multi-modal segmentation model. During implementation, multi-step training can be performed, that is, the parameters of the local multi-modal segmentation model are updated multiple times, and the parameters of the last update are denoted as , which represents the parameters of the local multi-modal segmentation model of the i-th data center.

[0069] After updating the parameters of the local multi-modal segmentation model, each selected data center calculates the loss on the local query set based on the updated local multi-modal segmentation model.

[0070] The samples of the multi-modal CTV dataset after endometrial cancer surgery include paired plain CT images and enhanced CT images of endometrial cancer, and some samples also include mask images corresponding to the plain CT images and enhanced CT images.

[0071] The multi-modal segmentation model includes a first segmentation module and a second segmentation module; the first segmentation module and the second segmentation module have the same structure.

[0072] The first segmentation module is used to perform target segmentation on the plain CT image to obtain a first segmentation image.

[0073] The second segmentation module is used to perform target segmentation on the enhanced CT image to obtain a first segmentation image.

[0074] During implementation, due to the large workload and low efficiency of manual delineation, some samples in the sample set are allowed to have no annotation data, that is, no corresponding mask image. Therefore, the samples of the multi-modal endometrial cancer sample set include paired plain CT images and enhanced CT images of endometrial cancer, and mask images corresponding to some paired plain CT images and enhanced CT images of endometrial cancer.

[0075] During implementation, the first segmentation model and the second segmentation model can adopt existing segmentation image segmentation models, such as the Unet model.

[0076] During implementation, for the i-th data center, input the samples on the local query set into the local multi-modal segmentation model (the model parameters are ), and calculate the loss according to the first segmentation image and the second segmentation image output by the model.

[0077] Specifically, for the i-th data center, calculate the loss based on the updated local multi-modal segmentation model on the local query set. Specifically, use the following formula to calculate the loss:

[0078] ;

[0079] Among them, represents the multi-modal image segmentation loss, represents the multi-modal cross loss, represents the contrast loss.

[0080] Specifically, for the labeled samples, that is, the samples with corresponding mask images, calculate the multi-modal segmentation loss based on the first segmentation image and the second segmentation image obtained by the first segmentation module and the second segmentation module.

[0081] Specifically, use the following formula to calculate the multi-modal image segmentation loss:

[0082] ;

[0083] Among them, represents the number of samples with corresponding mask images in the query set of the i-th data center, represents the first segmentation image of the k-th sample with a corresponding mask image in the query set, represents the mask image corresponding to the plain CT image of the k-th sample with a corresponding mask image in the query set, represents the second segmentation image of the k-th sample with a corresponding mask image in the query set, represents the mask image corresponding to the enhanced CT image of the k-th sample with a corresponding mask image in the query set, and represent the weight parameters.

[0084] During implementation, and are used to control false negatives and false positives respectively, and adjusting and can control the trade-off between false negatives and false positives. represents and 's intersection.

[0085] For the samples without labels, that is, the samples without corresponding mask images, calculate the loss through multi-modal mutual learning and contrast learning to obtain the multi-modal cross loss and the contrast loss.

[0086] Specifically, use the following formula to calculate the multi-modal cross loss:

[0087] ;

[0088] Among them, represents the number of samples in the query set of the i-th data center that do not have corresponding mask images, represents the element value at the position in the first segmentation image of the k-th sample without a corresponding mask image, represents the element value at the position in the second segmentation image of the k-th sample without a corresponding mask image, and m represents the number of elements in the first segmentation image.

[0089] Through the multi-modal cross-loss, train the images segmented by the first segmentation module and the second segmentation module to be as similar as possible.

[0090] Specifically, the contrast loss is calculated in the following way:

[0091] For each sample in the query set without a corresponding mask image, use the first segmentation image and the second segmentation image of this sample as the positive sample pair; use the first segmentation image of this sample and the second segmentation images of other samples in the query set without corresponding mask images as the negative sample pairs; obtain the positive sample pairs and negative sample pairs corresponding to each sample in the query set without a corresponding mask image;

[0092] Calculate the contrast loss based on the positive sample pairs and negative sample pairs corresponding to each sample in the query set without a corresponding mask image.

[0093] For example, the number of samples in the query set of the i-th data center without corresponding mask images is .

[0094] For the k-th sample without a corresponding mask image, take the first segmentation image predicted by the model and the second segmentation image as the positive sample pair corresponding to this sample . Then, form the negative sample pair corresponding to this sample by using the first segmentation image and the second segmentation images of other samples in the query set without corresponding mask images, .

[0095] Then, calculate the contrast loss based on the positive sample pairs and negative sample pairs corresponding to each sample in the query set without a corresponding mask image in the following way:

[0096] ;

[0097] Among them, represents the number of samples in the query set of the i-th data center that do not have corresponding mask images, represents the first segmentation image of the k-th sample without a corresponding mask image, The second segmentation image representing the k-th sample without a corresponding mask image The second segmentation image representing the j-th sample without a corresponding mask image, τ represents a tuning parameter, and dice(·,·) represents the DICE loss.

[0098] The contrastive loss makes the distance between positive samples closer and the distance between negative samples farther, thereby deepening the mutual learning between different modalities and improving the segmentation performance of the model.

[0099] After calculating the loss in N data, update the initial parameters based on the loss of the N data centers.

[0100] During implementation, since the sample sizes and data qualities of different data centers are different, the weights of the losses of the N data centers for updating the initial parameters are different.

[0101] Specifically, the following method is used to obtain the weight of each selected data center:

[0102] S241. Pre-update the parameters of the multi-modal segmentation model of the N selected data centers based on the losses of the N selected data centers;

[0103] S242. For each selected data center, calculate the marginal loss on the local query set based on the pre-updated local multi-modal segmentation model;

[0104] S243. Calculate the weight of each selected data center based on the marginal loss of each selected data center.

[0105] During implementation, pre-updating the parameters of the multi-modal segmentation model of the N selected data centers based on the losses of the N selected data centers includes:

[0106] Calculate the first pre-update parameter based on the losses of the N selected data centers;

[0107] For the i-th selected data center, the second pre-update parameter of the i-th selected data center obtained based on the losses of the other N - 1 selected data centers;

[0108] For the i-th selected data center, pre-update the local multi-modal segmentation model based on the first pre-update parameter and the second pre-update parameter respectively to obtain the first pre-update model and the second pre-update model.

[0109] Specifically, the first pre-update parameter is calculated based on the losses of the N selected data centers using the following formula:

[0110] ;

[0111] ;

[0112] Among them, represents the total loss of N selected data centers, represents the total loss gradient of, represents the loss of the i-th selected data center, β represents the learning rate, represents the initial parameter, represents the first pre-updated parameter.

[0113] Specifically, the second pre-updated parameter of the i-th selected data center obtained by the following formula based on the losses of the other N - 1 selected data centers:

[0114] ; ;

[0115] Among them, represents the total loss of N - 1 selected data centers, represents the total loss gradient of, represents the loss of the i-th selected data center, β represents the learning rate, represents the initial parameter, represents the second pre-updated parameter of the i-th selected data center.

[0116] During implementation, update the parameters of the local multi-modal segmentation model of the i-th selected data center to the first pre-updated parameter to obtain the first pre-updated model (parameters are ), update the parameters of the local multi-modal segmentation model of the i-th selected data center to the second pre-updated parameter to obtain the second pre-updated model (parameters are ).

[0117] Specifically, for the i-th selected data center, calculate the marginal loss based on the pre-updated local multi-modal segmentation model on the local query set, including:

[0118] Calculate the first loss by calculating the loss based on the first pre-updated model on the local query set;

[0119] Calculate the second loss by calculating the loss based on the second pre-updated model on the local query set;

[0120] The difference between the second loss and the first loss gives the marginal loss.

[0121] During implementation, both the first loss and the second loss can be calculated according to the formula of Equation (1).

[0122] Specifically, the weight of each selected data center is calculated using the following formula based on the marginal loss of each selected data center:

[0123] ;

[0124] ;

[0125] wherein, represents the data quality of the i-th selected data center, represents the marginal loss of the i-th selected data center, and γ represents the adjustment parameter.

[0126] After obtaining the weights, the initial parameters are updated using the following formula based on the losses of the N data centers:

[0127] ; ;

[0128] wherein, represents the total loss of the N selected data centers, represents the total loss of the N selected data centers of the gradient, represents the loss of the i-th selected data center, represents the weight of the i-th selected data center, β represents the learning rate, represents the initial parameter.

[0129] It should be noted that, represents the loss of the i-th selected data center, which is the loss calculated in step S23, that is, the loss calculated when the parameter is used.

[0130] By calculating the weights according to the data quality of different data centers and dynamically calculating the model parameters during each backpropagation, the weights of the data centers with high quality are improved, thereby improving the accuracy of model segmentation and the performance of the model.

[0131] After updating the initial parameters, the parameters of the local multi-modal segmentation models of all data centers are updated to the initial parameters, thereby completing one backpropagation of the multi-modal segmentation model.

[0132] After completing one backpropagation of the multi-modal segmentation model, if the current stopping condition is reached, the training is ended, and a pre-trained CTV segmentation model for endometrial cancer after surgery is obtained. If not, return to step S21 to continue training until the stopping condition of model training is reached. In implementation, the stopping condition can be reaching a preset number of training times or the total loss reaching a preset accuracy.

[0133] Due to the different image styles of different data centers, by screening the features that can represent the styles of different data centers, the fine-tuning of the pre-trained model can be more precisely guided, making the obtained model more adaptable to the new data center. Therefore, style features are screened based on the multi-modal CTV sample sets of endometrial cancer after surgery from multiple data centers.

[0134] Specifically, the following method is used to screen style features based on the multi-modal CTV sample sets of endometrial cancer after surgery from multiple data centers:

[0135] S31. For the multi-modal CTV sample set of each data center, extract the samples with corresponding mask images, take the intersection of the plain CT image and the enhanced CT image of the sample with the corresponding mask image respectively to obtain the plain CTV internal image and the enhanced CTV internal image; the plain CTV internal image and the enhanced CTV internal image are spliced to obtain the CTV internal image of the sample;

[0136] S32. Calculate various radiomics features of each CTV internal image;

[0137] S33. For each data center, based on each radiomics feature of each CTV internal image in this data center, construct each radiomics feature matrix of this data center;

[0138] S34. Calculate the style representativeness of each radiomics feature based on each radiomics feature matrix of each data center;

[0139] S35. Select a preset number of radiomics features with larger style representativeness as style features.

[0140] During implementation, for each data center, first extract the samples with corresponding mask images, take the intersection of the plain CT image and the enhanced CT image of the sample with the corresponding mask image respectively to obtain the plain CTV internal image and the enhanced CTV internal image; the plain CTV internal image and the enhanced CTV internal image are spliced to obtain the CTV internal image of the sample.

[0141] Then calculate various radiomics features of each CTV internal image. During implementation, various radiomics features include the energy, entropy, maximum value, minimum value, mean value, gray level co-occurrence matrix, gray level run length matrix, etc. of the image, which are not limited in this invention. The pyradiomics algorithm can be used to calculate various radiomics features during implementation.

[0142] For each data center, based on each radiomics feature of each CTV internal image in this data center, construct each radiomics feature matrix of this data center.

[0143] For example, for the i-th data center, for the A radiomics feature, as a row element of the th radiomics feature matrix of this data center, constructs the th radiomics feature matrix of this data center.

[0144] Then, based on each radiomics feature matrix of each data center, calculate the style representativeness of each radiomics feature.

[0145] Specifically, use the following formula to calculate the style representativeness of each radiomics feature:

[0146] ;

[0147] where, represents the style representativeness of the th radiomics feature, N represents the number of data centers, represents the KL divergence, represents the th radiomics feature matrix of the i-th data center, represents the th radiomics feature matrix of the j-th data center.

[0148] During implementation, calculate the distribution difference of the th radiomics feature on any two data centers through the KL divergence. The greater the distribution difference of the feature on different data centers, the more it can represent the styles of different data centers.

[0149] Sort according to the style representativeness from large to small, and select a preset number of radiomics features with larger style representativeness as style features. During implementation, the preset number is set according to the calculation efficiency requirements.

[0150] For a new medical center, a small amount of multi-modal CTV samples after endometrial cancer surgery can be collected to construct a training sample set for micro-calling. Based on the center style features, fine-tune the pre-trained CTV segmentation model after endometrial cancer surgery, and then a trained CTV segmentation model after endometrial cancer surgery can be obtained quickly. Input the multi-modal image after multi-modal endometrial cancer surgery to be segmented on this center into the trained CTV segmentation model after endometrial cancer surgery, and the target area image can be obtained quickly and accurately.

[0151] For the target data center, fine-tuning the pre-trained CTV segmentation model after endometrial cancer surgery based on the center style features includes:

[0152] S41. Take the intersection of the plain CT image and the contrast-enhanced CT image of the training samples in the target data center with the corresponding mask images respectively to obtain the plain CTV internal image and the contrast-enhanced CTV internal image; splice the plain CTV internal image and the contrast-enhanced CTV internal image to obtain the true CTV internal image of the sample.

[0153] S42. Input the plain CT image and the contrast-enhanced CT image of the training sample into the pre-trained CTV segmentation model for endometrial cancer after surgery to obtain the first segmentation image and the second segmentation image; take the intersection of the plain CT image and the first segmentation image to obtain the predicted plain CTV internal image, take the intersection of the contrast-enhanced CT image and the second segmentation image to obtain the predicted contrast-enhanced CTV internal image, and splice the predicted plain CTV internal image and the predicted contrast-enhanced CTV internal image to obtain the predicted CTV internal image of the sample.

[0154] S43. Calculate the loss based on the true CTV internal image and the predicted CTV internal image of the sample, update the parameters of the pre-trained CTV segmentation model for endometrial cancer after surgery, and fine-tune the pre-trained CTV segmentation model for endometrial cancer after surgery.

[0155] During implementation, obtain the true CTV image of the training samples in the target data center in the manner of the aforementioned step S31.

[0156] For each sample, obtain the first segmentation image and the second segmentation image through the pre-trained CTV segmentation model for endometrial cancer after surgery, that is, according to the target region mask images corresponding to the plain CT image and the contrast-enhanced CT image respectively. In the same way, obtain the predicted CTV internal image of the sample.

[0157] Calculate the loss based on the true CTV internal image and the predicted CTV internal image of the sample, update the parameters of the pre-trained CTV segmentation model for endometrial cancer after surgery, and fine-tune the pre-trained CTV segmentation model for endometrial cancer after surgery. During implementation, existing loss functions can be used, such as the cross-entropy loss function.

[0158] During implementation, the model parameters can be updated once according to the loss of one training sample, or the parameters can be updated once according to the loss of a batch of training samples.

[0159] Stop training after reaching the stop condition for model training to obtain the trained CTV segmentation model for endometrial cancer after surgery. During implementation, the stop condition can be reaching the preset number of training times, or the loss reaching the preset accuracy.

[0160] Through fine-tuning based on radiomics features, it better adapts to the style of the new data center and improves the accuracy of the CTV segmentation model for endometrial cancer after surgery on the new data center.

[0161] Those skilled in the art can understand that all or part of the processes of implementing the methods in the above embodiments can be completed by instructing relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium. Among them, the computer-readable storage medium is a disk, an optical disc, a read-only memory or a random access memory, etc.

[0162] As mentioned above, the above are only the preferred specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention.

Claims

1. A training method for a postoperative CTV segmentation model for endometrial cancer, characterized in that: The following steps are involved: Build multiple data centers, each of which includes a multimodal segmentation model and a multimodal endometrial cancer postoperative CTV sample set; Pre-training the multimodal segmentation model based on the multiple data centers to obtain a pre-trained endometrial cancer postoperative CTV segmentation model; Screening style features of multimodal postoperative CTV sample sets for endometrial cancer based on multiple data centers; For the target data center, fine-tuning the pre-trained endometrial cancer postoperative CTV segmentation model based on the center style feature to obtain a trained endometrial cancer postoperative CTV segmentation model; The sample set includes a support set and a query set; the multimodal segmentation model is pre-trained based on the multiple data centers to obtain a pre-trained endometrial cancer postoperative CTV segmentation model, including: S21. Randomly select N data centers and record the parameters of the current multimodal segmentation model as initial parameters; S22. For each selected data center, training a local multimodal segmentation model based on a local support set, and updating parameters of the local multimodal segmentation model; S23. For each selected data center, calculate the loss on the local query set based on the updated local multimodal segmentation model; S24, updating the initial parameters based on the loss and weight of each selected data center; updating the parameters of the multimodal segmentation model of each data center to the initial parameters; S25, determining whether the stop condition is met, if so, stopping the training to obtain the pre-trained postoperative CTV segmentation model for endometrial cancer, otherwise, returning to step S21; The weight of each selected data center is obtained in the following way: Pre-updating parameters of the multimodal segmentation model of the N selected data centers based on the losses of the N selected data centers; For each selected data center, the marginal loss is calculated on the local query set based on the pre-updated local multimodal segmentation model; The weight of each selected data center is calculated based on its marginal loss.

2. The training method for the postoperative CTV segmentation model for endometrial cancer according to claim 1, characterized in that: The samples of the multimodal postoperative CTV dataset for endometrial cancer include paired plain scan CT images and enhanced CT images of endometrial cancer; some samples also include mask images corresponding to the plain scan CT images and enhanced CT images; The multimodal postoperative CTV sample set for endometrial cancer based on multiple data centers uses the following method to screen style features: For each data center's multimodal postoperative CTV sample set for endometrial cancer, samples with corresponding mask images are extracted, and the plain scan CT image and enhanced CT image of the sample are intersected with the corresponding mask image to obtain the plain scan CTV internal image and the enhanced CTV internal image; the plain scan CTV internal image and the enhanced CTV internal image are spliced ​​to obtain the sample's CTV internal image; Calculate multiple radiomic features of images within each CTV; For each data center, a matrix of each radiomics feature of each CTV in the data center is constructed based on each radiomics feature of the image in the CTV of the data center. The style representativeness of each radiomics feature is calculated based on each radiomics feature matrix of each data center; A preset number of radiomics features with greater style representativeness are selected as style features.

3. The training method for the postoperative CTV segmentation model for endometrial cancer according to claim 2, characterized in that: The style representativeness of each radiomics feature is calculated using the following formula: ； in, Indicates The style representativeness of the radiomics features, N represents the number of data centers, represents the KL divergence, represents the first A radiomics feature matrix, represents the jth data center A radiomics feature matrix.

4. The training method for the postoperative CTV segmentation model for endometrial cancer according to claim 1, characterized in that: For the target data center, the pre-trained endometrial cancer postoperative CTV segmentation model is fine-tuned based on the central style features, including: The plain scan CT image and enhanced CT image of the training sample in the target data center are respectively intersected with the corresponding mask image to obtain the plain scan CTV image and the enhanced CTV image; the plain scan CTV image and the enhanced CTV image are spliced ​​to obtain the real CTV image of the sample; The plain scan CT image and the enhanced CT image of the training sample are input into the pre-trained endometrial cancer postoperative CTV segmentation model to obtain a first segmented image and a second segmented image; the plain scan CT image and the first segmented image are intersected to obtain a predicted plain scan CTV internal image, the enhanced CT image and the second segmented image are intersected to obtain a predicted enhanced CTV internal image, and the predicted plain scan CTV internal image and the predicted enhanced CTV internal image are spliced ​​to obtain a predicted CTV internal image of the sample; The loss is calculated based on the real CTV image and the predicted CTV image of the sample, the parameters of the pre-trained endometrial cancer postoperative CTV segmentation model are updated, and the pre-trained endometrial cancer postoperative CTV segmentation model is fine-tuned.

5. The training method for the postoperative CTV segmentation model for endometrial cancer according to claim 1, characterized in that: The initial parameters are updated based on the loss and weight of each selected data center using the following formula: ； ； in, represents the total loss of N selected data centers, Represents the total loss of N selected data centers The gradient of represents the loss of the i-th selected data center, represents the weight of the i-th selected data center, β represents the learning rate, Represents the initial parameters.

6. The training method for the postoperative CTV segmentation model for endometrial cancer according to claim 1, characterized in that: The loss is calculated using the following formula: ； in, represents the multimodal image segmentation loss, represents the multimodal cross loss, represents contrast loss.

7. The training method for the postoperative CTV segmentation model for endometrial cancer according to claim 6, characterized in that: The multimodal cross loss is calculated using the following formula: ； in, Indicates the number of samples that do not have corresponding mask images in the query set of the i-th data center, In the first segmented image, the kth sample without a corresponding mask image is represented The element value at position, In the second segmented image, the kth sample without a corresponding mask image is represented The element value of the position, m represents the number of elements of the first segmented image.

8. The training method for the postoperative CTV segmentation model for endometrial cancer according to claim 6, characterized in that: The contrast loss is calculated in the following way: For each sample in the query set without a corresponding mask image, the first segmented image and the second segmented image of the sample are used as a positive sample pair; the first segmented image of the sample and the second segmented image of other samples in the query set without a corresponding mask image are used as negative sample pairs; Get the positive sample pairs and negative sample pairs corresponding to each sample in the query set that has no corresponding mask image; The contrast loss is calculated based on the positive and negative pairs corresponding to each sample in the query set that does not have a corresponding mask image.

9. The training method for the postoperative CTV segmentation model for endometrial cancer according to claim 8, characterized in that: The contrast loss is calculated based on the positive and negative pairs corresponding to each sample in the query set that does not have a corresponding mask image in the following way: ； in, Indicates the number of samples that do not have corresponding mask images in the query set of the i-th data center, represents the first segmented image of the kth sample without a corresponding mask image, represents the second segmented image of the kth sample without a corresponding mask image, represents the second segmented image of the jth sample without a corresponding mask image, τ represents the adjustment parameter, and dice(·,·) represents the DICE loss.

Images