Medical image processing method and device, computer device and storage medium

Abstract

The application relates to a medical image processing method and device, computer equipment, a storage medium and a computer program product. The method comprises the following steps: acquiring an organ image segmentation model constructed in a tree structure; the organ image segmentation model comprises trained organ image segmentation sub-models corresponding to each leaf node in the tree structure; inputting a sample image set into each organ image segmentation sub-model to obtain a predicted segmentation image corresponding to each sample image; adjusting a current integration strategy of the organ image segmentation model according to a segmentation annotation image and the predicted segmentation image to obtain a target integration strategy; combining predicted segmentation results output by each organ image segmentation sub-model for a to-be-segmented organ image according to the target integration strategy to obtain a target organ segmentation image; and the segmentation style corresponding to the target organ segmentation image matches the segmentation style corresponding to the segmentation annotation image. The method can improve the accuracy of the segmentation result.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a medical image processing method, apparatus, computer equipment, storage medium, and computer program product. Background Technology

[0002] In recent years, with the continuous development and progress of medical imaging technology and computer technology, organ segmentation of medical images based on artificial intelligence technology has become the main application area of ​​current medical image analysis.

[0003] In organ segmentation tasks, segmentation models can provide personalized segmentation results based on the delineation characteristics of different doctors or hospitals, thereby improving the doctor's experience. Related technologies often involve learning a segmentation model from a large amount of organ segmentation sample data with different delineation styles. When new data arrives, the existing segmentation model is adapted and its parameters are fine-tuned on the new data.

[0004] However, this method requires a large amount of data to fine-tune the model, and segmentation models, as typical deep learning models, inherently have a very large number of parameters. Therefore, when the amount of data is small, the model cannot be fully trained; a small number of new data samples means that there may be a bias in the sampling itself, causing the model to learn an overall bias, resulting in an inability to accurately output organ segmentation results that match the style of the new data.

[0005] Therefore, traditional techniques suffer from the problem that segmentation models cannot accurately output segmentation results. Summary of the Invention

[0006] Therefore, it is necessary to provide a medical image processing method, apparatus, computer equipment, computer-readable storage medium, and computer program product that can improve the accuracy of segmentation results of segmentation models, in order to address the aforementioned technical problems.

[0007] Firstly, this application provides a medical image processing method. The method includes:

[0008] Obtain an organ image segmentation model constructed with a tree structure; the organ image segmentation model includes trained organ image segmentation sub-models corresponding to each leaf node in the tree structure; wherein, each organ image segmentation sub-model corresponds to a different segmentation style;

[0009] The sample image set is input into each of the organ image segmentation sub-models to obtain the predicted segmentation image corresponding to each sample image in the sample image set;

[0010] Based on the segmentation annotation image corresponding to each sample image and the predicted segmentation image corresponding to each sample image, the current ensemble strategy of the organ image segmentation model is adjusted to obtain the target ensemble strategy.

[0011] According to the target integration strategy, the predicted segmentation results output by each organ image segmentation sub-model for the organ image to be segmented are combined to obtain the target organ segmentation image corresponding to the organ image to be segmented; the segmentation style corresponding to the target organ segmentation image is matched with the segmentation style corresponding to the segmentation annotation image.

[0012] In one embodiment, the step of inputting the sample image set into each of the organ image segmentation sub-models to obtain the predicted segmented image corresponding to each sample image in the sample image set includes:

[0013] The sample image set is input into each of the organ image segmentation sub-models to obtain the predicted segmentation results output by each organ image segmentation sub-model for each of the sample images;

[0014] Based on the current integration strategy, the predicted segmentation results corresponding to each of the sample images are combined to obtain the predicted segmentation image corresponding to each of the sample images.

[0015] In one embodiment, the current integration strategy includes segmentation result weights corresponding to each of the organ image segmentation sub-models; adjusting the current integration strategy of the organ image segmentation model based on the segmentation annotation image corresponding to each of the sample images and the predicted segmentation image corresponding to each of the sample images to obtain the target integration strategy includes:

[0016] Determine the image differences between the segmentation annotation image corresponding to each sample image and the predicted segmentation image corresponding to each sample image;

[0017] Based on the image differences, adjust the segmentation result weights corresponding to each organ image segmentation sub-model to obtain the target segmentation result weights corresponding to each organ image segmentation sub-model.

[0018] The target ensemble strategy is obtained based on the target segmentation result weights corresponding to each organ image segmentation sub-model.

[0019] In one embodiment, adjusting the segmentation result weights corresponding to each organ image segmentation sub-model based on the image differences to obtain the target segmentation result weights corresponding to each organ image segmentation sub-model includes:

[0020] Based on the image differences, the segmentation result weights corresponding to each organ image segmentation sub-model are adjusted to obtain the adjusted weights corresponding to each organ image segmentation sub-model.

[0021] Based on the adjusted weights corresponding to each organ image segmentation sub-model, the predicted segmentation results output by each organ image segmentation sub-model are combined to obtain the adjusted predicted segmentation image corresponding to each sample image.

[0022] The adjusted predicted segmentation image is used as the predicted segmentation image. The step of adjusting the segmentation result weights corresponding to each organ image segmentation sub-model according to the image differences is returned until the image differences meet the preset adjustment end condition, and the target segmentation result weights corresponding to each organ image segmentation sub-model are obtained.

[0023] In one embodiment, obtaining the organ image segmentation model constructed with a tree structure includes:

[0024] Obtain a training sample image set, a sample segmentation labeled image set, and an organ image segmentation sub-model to be trained in the initial segmentation model tree; wherein, the sample segmentation labeled image set includes sample segmentation labeled images corresponding to each training sample image in the training sample image set; the sample segmentation labeled images are obtained by labeling the actual region where the object to be segmented is located in the training sample image;

[0025] Based on the training sample image set and the sample segmentation labeled image set, the organ image segmentation sub-model to be trained in the initial segmentation model tree is trained to obtain the target segmentation model tree constructed with the tree structure; the segmentation model tree includes the trained organ image segmentation sub-model corresponding to each of the leaf nodes in the tree structure;

[0026] The training sample image set is input into the trained organ image segmentation sub-model corresponding to each leaf node to obtain the sample prediction segmentation result output by each organ image segmentation sub-model for the training sample image;

[0027] The organ image segmentation model is constructed based on the sample prediction segmentation results output by each organ image segmentation sub-model.

[0028] In one embodiment, constructing the organ image segmentation model based on the sample prediction segmentation results output by each of the organ image segmentation sub-models includes:

[0029] Determine the sample prediction probability image in each of the sample prediction segmentation results;

[0030] By stitching together the predicted probability images of each sample, a stitched predicted probability image of the sample is obtained.

[0031] A preset ensemble model is constructed on the stitched sample prediction probability image to obtain the organ image segmentation model.

[0032] In one embodiment, the initial segmentation model tree includes the organ image segmentation sub-model to be trained corresponding to the root node in the tree structure; the step of training the organ image segmentation sub-model to be trained in the initial segmentation model tree according to the training sample image set and the sample segmentation labeled image set to obtain the target segmentation model tree constructed with the tree structure includes:

[0033] The organ image segmentation sub-model to be trained corresponding to the root node is trained based on the training sample image set and the sample segmentation labeled image set to obtain the trained organ image segmentation sub-model corresponding to the root node.

[0034] Obtain the texture feature image corresponding to each residual image; each residual image is obtained based on the difference between the root node prediction probability image corresponding to each training sample image and the sample segmentation annotation image corresponding to each training sample image; the root node prediction probability image is the sample prediction probability image in the sample prediction segmentation result output by the trained organ image segmentation sub-model corresponding to the root node for the training sample image.

[0035] The target segmentation model tree is obtained based on the clustering results of each texture feature image.

[0036] In one embodiment, obtaining the target segmentation model tree based on the clustering results of each of the texture feature images includes:

[0037] Based on the clustering results of each texture feature image, the training sample image set is divided into a first training sample image set and a second training sample image set, and the first child node and the second child node of the root node are split from the root node.

[0038] The first child node is used as the root node, and the first training sample image set is used as the training sample image set;

[0039] And, the second child node is used as the root node, and the second training sample image set is used as the training sample image set;

[0040] The step of training the organ image segmentation sub-model corresponding to the root node based on the training sample image set and the sample segmentation labeled image set is performed until the number of the first training sample image set or the second training sample image set is less than a preset number, thereby obtaining the trained organ image segmentation sub-model corresponding to each node in the tree structure.

[0041] The target segmentation model tree is determined based on the trained organ image segmentation sub-models corresponding to each node.

[0042] Secondly, this application also provides a medical image processing apparatus. The apparatus includes:

[0043] The acquisition module is used to acquire an organ image segmentation model constructed in a tree structure; the organ image segmentation model includes a trained organ image segmentation sub-model corresponding to each leaf node in the tree structure; wherein, each organ image segmentation sub-model corresponds to a different segmentation style;

[0044] The input module is used to input the sample image set into each of the organ image segmentation sub-models respectively, and obtain the predicted segmentation results output by each organ image segmentation sub-model for the sample images in the sample image set;

[0045] The combination module is used to combine the predicted segmentation results output by each organ image segmentation sub-model according to the current integration strategy of the organ image segmentation model to obtain the predicted segmentation image corresponding to each sample image.

[0046] An adjustment module is used to adjust the current ensemble strategy based on the segmentation annotation image corresponding to each sample image and the predicted segmentation image corresponding to each sample image to obtain a target ensemble strategy; the target ensemble strategy is used for the organ image segmentation model to output a target organ segmentation image for the organ image to be segmented; the segmentation style corresponding to the target organ segmentation image matches the segmentation style corresponding to the segmentation annotation image.

[0047] Thirdly, this application also provides a computer device. The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of the method described above.

[0048] Fourthly, this application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program thereon, which, when executed by a processor, implements the steps of the above-described method.

[0049] Fifthly, this application also provides a computer program product. The computer program product includes a computer program that, when executed by a processor, implements the steps of the above-described method.

[0050] The aforementioned medical image processing method, apparatus, computer equipment, storage medium, and computer program product acquire an organ image segmentation model constructed with a tree structure. The organ image segmentation model includes trained organ image segmentation sub-models corresponding to each leaf node in the tree structure. Each organ image segmentation sub-model has a different segmentation style. A set of sample images is input into each organ image segmentation sub-model to obtain a predicted segmentation image corresponding to each sample image in the set. Based on the segmentation annotation image and the predicted segmentation image corresponding to each sample image, the current ensemble strategy of the organ image segmentation model is adjusted to obtain a target ensemble strategy. According to the target ensemble strategy, the predicted segmentation results output by each organ image segmentation sub-model for the organ image to be segmented are combined to obtain a target organ segmentation image corresponding to the organ image to be segmented. The segmentation style of the target organ segmentation image matches the segmentation style of the segmentation annotation image.

[0051] Thus, by using the segmentation annotation images corresponding to the sample images in the sample image set and the predicted segmentation images corresponding to the sample images, the current ensemble strategy of the existing organ image segmentation model is fine-tuned to obtain the target ensemble strategy. Since there is no need to retrain a segmentation mapping relationship for the organ image segmentation model, but rather to fine-tune the current ensemble strategy, even when the number of sample images in the sample image set is small, the target ensemble strategy can still enable the organ image segmentation model to accurately output a target organ segmentation image that matches the segmentation style corresponding to the segmentation annotation image for the organ image to be segmented. This achieves accurate provision of personalized segmentation results based on the delineation characteristics of the segmentation annotation image. Even when the number of sample images is small, there will be no overall shift in the organ image segmentation model during training, thereby improving the accuracy of the segmentation results output by the organ image segmentation model. This invention addresses the problem in related technologies where, when the number of sample images in the sample image set is small, the organ image segmentation model cannot be sufficiently trained. Furthermore, the small number of sample images leads to a bias in the sample image set itself, causing the organ image segmentation model to exhibit an overall bias during the learning process. Consequently, the segmentation style corresponding to the final output of the organ image segmentation model cannot accurately match the segmentation style corresponding to the segmentation annotation image, resulting in poor segmentation accuracy. Attached Figure Description

[0052] Figure 1 This is a flowchart illustrating a medical image processing method in one embodiment;

[0053] Figure 2 This is a flowchart illustrating the steps to obtain the target integration strategy in one embodiment;

[0054] Figure 3This is a flowchart illustrating the steps of acquiring an organ image segmentation model using a tree structure in one embodiment;

[0055] Figure 4 This is a flowchart illustrating the steps involved in obtaining a target segmentation model tree constructed using a tree structure, as shown in one embodiment.

[0056] Figure 5 This is a schematic diagram of the clustering result corresponding to a node in a tree structure in one embodiment;

[0057] Figure 6 This is a flowchart illustrating a medical image processing method in another embodiment;

[0058] Figure 7 This is a structural block diagram of a medical image processing device in one embodiment;

[0059] Figure 8 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0060] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0061] It should be noted that the terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims.

[0062] In one embodiment, such as Figure 1 As shown, a medical image processing method is provided, applied to a computer device. In practical applications, the computer device can be a user terminal, or it can be implemented using a standalone server or a server cluster composed of multiple servers. The user terminal can be, but is not limited to, various personal computers, laptops, smartphones, tablets, and portable wearable devices. In this embodiment, the method includes the following steps:

[0063] Step S110: Obtain the organ image segmentation model constructed with a tree structure.

[0064] The organ image segmentation model includes trained organ image segmentation sub-models corresponding to each leaf node in the tree structure.

[0065] Among them, the segmentation styles corresponding to the segmentation sub-models of the above organ images are different.

[0066] Among them, the organ image segmentation model is an ensemble model.

[0067] In practice, the computer device can acquire an organ image segmentation model constructed in a tree structure. This organ image segmentation model is an ensemble model, and it includes trained organ image segmentation sub-models corresponding to each leaf node in the tree structure. Furthermore, each organ image segmentation sub-model corresponds to a different segmentation style.

[0068] Step S120: Input the sample image set into the segmentation sub-model of each organ image respectively to obtain the predicted segmentation image corresponding to each sample image in the sample image set.

[0069] The number of sample images in the sample image set meets a preset limit, which is not restricted here. For example, when the sample image set is a small sample set, the preset number can be 10 or less.

[0070] The sample image is an image containing the object to be segmented.

[0071] The object to be segmented can be an organ in an organism, such as the liver or gallbladder.

[0072] In practice, the computer device can input the sample image set into the segmentation sub-model of each organ image, and use the segmentation sub-model of each organ image to predict each sample image in the sample image set, so as to obtain the predicted segmented image corresponding to each sample image in the sample image set.

[0073] Step S130: Adjust the current ensemble strategy of the organ image segmentation model based on the segmentation annotation image corresponding to each sample image and the predicted segmentation image corresponding to each sample image to obtain the target ensemble strategy.

[0074] In practical applications, the segmented and labeled image can be named the gold standard segmentation result.

[0075] In practice, the computer device can adjust the current ensemble strategy of the organ image segmentation model based on the segmentation annotation image corresponding to each sample image and the predicted segmentation image corresponding to each sample image, so as to obtain the target ensemble strategy.

[0076] Step S140: According to the target integration strategy, combine the predicted segmentation results output by each organ image segmentation sub-model for the organ image to be segmented to obtain the target organ segmentation image corresponding to the organ image to be segmented.

[0077] Among them, the organ image to be segmented is an image containing the object to be segmented.

[0078] In this context, the segmentation annotation images corresponding to each sample image belong to the same segmentation style. For example, the segmentation annotation images corresponding to each sample image may come from the same doctor or the same hospital.

[0079] In this process, the segmentation style corresponding to the segmentation image of the target organ is matched with the segmentation style corresponding to the segmentation annotation image.

[0080] Among them, the target organ segmentation image can represent the predicted region of the object to be segmented in the image of the organ to be segmented.

[0081] Among them, the target organ segmentation image can also be named the target organ probability image, which can characterize the target probability of each pixel in the image of the organ to be segmented belonging to the object to be segmented.

[0082] In practice, after the computer device acquires the image of the organ to be segmented, it can input the image of the organ to be segmented into the trained organ image segmentation sub-models corresponding to each leaf node in the organ image segmentation model. Through the target ensemble strategy, the predicted segmentation results output by each organ image segmentation sub-model for the organ to be segmented are combined to obtain the target organ segmentation image output by the organ image segmentation model for the organ to be segmented. The segmentation style corresponding to the target organ segmentation image matches the segmentation style corresponding to the above-mentioned segmentation annotation image.

[0083] In the aforementioned medical image processing method, an organ image segmentation model constructed using a tree structure is obtained. This model includes trained organ image segmentation sub-models corresponding to each leaf node in the tree structure, each sub-model having a different segmentation style. Sample image sets are input into each sub-model to obtain predicted segmentation images for each sample image. Based on the segmentation annotation images and predicted segmentation images for each sample image, the current ensemble strategy of the organ image segmentation model is adjusted to obtain a target ensemble strategy. According to the target ensemble strategy, the predicted segmentation results output by each sub-model for the organ image to be segmented are combined to obtain the target organ segmentation image. The segmentation style of the target organ segmentation image matches the segmentation style of the annotation images.

[0084] Thus, by using the segmentation annotation images corresponding to the sample images in the sample image set and the predicted segmentation images corresponding to the sample images, the current ensemble strategy of the existing organ image segmentation model is fine-tuned to obtain the target ensemble strategy. Since there is no need to retrain a segmentation mapping relationship for the organ image segmentation model, but rather to fine-tune the current ensemble strategy, even when the number of sample images in the sample image set is small, the target ensemble strategy can still enable the organ image segmentation model to accurately output a target organ segmentation image that matches the segmentation style corresponding to the segmentation annotation image for the organ image to be segmented. This achieves accurate provision of personalized segmentation results based on the delineation characteristics of the segmentation annotation image. Even when the number of sample images is small, there will be no overall shift in the organ image segmentation model during training, thereby improving the accuracy of the segmentation results output by the organ image segmentation model. This invention addresses the problem in related technologies where, when the number of sample images in the sample image set is small, the organ image segmentation model cannot be sufficiently trained. Furthermore, the small number of sample images leads to a bias in the sample image set itself, causing the organ image segmentation model to exhibit an overall bias during the learning process. Consequently, the segmentation style corresponding to the final output of the organ image segmentation model cannot accurately match the segmentation style corresponding to the segmentation annotation image, resulting in poor segmentation accuracy.

[0085] In one embodiment, the sample image set is input into each organ image segmentation sub-model to obtain the predicted segmentation image corresponding to each sample image in the sample image set, including: inputting the sample image set into each organ image segmentation sub-model to obtain the predicted segmentation result output by each organ image segmentation sub-model for each sample image; and combining the predicted segmentation results corresponding to each sample image according to the current integration strategy to obtain the predicted segmentation image corresponding to each sample image.

[0086] The predicted segmentation results include a predicted probability image.

[0087] The predicted probability image represents the probability that each pixel in the sample image belongs to the object to be segmented. The numerical value of each pixel ranges from 0 to 1.

[0088] In specific implementation, during the process of inputting the sample image set into each organ image segmentation sub-model to obtain the predicted segmentation image corresponding to each sample image in the sample image set, the computer device can input the sample image set into each organ image segmentation sub-model to obtain the predicted segmentation result output by each organ image segmentation sub-model for each sample image. The predicted segmentation result includes a predicted probability image. Through the current integration strategy of the organ image segmentation model, the predicted probability images output by each organ image segmentation sub-model for each sample image are combined to obtain the predicted segmentation image corresponding to each sample image.

[0089] The technical solution of this embodiment involves inputting sample image sets into each organ image segmentation sub-model to obtain the predicted segmentation results output by each organ image segmentation sub-model for each sample image; according to the current integration strategy, the predicted segmentation results corresponding to each sample image are combined to obtain the predicted segmented image corresponding to each sample image; in this way, the current integration strategy can realize the integrated learning of the predicted segmentation results output by each organ image segmentation sub-model. Since each organ image segmentation sub-model has a different segmentation style, compared with using an organ image segmentation model with a single segmentation style for prediction, the accuracy of the prediction results corresponding to each sample image can be effectively improved.

[0090] In one embodiment, such as Figure 2 As shown, in step S130, based on the segmentation annotation image corresponding to each sample image and the predicted segmentation image corresponding to each sample image, the current ensemble strategy of the organ image segmentation model is adjusted to obtain the target ensemble strategy, which specifically includes:

[0091] Step S210: Determine the image difference between the segmentation annotation image corresponding to each sample image and the predicted segmentation image corresponding to each sample image.

[0092] The current integration strategy includes the segmentation result weights corresponding to the segmentation sub-models for each organ image.

[0093] In the process of combining the predicted segmentation results output by each organ image segmentation sub-model for each sample image according to the current integration strategy of the organ image segmentation model, and obtaining the predicted segmentation image corresponding to each sample image, the computer equipment can perform weighted combination of the predicted probability images output by each organ image segmentation sub-model for each sample image according to the segmentation result weights corresponding to each organ image segmentation sub-model, and obtain the predicted segmentation image corresponding to each sample image.

[0094] Among them, the predicted segmentation image corresponding to each sample image can be named the target prediction probability image, which can characterize the probability that each pixel in the sample image belongs to the target object to be segmented.

[0095] The segmentation annotation image is obtained by annotating the actual region where the object to be segmented is located in the sample image. The segmentation annotation image can represent the actual region corresponding to the object to be segmented in the sample image.

[0096] Among them, the segmentation and annotation image is a mask image.

[0097] Among them, the predicted segmentation image can represent the predicted region of the object to be segmented in the sample image.

[0098] In practice, during the process of adjusting the current ensemble strategy of the organ image segmentation model based on the segmentation annotation image and the predicted segmentation image corresponding to each sample image to obtain the target ensemble strategy, the computer device can determine the image differences between the segmentation annotation image and the predicted segmentation image corresponding to each sample image.

[0099] Step S220: Adjust the segmentation result weights of each organ image segmentation sub-model according to the image differences to obtain the target segmentation result weights of each organ image segmentation sub-model.

[0100] In practice, the computer device can adjust the segmentation result weights of each organ image segmentation sub-model according to the image differences until the preset adjustment end condition is met, and obtain the target segmentation result weights of each organ image segmentation sub-model.

[0101] Step S230: Obtain the target ensemble strategy based on the target segmentation result weights corresponding to the segmentation sub-models of each organ image.

[0102] After obtaining the target segmentation result weights corresponding to each organ image segmentation sub-model, the computer equipment can determine the target ensemble strategy of the organ image segmentation model based on the target segmentation result weights corresponding to each organ image segmentation sub-model.

[0103] The technical solution of this embodiment includes the following integration strategy: the segmentation result weights corresponding to each organ image segmentation sub-model; determining the image difference between the segmentation annotation image corresponding to each sample image and the predicted segmentation image corresponding to each sample image; adjusting the segmentation result weights corresponding to each organ image segmentation sub-model according to the image difference to obtain the target segmentation result weights corresponding to each organ image segmentation sub-model; and obtaining the target integration strategy according to the target segmentation result weights corresponding to each organ image segmentation sub-model. In this way, the segmentation result weights corresponding to each organ image segmentation sub-model can be accurately adjusted based on the image difference between the segmentation annotation image and the predicted segmentation image to obtain the target segmentation result weights corresponding to each organ image segmentation sub-model. Thus, the target integration strategy can be obtained based on the target segmentation result weights, and then the target organ segmentation image that matches the segmentation style corresponding to the segmentation annotation image can be accurately output based on the target integration strategy.

[0104] In one embodiment, step S220, adjusting the segmentation result weights corresponding to each organ image segmentation sub-model based on image differences to obtain the target segmentation result weights corresponding to each organ image segmentation sub-model, specifically includes: adjusting the segmentation result weights corresponding to each organ image segmentation sub-model based on image differences to obtain the adjusted weights corresponding to each organ image segmentation sub-model; combining the predicted segmentation results output by each organ image segmentation sub-model based on the adjusted weights corresponding to each organ image segmentation sub-model to obtain the adjusted predicted segmentation image corresponding to each sample image; using the adjusted predicted segmentation image as the predicted segmentation image, returning to the step of adjusting the segmentation result weights corresponding to each organ image segmentation sub-model based on image differences, until the image differences meet the preset adjustment termination condition, to obtain the target segmentation result weights corresponding to each organ image segmentation sub-model.

[0105] In the specific implementation, during the process of adjusting the segmentation result weights of each organ image segmentation sub-model according to the image differences to obtain the target segmentation result weights of each organ image segmentation sub-model, the computer device can adjust the segmentation result weights of each organ image segmentation sub-model according to the image differences to obtain the adjusted weights of each organ image segmentation sub-model. Then, the computer device can use the adjusted weights of each organ image segmentation sub-model to perform weighted combination of the predicted probability images in the predicted segmentation results output by each organ image segmentation sub-model for the sample images to obtain the adjusted predicted segmentation images corresponding to each sample image.

[0106] Then, the computer device can use the adjusted predicted segmentation image corresponding to each sample image as the predicted segmentation image corresponding to each sample image, and return the step of adjusting the segmentation result weights corresponding to each organ image segmentation sub-model according to the image difference, and obtaining the adjusted weights corresponding to each organ image segmentation sub-model. This process continues until the image difference between the segmentation annotation image corresponding to each sample image and the predicted segmentation image corresponding to each sample image is less than the preset difference. Then, it is determined that the image difference meets the preset adjustment termination condition, and the target segmentation result weights corresponding to each organ image segmentation sub-model are obtained.

[0107] The technical solution of this embodiment adjusts the segmentation result weights of each organ image segmentation sub-model according to image differences to obtain the adjusted weights of each organ image segmentation sub-model; based on the adjusted weights of each organ image segmentation sub-model, the predicted segmentation results output by each organ image segmentation sub-model are combined to obtain the adjusted predicted segmentation image corresponding to each sample image; using the adjusted predicted segmentation image as the predicted segmentation image, the step of adjusting the segmentation result weights of each organ image segmentation sub-model according to image differences is returned until the image differences meet the preset adjustment termination condition to obtain the target segmentation result weights of each organ image segmentation sub-model; thus, by iteratively adjusting the segmentation result weights of each organ image segmentation sub-model until the image differences between the segmentation annotation image and the predicted segmentation image meet the preset adjustment termination condition, the target segmentation result weights of each organ image segmentation sub-model can be accurately obtained. Therefore, the target ensemble strategy obtained based on the target segmentation result weights can accurately output the target organ segmentation image that matches the segmentation style corresponding to the segmentation annotation image, thereby improving the accuracy of the segmentation results output by the organ image segmentation model.

[0108] In one embodiment, such as Figure 3 As shown, step S110, obtaining an organ image segmentation model constructed with a tree structure, specifically includes:

[0109] Step S310: Obtain the training sample image set, the sample segmentation labeled image set, and the organ image segmentation sub-model to be trained in the initial segmentation model tree.

[0110] The sample segmentation labeled image set includes sample segmentation labeled images corresponding to each training sample image in the training sample image set.

[0111] The training sample images are images containing the objects to be segmented.

[0112] Among them, the sample segmentation annotation image is obtained by annotating the actual region where the object to be segmented is located in the training sample image. The sample segmentation annotation image can represent the actual region corresponding to the object to be segmented in the training sample image.

[0113] In practical applications, the sample segmentation labeled image can be named the gold standard segmentation result.

[0114] Among them, the sample segmentation annotation image is a mask image.

[0115] The sample segmentation labeled image set corresponds to at least two segmentation styles.

[0116] The organ image segmentation sub-model to be trained can be any image segmentation model, such as the UNet segmentation model (an image segmentation model).

[0117] In practice, during the process of acquiring an organ image segmentation model constructed in a tree structure, the computer device can acquire a training sample image set, a sample segmentation labeled image set, and the organ image segmentation sub-model to be trained in the initial segmentation model tree. The sample segmentation labeled image set includes sample segmentation labeled images corresponding to each training sample image in the training sample image set; these image sets are obtained by labeling the actual regions where the object to be segmented is located in the training sample images.

[0118] In the process of acquiring training sample images from the training sample image set, the computer device can first acquire an initial sample image set and perform a unified segmentation preprocessing operation (including but not limited to data cleaning, contrast enhancement, data normalization, etc.) on each initial sample image in the initial sample image set to obtain the training sample image corresponding to each initial sample image.

[0119] Step S320: Based on the training sample image set and the sample segmentation labeled image set, train the organ image segmentation sub-model to be trained in the initial segmentation model tree to obtain the target segmentation model tree constructed with a tree structure.

[0120] The segmentation model tree includes trained organ image segmentation sub-models corresponding to each leaf node in the tree structure.

[0121] Among them, the trained organ image segmentation sub-models corresponding to each leaf node can be used as model clusters.

[0122] In practice, the computer device can train the organ image segmentation sub-model to be trained in the initial segmentation model tree based on the training sample image set and the sample segmentation labeled image set. When the trained organ image segmentation sub-model meets the preset training termination condition, the trained organ image segmentation sub-model corresponding to each node in the tree structure corresponding to the organ image segmentation model can be obtained. Thus, based on the trained organ image segmentation sub-model corresponding to each node, the target segmentation model tree constructed with the tree structure can be obtained. The target segmentation model tree includes the trained organ image segmentation sub-model corresponding to each leaf node in the tree structure.

[0123] Step S330: Input the training sample image set into the trained organ image segmentation sub-model corresponding to each leaf node to obtain the sample prediction segmentation result output by each organ image segmentation sub-model for the training sample image.

[0124] In practice, the computer device can input the training sample image set into the pre-trained organ image segmentation sub-model corresponding to each leaf node, and use the pre-trained organ image segmentation sub-model corresponding to each leaf node to predict the training sample image, thereby obtaining the sample prediction segmentation result output by each organ image segmentation sub-model for the training sample image.

[0125] Step S340: Construct an organ image segmentation model based on the sample prediction segmentation results output by each organ image segmentation sub-model.

[0126] In practice, computer devices can construct organ image segmentation models by predicting segmentation results based on samples output by each organ image segmentation sub-model, taking advantage of the characteristics of ensemble learning.

[0127] The technical solution of this embodiment involves acquiring a training sample image set, a sample segmentation annotation image set, and an organ image segmentation sub-model to be trained in an initial segmentation model tree. The sample segmentation annotation image set includes sample segmentation annotation images corresponding to each training sample image in the training sample image set. These annotation images are obtained by annotating the actual regions where the object to be segmented is located in the training sample images. The organ image segmentation sub-model to be trained in the initial segmentation model tree is trained based on the training sample image set and the sample segmentation annotation image set, resulting in a target segmentation model tree constructed with a tree structure. The target segmentation model tree includes the leaf nodes corresponding to each leaf node in the tree structure. The trained organ image segmentation sub-model is used. The training sample image set is input into the trained organ image segmentation sub-model corresponding to each leaf node to obtain the sample prediction segmentation result output by each organ image segmentation sub-model for the training sample image. Based on the sample prediction segmentation result output by each organ image segmentation sub-model, an organ image segmentation model is constructed. In this way, based on the trained organ image segmentation sub-model corresponding to each leaf node in the target segmentation model tree constructed with a tree structure, and the sample prediction segmentation result output by the training sample image in the training sample image, the accuracy of the segmentation result output by the organ image segmentation model can be improved.

[0128] In one embodiment, step S340, constructing an organ image segmentation model based on the sample prediction segmentation results output by each organ image segmentation sub-model, includes: determining the sample prediction probability image in each sample prediction segmentation result; stitching together the sample prediction probability images to obtain a stitched sample prediction probability image; and constructing a preset ensemble model on the stitched sample prediction probability image to obtain the organ image segmentation model.

[0129] The predicted segmentation results for each sample include the sample prediction probability image output by the corresponding organ image segmentation sub-model for the training sample image.

[0130] The sample prediction probability image represents the probability that each pixel in the training sample image belongs to the object to be segmented. The numerical value of each pixel ranges from 0 to 1.

[0131] The sample prediction probability image has the same image size as the training sample image.

[0132] The preset ensemble model consists of a preset convolutional network.

[0133] In practice, during the process of constructing an organ image segmentation model based on the sample prediction segmentation results output by the segmentation sub-models of each organ image, the computer device can determine the sample prediction probability image in the sample prediction segmentation results and stitch together the sample prediction probability images to obtain the stitched sample prediction probability image.

[0134] Specifically, if the size of the training sample image is a*b (where a and b are natural numbers), then the size of the predicted probability image of each sample is a*b. If the number of trained organ image segmentation sub-models corresponding to each leaf node is m (where m is a natural number), then the size of the stitched sample predicted probability image is a*b*m.

[0135] Then, the computer device can construct a shallow ensemble model on the stitched sample prediction probability image. This ensemble model consists of a pre-defined convolutional network, which is composed of a 3x3 convolution and a 1x1 convolution. This enables the ensemble learning of the sample prediction segmentation results output by the trained organ image segmentation sub-model corresponding to each leaf node, resulting in a complete ensemble model, which serves as the organ image segmentation model constructed with the tree structure mentioned above.

[0136] The technical solution of this embodiment involves determining the sample prediction probability image in the predicted segmentation results of each sample; stitching together the predicted probability images of each sample to obtain a stitched sample prediction probability image; and constructing a preset convolutional network on the stitched sample prediction probability image to obtain an organ image segmentation model. In this way, the organ image segmentation model can integrate and learn the sample prediction segmentation results output by the organ image segmentation sub-model corresponding to each leaf node, thereby improving the accuracy of the segmentation results output by the organ image segmentation model.

[0137] In one embodiment, the initial segmentation model tree includes the organ image segmentation sub-model to be trained corresponding to the root node in the tree structure; such as Figure 4 As shown, step S320 involves training the organ image segmentation sub-model to be trained in the initial segmentation model tree based on the training sample image set and the sample segmentation labeled image set, to obtain the target segmentation model tree constructed with a tree structure, including:

[0138] Step S410: Train the organ image segmentation sub-model to be trained corresponding to the root node based on the training sample image set and the sample segmentation labeled image set, and obtain the trained organ image segmentation sub-model corresponding to the root node.

[0139] In the specific implementation, the computer device trains the organ image segmentation sub-model to be trained in the initial segmentation model tree based on the training sample image set and the sample segmentation labeled image set to obtain the target segmentation model tree constructed in a tree structure. The computer device can first train the organ image segmentation sub-model to be trained corresponding to the root node in the initial segmentation model tree based on the training sample image set and the sample segmentation labeled image set. When the trained organ image segmentation sub-model corresponding to the root node meets the preset training termination condition, the trained organ image segmentation sub-model corresponding to the root node is obtained.

[0140] Step S420: Obtain the texture feature image corresponding to each residual image.

[0141] Each residual image is obtained by comparing the root node prediction probability image corresponding to each training sample image with the sample segmentation annotation image corresponding to each training sample image.

[0142] The root node prediction probability image is the sample prediction probability image in the sample prediction segmentation result output by the trained organ image segmentation sub-model corresponding to the root node for the training sample image.

[0143] Among them, the sample segmentation annotation image is a mask image.

[0144] Among them, the sample segmentation labeled image can represent the actual region corresponding to the object to be segmented in the training sample image.

[0145] Among them, the sample prediction probability image can represent the predicted region of the object to be segmented in the training sample image.

[0146] In a specific implementation, the computer device can use the trained organ image segmentation sub-model corresponding to the root node to predict the training sample images in the training sample image set, and obtain the sample prediction segmentation result output by the trained organ image segmentation sub-model for the training sample images. The sample prediction probability image in the sample prediction segmentation result is used as the root node prediction probability image corresponding to the training sample image.

[0147] Then, the computer device can subtract the root node prediction probability image corresponding to each training sample image from the sample segmentation annotation image corresponding to each training sample image to obtain the difference between the root node prediction probability image corresponding to each training sample image and the sample segmentation annotation image corresponding to each training sample image, so as to determine the residual image corresponding to each training sample image.

[0148] Then, the computer device can extract the texture features of each residual image to obtain the texture feature image corresponding to each residual image. Specifically, the computer device can extract texture features based on the graycoprops function of the gray-level co-occurrence matrix.

[0149] Step S430: Based on the clustering results of each texture feature image, obtain the target segmentation model tree.

[0150] In specific implementation, computer equipment can use K-Means (K-means clustering algorithm) (K=2), that is, use an unsupervised method to divide each texture feature image to obtain the 2-Means clustering result of each texture feature image. Based on the clustering result of each texture feature image, train the organ image segmentation sub-model corresponding to the other nodes in the tree structure except the root node to obtain the target segmentation model tree constructed with the tree structure.

[0151] The technical solution of this embodiment includes an initial segmentation model tree comprising an organ image segmentation sub-model to be trained corresponding to the root node in the tree structure; training the organ image segmentation sub-model to be trained corresponding to the root node based on a set of training sample images and a set of sample segmentation annotation images to obtain a trained organ image segmentation sub-model corresponding to the root node; obtaining texture feature images corresponding to each residual image; each residual image is obtained based on the difference between the root node prediction probability image corresponding to each training sample image and the sample segmentation annotation image corresponding to each training sample image; the root node prediction probability image is the sample prediction probability image in the sample prediction segmentation result output by the trained organ image segmentation sub-model corresponding to the root node for the training sample images; and obtaining a target segmentation model tree based on the clustering results of each texture feature image. In this way, training sample images with different texture features can be separated to train a target segmentation model tree, so that the organ image segmentation sub-model corresponding to each leaf node in the target segmentation model tree can be matched to different segmentation styles, thereby accurately providing personalized segmentation results.

[0152] In one embodiment, obtaining a target segmentation model tree based on the clustering results of each texture feature image includes: dividing the training sample image set into a first training sample image set and a second training sample image set based on the clustering results of each texture feature image set, and splitting the root node into a first child node and a second child node; using the first child node as the root node and the first training sample image set as the training sample image set; and using the second child node as the root node and the second training sample image set as the training sample image set; performing the step of training the organ image segmentation sub-model to be trained corresponding to the root node based on the training sample image set and the sample segmentation annotation image set, until the number of the first training sample image set or the second training sample image set is less than a preset number, obtaining the trained organ image segmentation sub-model corresponding to each node in the tree structure; and determining the target segmentation model tree based on the trained organ image segmentation sub-model corresponding to each node.

[0153] In specific implementation, during the process of obtaining the target segmentation model tree based on the clustering results of each texture feature image, the computer device can divide the training sample image set into a first training sample image set and a second training sample image set based on the 2-Means clustering results of each texture feature image, and split the first child node and the second child node of the root node from the root node; then, the first subtree and the second subtree of the root node are established, that is, the first child node is used as the root node and the first training sample image set is used as the training sample image set; and the second child node is used as the root node and the second training sample image set is used as the training sample image set, and the process of dividing the training sample image set into the second subtree is executed. The process of training the organ image segmentation sub-model corresponding to the root node with the sample segmentation labeled image set continues until the number of images in the first training sample image set corresponding to the first sub-node or the second training sample image set corresponding to the second sub-node is less than a preset number (e.g., less than 30). At this point, the splitting of the root node is stopped, and the trained organ image segmentation sub-models corresponding to each node in the tree structure can be obtained. Then, based on the trained organ image segmentation sub-models corresponding to each node, the target segmentation model tree can be determined. The target segmentation model tree includes a set of organ image segmentation sub-models with different segmentation styles, which are the trained organ image segmentation sub-models corresponding to each leaf node in the tree structure.

[0154] For the ease of understanding of those skilled in the art, Figure 5 This provides a schematic diagram of the clustering results corresponding to a node in a tree structure. For example... Figure 5 As shown, the data to the left of the dashed line represents the residual image from the first doctor (Doctor A), indicated by a "·"; the data to the right of the dashed line represents the residual image from the second doctor (Doctor B), indicated by a "+"; the horizontal axis represents different images. Figure 5 As shown, the residual images can be classified into different categories based on the 2-Means clustering results of each texture feature image, such as... Figure 5 Hollow markers represent the first category, and solid markers represent the second category. It can be seen that in the clustering results corresponding to this node, the residual image of the second doctor can be classified well.

[0155] The technical solution of this embodiment divides the training sample image set into a first training sample image set and a second training sample image set according to the clustering results of each texture feature image, and splits the first child node and the second child node of the root node from the root node; the first child node is used as the root node, and the first training sample image set is used as the training sample image set; and the second child node is used as the root node, and the second training sample image set is used as the training sample image set; the step of training the organ image segmentation sub-model corresponding to the root node according to the training sample image set and the sample segmentation annotation image set is performed until the number of the first training sample image set or the second training sample image set is less than a preset number, so as to obtain the trained organ image segmentation sub-model corresponding to each node in the tree structure; according to each node... The corresponding trained organ image segmentation sub-model is used to determine the target segmentation model tree. In this way, training sample images with different texture features can be divided into different categories, resulting in a first training sample image set and a second training sample image set belonging to different categories. These are then used to train the organ image segmentation sub-models corresponding to the first and second child nodes of the root node, thereby recursively constructing the trained organ image segmentation sub-models corresponding to each node. This allows the organ image segmentation sub-models corresponding to each leaf node in the tree structure to accurately match different segmentation styles. Furthermore, based on the segmentation results with different segmentation styles output by the organ image segmentation sub-models, the target organ segmentation image that matches the target segmentation style can be accurately output.

[0156] In another embodiment, such as Figure 6 As shown, a medical image processing method is provided, and the method is illustrated using an application to a computer device as an example. The method includes the following steps:

[0157] Step S602: Obtain the training sample image set, the sample segmentation labeled image set, and the organ image segmentation sub-model to be trained in the initial segmentation model tree.

[0158] Step S604: Train the organ image segmentation sub-model to be trained in the initial segmentation model tree according to the training sample image set and the sample segmentation labeled image set to obtain the target segmentation model tree constructed with a tree structure.

[0159] Step S606: Input the training sample image set into the trained organ image segmentation sub-model corresponding to each leaf node to obtain the sample prediction segmentation result output by each organ image segmentation sub-model for the training sample image.

[0160] Step S608: Determine the sample prediction probability image in the prediction segmentation results of each sample.

[0161] Step S610: stitch together the predicted probability images of each sample to obtain the stitched sample predicted probability image.

[0162] Step S612: Construct a preset ensemble model on the stitched sample prediction probability image to obtain an organ image segmentation model constructed with a tree structure.

[0163] Step S614: Input the sample image set into the segmentation sub-model of each organ image respectively to obtain the predicted segmentation image corresponding to each sample image in the sample image set.

[0164] Step S616: Adjust the current ensemble strategy of the organ image segmentation model based on the segmentation annotation image corresponding to each sample image and the predicted segmentation image corresponding to each sample image to obtain the target ensemble strategy.

[0165] Step S618: According to the target ensemble strategy, combine the predicted segmentation results output by each organ image segmentation sub-model for the organ image to be segmented to obtain the target organ segmentation image corresponding to the organ image to be segmented.

[0166] It should be noted that the specific limitations of the above steps can be found in the specific limitations of a medical image processing method described above.

[0167] In this embodiment, the organ image segmentation model is obtained by integrating and learning various organ image segmentation sub-models. That is, the organ image segmentation model is an ensemble model. Since the ensemble model has fewer model parameters, the training speed is faster during the training process based on the sample image set. The current ensemble strategy can be quickly fine-tuned to obtain the target ensemble strategy. This solves the problem in related technologies where the training speed of a single segmentation model is slow due to the large number of parameters in the segmentation model when training on the sample image set. In this way, the training efficiency of the model can be improved, and an organ image segmentation model that matches the segmentation style corresponding to the segmentation annotation image can be obtained quickly.

[0168] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0169] Based on the same inventive concept, this application also provides a medical image processing apparatus for implementing the aforementioned medical image processing method. The solution provided by this apparatus is similar to the implementation described in the above method; therefore, the specific limitations in the one or more medical image processing apparatus embodiments provided below can be found in the above-described limitations of the medical image processing method, and will not be repeated here.

[0170] In one embodiment, such as Figure 7 As shown, a medical image processing device is provided, including: an acquisition module 710, an input module 720, an adjustment module 730, and a combination module 740, wherein:

[0171] The acquisition module 710 is used to acquire an organ image segmentation model constructed with a tree structure; the organ image segmentation model includes a trained organ image segmentation sub-model corresponding to each leaf node in the tree structure; wherein, each organ image segmentation sub-model corresponds to a different segmentation style.

[0172] The input module 720 is used to input the sample image set into each of the organ image segmentation sub-models to obtain the predicted segmentation image corresponding to each sample image in the sample image set.

[0173] The adjustment module 730 is used to adjust the current ensemble strategy of the organ image segmentation model based on the segmentation annotation image corresponding to each sample image and the predicted segmentation image corresponding to each sample image, so as to obtain the target ensemble strategy.

[0174] The combination module 740 is used to combine the predicted segmentation results output by each organ image segmentation sub-model for the organ image to be segmented according to the target integration strategy, so as to obtain the target organ segmentation image corresponding to the organ image to be segmented; the segmentation style corresponding to the target organ segmentation image is matched with the segmentation style corresponding to the segmentation annotation image.

[0175] In one embodiment, the input module 720 is specifically used to input the sample image set into each of the organ image segmentation sub-models to obtain the predicted segmentation results output by each organ image segmentation sub-model for each of the sample images; and to combine the predicted segmentation results corresponding to each of the sample images according to the current integration strategy to obtain the predicted segmented image corresponding to each of the sample images.

[0176] In one embodiment, the current integration strategy includes segmentation result weights corresponding to each organ image segmentation sub-model; the adjustment module 730 is specifically used to determine the image difference between the segmentation annotation image corresponding to each sample image and the predicted segmentation image corresponding to each sample image; adjust the segmentation result weights corresponding to each organ image segmentation sub-model according to the image difference to obtain the target segmentation result weights corresponding to each organ image segmentation sub-model; and obtain the target integration strategy according to the target segmentation result weights corresponding to each organ image segmentation sub-model.

[0177] In one embodiment, the adjustment module 730 is specifically configured to: adjust the segmentation result weights corresponding to each organ image segmentation sub-model according to the image differences to obtain the adjusted weights corresponding to each organ image segmentation sub-model; combine the predicted segmentation results output by each organ image segmentation sub-model according to the adjusted weights corresponding to each organ image segmentation sub-model to obtain the adjusted predicted segmentation image corresponding to each sample image; use the adjusted predicted segmentation image as the predicted segmentation image, and return to the step of adjusting the segmentation result weights corresponding to each organ image segmentation sub-model according to the image differences, until the image differences meet the preset adjustment termination condition, and obtain the target segmentation result weights corresponding to each organ image segmentation sub-model.

[0178] In one embodiment, the acquisition module 710 is specifically used to acquire a training sample image set, a sample segmentation annotation image set, and an organ image segmentation sub-model to be trained in an initial segmentation model tree; wherein, the sample segmentation annotation image set includes sample segmentation annotation images corresponding to each training sample image in the training sample image set; the sample segmentation annotation images are obtained by annotating the actual region where the object to be segmented is located in the training sample image; the organ image segmentation sub-model to be trained in the initial segmentation model tree is trained according to the training sample image set and the sample segmentation annotation image set to obtain a target segmentation model tree constructed with the tree structure; the target segmentation model tree includes trained organ image segmentation sub-models corresponding to each leaf node in the tree structure; the training sample image set is input into each trained organ image segmentation sub-model corresponding to each leaf node to obtain the sample prediction segmentation result output by each organ image segmentation sub-model for the training sample image; and the organ image segmentation model is constructed according to the sample prediction segmentation result output by each organ image segmentation sub-model.

[0179] In one embodiment, the acquisition module 710 is specifically used to determine the sample prediction probability image in each of the sample prediction segmentation results; stitch together each of the sample prediction probability images to obtain a stitched sample prediction probability image; and construct a preset ensemble model on the stitched sample prediction probability image to obtain the organ image segmentation model.

[0180] In one embodiment, the initial segmentation model tree includes an organ image segmentation sub-model to be trained corresponding to the root node in the tree structure; the acquisition module 710 is specifically used to train the organ image segmentation sub-model to be trained corresponding to the root node according to the training sample image set and the sample segmentation annotation image set, to obtain the trained organ image segmentation sub-model corresponding to the root node; acquire texture feature images corresponding to each residual image; each residual image is obtained based on the difference between the root node prediction probability image corresponding to each training sample image and the sample segmentation annotation image corresponding to each training sample image; the root node prediction probability image is the sample prediction probability image in the sample prediction segmentation result output by the trained organ image segmentation sub-model corresponding to the root node for the training sample image; and obtain the target segmentation model tree according to the clustering result of each texture feature image.

[0181] In one embodiment, the acquisition module 710 is specifically configured to: divide the training sample image set into a first training sample image set and a second training sample image set according to the clustering results of each texture feature image; and split the first child node and the second child node of the root node from the root node; use the first child node as the root node and the first training sample image set as the training sample image set; and use the second child node as the root node and the second training sample image set as the training sample image set; execute the step of training the organ image segmentation sub-model to be trained corresponding to the root node according to the training sample image set and the sample segmentation annotation image set, until the number of the first training sample image set or the second training sample image set is less than a preset number, thereby obtaining the trained organ image segmentation sub-model corresponding to each node in the tree structure; and determine the target segmentation model tree according to the trained organ image segmentation sub-model corresponding to each node.

[0182] The various modules in the aforementioned medical image processing device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0183] In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 8 As shown, the computer device includes a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, NFC (Near Field Communication), or other technologies. When the computer program is executed by the processor, it implements a medical image processing method. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad located on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0184] Those skilled in the art will understand that Figure 8 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0185] In one embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.

[0186] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the steps in the above method embodiments.

[0187] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0188] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0189] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0190] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0191] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A medical image processing method, characterized in that, The method includes: Obtain an organ image segmentation model constructed with a tree structure; the organ image segmentation model includes trained organ image segmentation sub-models corresponding to each leaf node in the tree structure; wherein, each organ image segmentation sub-model corresponds to a different segmentation style; including: classifying different training sample images into different categories according to corresponding texture features to obtain a first training sample image set and a second training sample image set belonging to different categories, so as to train the organ image segmentation sub-model to be trained corresponding to the first child node of the root node in the tree structure and the organ image segmentation sub-model to be trained corresponding to the second child node of the root node in the tree structure respectively, and recursively constructing the trained organ image segmentation sub-models corresponding to each node in the tree structure; The sample image set is input into each of the organ image segmentation sub-models to obtain the predicted segmentation image corresponding to each sample image in the sample image set; Based on the segmentation annotation image corresponding to each sample image and the predicted segmentation image corresponding to each sample image, the current ensemble strategy of the organ image segmentation model is adjusted to obtain the target ensemble strategy. According to the target integration strategy, the predicted segmentation results output by each organ image segmentation sub-model for the organ image to be segmented are combined to obtain the target organ segmentation image corresponding to the organ image to be segmented; the segmentation style corresponding to the target organ segmentation image is matched with the segmentation style corresponding to the segmentation annotation image.

2. The method according to claim 1, characterized in that, The current ensemble strategy includes the segmentation result weights corresponding to each organ image segmentation sub-model; adjusting the current ensemble strategy of the organ image segmentation model based on the segmentation annotation image corresponding to each sample image and the predicted segmentation image corresponding to each sample image to obtain the target ensemble strategy includes: Determine the image differences between the segmentation annotation image corresponding to each sample image and the predicted segmentation image corresponding to each sample image; Based on the image differences, adjust the segmentation result weights corresponding to each organ image segmentation sub-model to obtain the target segmentation result weights corresponding to each organ image segmentation sub-model. The target ensemble strategy is obtained based on the target segmentation result weights corresponding to each organ image segmentation sub-model.

3. The method according to claim 2, characterized in that, The step of adjusting the segmentation result weights corresponding to each organ image segmentation sub-model based on the image differences to obtain the target segmentation result weights corresponding to each organ image segmentation sub-model includes: Based on the image differences, the segmentation result weights corresponding to each organ image segmentation sub-model are adjusted to obtain the adjusted weights corresponding to each organ image segmentation sub-model. Based on the adjusted weights corresponding to each organ image segmentation sub-model, the predicted segmentation results output by each organ image segmentation sub-model are combined to obtain the adjusted predicted segmentation image corresponding to each sample image. The adjusted predicted segmentation image is used as the predicted segmentation image. The step of adjusting the segmentation result weights corresponding to each organ image segmentation sub-model according to the image differences is returned until the image differences meet the preset adjustment end condition, and the target segmentation result weights corresponding to each organ image segmentation sub-model are obtained.

4. The method according to claim 1, characterized in that, The acquisition of the organ image segmentation model constructed with a tree structure includes: Obtain a training sample image set, a sample segmentation labeled image set, and an organ image segmentation sub-model to be trained in the initial segmentation model tree; wherein, the sample segmentation labeled image set includes sample segmentation labeled images corresponding to each training sample image in the training sample image set; the sample segmentation labeled images are obtained by labeling the actual region where the object to be segmented is located in the training sample image; Based on the training sample image set and the sample segmentation labeled image set, the organ image segmentation sub-model to be trained in the initial segmentation model tree is trained to obtain the target segmentation model tree constructed with the tree structure; the target segmentation model tree includes the trained organ image segmentation sub-model corresponding to each of the leaf nodes in the tree structure; The training sample image set is input into the trained organ image segmentation sub-model corresponding to each leaf node to obtain the sample prediction segmentation result output by each organ image segmentation sub-model for the training sample image; The organ image segmentation model is constructed based on the sample prediction segmentation results output by each organ image segmentation sub-model.

5. The method according to claim 4, characterized in that, The step of constructing the organ image segmentation model based on the sample prediction segmentation results output by each of the organ image segmentation sub-models includes: Determine the sample prediction probability image in each of the sample prediction segmentation results; By stitching together the predicted probability images of each sample, a stitched predicted probability image of the sample is obtained. A preset ensemble model is constructed on the stitched sample prediction probability image to obtain the organ image segmentation model.

6. The method according to claim 4, characterized in that, The initial segmentation model tree includes the organ image segmentation sub-model to be trained corresponding to the root node in the tree structure; the step of training the organ image segmentation sub-model to be trained in the initial segmentation model tree according to the training sample image set and the sample segmentation labeled image set to obtain the target segmentation model tree constructed with the tree structure includes: The organ image segmentation sub-model to be trained corresponding to the root node is trained based on the training sample image set and the sample segmentation labeled image set to obtain the trained organ image segmentation sub-model corresponding to the root node. Obtain the texture feature image corresponding to each residual image; each residual image is obtained based on the difference between the root node prediction probability image corresponding to each training sample image and the sample segmentation annotation image corresponding to each training sample image; the root node prediction probability image is the sample prediction probability image in the sample prediction segmentation result output by the trained organ image segmentation sub-model corresponding to the root node for the training sample image. The target segmentation model tree is obtained based on the clustering results of each texture feature image.

7. The method according to claim 6, characterized in that, The step of obtaining the target segmentation model tree based on the clustering results of each texture feature image includes: Based on the clustering results of each texture feature image, the training sample image set is divided into a first training sample image set and a second training sample image set, and the first child node and the second child node of the root node are split from the root node. The first child node is used as the root node, and the first training sample image set is used as the training sample image set; And, the second child node is used as the root node, and the second training sample image set is used as the training sample image set; The step of training the organ image segmentation sub-model corresponding to the root node based on the training sample image set and the sample segmentation labeled image set is performed until the number of the first training sample image set or the second training sample image set is less than a preset number, thereby obtaining the trained organ image segmentation sub-model corresponding to each node in the tree structure. The target segmentation model tree is determined based on the trained organ image segmentation sub-models corresponding to each node.

8. A medical image processing device, characterized in that, The device includes: The acquisition module is used to acquire an organ image segmentation model constructed in a tree structure; the organ image segmentation model includes a trained organ image segmentation sub-model corresponding to each leaf node in the tree structure; wherein, each organ image segmentation sub-model corresponds to a different segmentation style; The acquisition module is specifically used to divide different training sample images into different categories according to the corresponding texture features, and obtain a first training sample image set and a second training sample image set belonging to different categories, so as to train the organ image segmentation sub-model to be trained corresponding to the first child node of the root node in the tree structure and the organ image segmentation sub-model to be trained corresponding to the second child node of the root node, and recursively construct the trained organ image segmentation sub-model corresponding to each node in the tree structure. The input module is used to input the sample image set into each of the organ image segmentation sub-models to obtain the predicted segmentation image corresponding to each sample image in the sample image set; The adjustment module is used to adjust the current ensemble strategy of the organ image segmentation model based on the segmentation annotation image corresponding to each sample image and the predicted segmentation image corresponding to each sample image, so as to obtain the target ensemble strategy. The combination module is used to combine the predicted segmentation results output by each organ image segmentation sub-model for the organ image to be segmented according to the target integration strategy, so as to obtain the target organ segmentation image corresponding to the organ image to be segmented; the segmentation style corresponding to the target organ segmentation image is matched with the segmentation style corresponding to the segmentation annotation image.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images