A medical image segmentation method and system based on uncertainty knowledge modeling

Abstract

The present invention belongs to the field of image processing technology, and discloses a medical image segmentation method and system based on uncertainty knowledge modeling, which fuses the acquired image and label, extracts the image features of the fused image and the original image, obtains hidden features based on the image and the zero matrix, operates the image features and the hidden features, obtains two new image features and new hidden features; decodes the above features, obtains two confidences and calculates them, obtains two uncertainty maps, fuses the two uncertainty maps and the two new image features, obtains the fused feature map, performs segmentation operation on the fused feature map, obtains the segmentation result; and trains the model. The present invention encourages the current layer to reuse and re-explore the historical information of the previous layer, allows the deep layer to learn more comprehensive features, helps image segmentation with the help of uncertainty knowledge, improves the image segmentation accuracy, and improves the final segmentation result.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to a medical image segmentation method and system based on uncertainty knowledge modeling. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] In the field of computer vision, one of the main challenges is medical image segmentation, such as skin lesion segmentation in dermoscopic images, polyp segmentation in colonoscopy images, brain tumor segmentation in magnetic resonance images, and multi-organ segmentation in abdominal CT images. These segmentation results provide valuable information for detailed analysis of the anatomical regions required and greatly help doctors describe injuries, monitor disease progression, and assess the need for appropriate treatment. With the growing demand for intelligent medical image analysis, accurate and robust segmentation methods are becoming increasingly important.

[0004] To meet this demand, many deep learning methods with encoder and decoder structures have been proposed. Their encoders are usually used to extract image features, and the decoder is used to restore the extracted features to the original image size and output the final segmentation results. As a representative of encoder and decoder networks, U-Net has demonstrated its effectiveness in many medical segmentation tasks, which has inspired the development of a series of efficient segmentation networks, such as Attention U-Net, Res-UNet and DR-UNet.

[0005] Although these networks have been widely used in medical image segmentation, they still have a problem that cannot be ignored, that is, the segmentation effect is not ideal for lesions with blurred boundaries and irregular shapes. The main reasons for this result are: (1) The features extracted by the encoders of these networks are almost dominated by abstract semantic information, and a large amount of detailed content description is lost; (2) Although the skip connection in the decoder can fuse low-resolution and high-resolution information to improve feature learning, it is limited by the huge semantic gap between low-resolution and high-resolution features, resulting in blurred feature mapping and suboptimal segmentation results. As the network deepens, detailed information (such as boundaries, shapes, textures) becomes less and less, which is obviously not conducive to dense prediction tasks. Summary of the invention

[0006] In order to solve the above problems, the present invention proposes a medical image segmentation method and system based on uncertainty knowledge modeling, constructs a model based on uncertainty knowledge modeling, encourages the current layer to reuse and re-explore the historical information of the previous layer, allows deep learning of more comprehensive features, and uses uncertainty knowledge to assist image segmentation, improve image segmentation accuracy, and improve the final segmentation results.

[0007] In order to achieve the above object, the present invention adopts the following technical solution:

[0008] In a first aspect, the present invention provides a medical image segmentation method based on uncertainty knowledge modeling, comprising the following steps:

[0009] Get the image to be segmented and its true segmentation label;

[0010] The image to be segmented and the true segmentation label are fused to obtain a first image, and features are extracted from the first image and the original image to obtain first image features and second image features, and hidden features are obtained based on the first image, the original image, and a zero matrix of the same size as the image to be segmented, and the first image features and the second image features are fused with the hidden features and then separated to obtain third image features, fourth image features, and new hidden features;

[0011] The third image feature, the fourth image feature, and the new hidden feature are decoded to obtain the confidence of the third image feature and the confidence of the fourth image feature, the two confidences are calculated by Shannon entropy to obtain two uncertainty maps focusing on different areas, the two uncertainty maps, the third image feature, and the fourth image feature are fused to obtain a fused feature map, and the fused feature map is segmented to obtain a segmentation result;

[0012] Define the loss function, optimize the model parameters, and obtain a trained medical image segmentation model.

[0013] As an optional implementation, based on the first image, the original image, and a zero matrix of the same size as the image to be segmented, a hidden feature is obtained, specifically: the first image and the zero matrix and the original image and the zero matrix are spliced, respectively, and the spliced ​​image features are extracted, and a copy of the spliced ​​image feature map is connected to the spliced ​​image feature through a jump connection to generate a new image feature. At the same time, another copy of the spliced ​​image feature is compressed through a shared 1×1 convolution to compress its feature channel, and then summed with the zero matrix, and finally a pre-activation unit is used to generate a hidden feature.

[0014] In a second aspect, the present invention provides a medical image segmentation system based on uncertainty knowledge modeling, comprising:

[0015] The data acquisition module is configured to: acquire the image to be segmented and its true segmentation label;

[0016] The encoder module is configured to: fuse the image to be segmented and the true segmentation label to obtain a first image, extract features from the first image and the original image respectively to obtain first image features and second image features, obtain hidden features based on the first image, the original image and a zero matrix of the same size as the image to be segmented, fuse the first image features and the second image features with the hidden features respectively and then split them to obtain third image features, fourth image features and new hidden features;

[0017] The decoder module is configured to: perform a decoding operation on the third image feature, the fourth image feature, and the new hidden feature to obtain a confidence of the third image feature and a confidence of the fourth image feature, perform Shannon entropy calculation on the two confidences to obtain two uncertainty maps focusing on different areas, perform a fusion operation on the two uncertainty maps, the third image feature, and the fourth image feature to obtain a fused feature map, and perform a segmentation operation on the fused feature map to obtain a segmentation result;

[0018] The model training module is configured to: define the loss function, optimize the model parameters, and obtain a trained medical image segmentation model.

[0019] As an optional implementation, the encoder includes an encoder information interaction module and a global information fusion model, and the decoder includes a decoder information interaction module and an uncertainty fusion module.

[0020] As an optional implementation, the encoder includes four encoder information interaction modules, the encoder information interaction module includes a residual module and an attention module, the global information fusion module is used to integrate image features and hidden features, the decoder includes four decoder information interaction modules, the decoder information interaction module includes a residual module and an attention module, and the uncertainty fusion module is used to fuse the feature map with the uncertainty map to obtain a fused feature map.

[0021] As an optional implementation, the path for processing image features is the image path, and the path for processing hidden features is the hidden state path. The number of feature channels on the image path changes with the change of the network stage, while the number of channels on the hidden state path remains unchanged.

[0022] As an alternative implementation, there is a 1×1 convolution at each stage in the hidden state path, with the same weights shared between layers.

[0023] In a third aspect, the present invention provides an electronic device comprising a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the method described in the first aspect is performed.

[0024] In a fourth aspect, the present invention provides a computer-readable storage medium for storing computer instructions, wherein when the computer instructions are executed by a processor, the method described in the first aspect is performed.

[0025] In a fifth aspect, the present invention provides a computer program product, comprising a computer program, which, when executed by a processor, implements the method described in the first aspect.

[0026] Compared with the prior art, the present invention has the following beneficial effects:

[0027] The present invention proposes a medical image segmentation method and system based on uncertainty knowledge modeling, which has a working principle similar to that of unfolding RNN and has the advantage of modeling the temporal order relationship between different network layers. Specifically, the present invention develops an information interaction module that enables the current layer to effectively reuse and re-explore the historical information of the previous layer, allowing deep learning of more comprehensive features containing low-level detail descriptions and high-level semantic abstractions. In addition, the present invention develops a global information fusion module to sensitively enhance the information extracted by the three-path encoder. The algorithm combines the advantages of local and global operations, can flexibly learn discriminative representations in a wider frequency range, and more effectively bridges the encoder and decoder. The segmentation method of the present invention has carried out a large number of experiments on four challenging tasks such as skin lesions, polyps, brain tumors, and abdominal multi-organ segmentation, which consistently proves that the model proposed by the present invention has excellent segmentation performance and generalization ability.

[0028] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0030] Figure 1 A flowchart of a medical image segmentation method based on uncertainty knowledge modeling provided in Example 1 of the present invention. DETAILED DESCRIPTION

[0031] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0032] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0033] It should be noted that the terms used herein are only for describing specific embodiments, and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that the terms "include" and "have" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0034] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0035] Example 1

[0036] like Figure 1 As shown, this embodiment provides a medical image segmentation method based on uncertainty knowledge modeling, comprising the following steps:

[0037] S1 obtains the image to be segmented and its true segmentation label;

[0038] S2 fuses the image to be segmented and the true segmentation label using a Bézier curve to obtain a first image, extracts features from the first image and the original image respectively, obtains first image features and second image features, obtains hidden features based on the first image, the original image and a zero matrix of the same size as the image to be segmented, fuses the first image features and the second image features with the hidden features respectively and then splits them to obtain third image features, fourth image features and new hidden features;

[0039] S3 performs a decoding operation on the third image feature, the fourth image feature, and the new hidden feature, and outputs the probability value after passing the feature output of the last layer of the decoding network through softmax to represent the confidence of the feature map, and obtains the confidence of the third image feature and the confidence of the fourth image feature, and performs Shannon entropy calculation on the two confidences, calculates the Shannon entropy of each pixel, and obtains two uncertainty maps focusing on different areas, and fuses the two uncertainty maps, the third image feature, and the fourth image feature to obtain a fused feature map, and performs a segmentation operation on the fused feature map to obtain a segmentation result;

[0040] S4 defines the loss function, optimizes the model parameters, and obtains a trained medical image segmentation model.

[0041] First, the present invention uses the public dataset ISIC2018 to train and verify the segmentation model.

[0042] In the embodiment of the present invention, the path for processing image features is called an image path, and the path for processing hidden features is called a hidden state path.

[0043] Three channels are used to extract two copies of image feature information and one copy of hidden features. In this step, there are three channels processing information, two paths are dedicated to image feature information like traditional U-Net, and the other path uses convolution kernels shared along the depth to hide state information. This structure allows the model to work similarly to an unfolded RNN and enjoy its advantages of modeling time series information and utilizing historical information. The model encourages the reuse and re-exploration of historical information through multi-faceted information interaction between the above three paths.

[0044] The method of the present invention comprises two parts, namely, an encoder and a decoder. The encoder comprises an encoder information interaction module and a global information fusion model, and the decoder comprises a decoder information interaction module and an uncertainty fusion module.

[0045] The encoder includes four encoder information interaction modules, which include a residual module and an attention module. The global information fusion module is used to integrate image features and hidden features. The decoder includes four decoder information interaction modules, which include a residual module and an attention module. The uncertainty fusion module is used to fuse the feature map with the uncertainty map to obtain a fused feature map.

[0046] Among them, the encoder information interaction module combines the input of the image path and the input of the state path by concatenation, and then passes it to the residual module and the attention module to extract discriminative new features. Then, a copy of this new feature map is connected to the old image feature (i.e., the input of the image path) through a jump connection to generate a new image feature. At the same time, another copy of this new feature compresses its feature channel through a shared 1×1 convolution, and then sums it with the old state feature (the input of the state path, i.e., the hidden feature of the previous layer), and finally passes through a pre-activation unit, which includes a non-shared batch normalization and a tanh activation unit to generate a new state feature, i.e., a new hidden feature.

[0047] The proposed encoder (ResNet-34) consists of four hierarchical stages with 3, 4, 6, and 3 residual blocks respectively.

[0048] That is, firstly, one information interaction module is used to reduce the feature size, and then multiple information interaction modules are used to obtain a deeper feature representation.

[0049] It is worth noting that 1×1 convolutions with the same weights are shared among layers within a stage. Due to this design, the model works similarly to an unrolled RNN with deep CNN layers as input, allowing the current layer to flexibly learn the historical information of the previous layer. This is the main difference between the proposed information interaction module and other feature fusion modules.

[0050] Image feature information and hidden feature information are processed based on the global information fusion module.

[0051] The spatial pyramid pooling and multi-scale natural convolution operations are embedded into the non-local block to form a global information fusion module.

[0052] For the input feature X∈ , where H, W, and C represent the spatial height, spatial width, and channel number, respectively. Two 1×1 convolutions are used and Transform X to a different embedding point ∈ and ∈ , among which =0.5Default.

[0053] An embedding point Flattened to size: N× ; Where: N = H × W, represents the total number of spatial positions. This flattened embedding is recorded as "query" (Q).

[0054] For another embedding algorithm, we use spatial pyramid pooling and multi-scale attribute convolution to capture its global and local context representations, respectively. Then, the captured representation is concatenated to a new embedding, recorded as “key” (K). K∈ ,in Indicates the number of points sampled by the spatial pyramid pooling operation.

[0055] We copy the "key" and record it as "value" (V). Note that the output calculation formula is:

[0056] Attention(Q,K,V)= Softmax( )V

[0057] Dim is the dimension for which the attention mechanism is required, and the dimension used here is K.

[0058] We reshape the attention we receive into , and then use a 1×1 convolution Expand its number of channels to C. Finally, we get the following output:

[0059] = (Reshape(Attention(Q,K,V)))+X

[0060] The information interaction module of the decoder is used to further extract features.

[0061] The information interaction module of the decoder increases the feature size through deconvolution. The information interaction module of the decoder is based on the information interaction module of the encoder, with the difference of residual blocks and jumper connections.

[0062] The number of the four residual blocks of the decoder is 1, and the connection method is deconvolution.

[0063] The uncertainty map is calculated using Shannon entropy and fused with the feature map in the uncertainty fusion module.

[0064] To estimate the uncertainty U, we use the Shannon entropy , where Pc is the softmax activation function of the segmenter for each segment c to obtain logits (i.e., posterior probabilities).

[0065] This method can obtain the uncertainty map of two feature maps from the trained decoder.

[0066] By considering U 1 and U 2 The heterogeneous aggregation between them fully fuses these maps to utilize these uncertainty maps as guidance to induce enhancement of confused regions.

[0067] Therefore, the method to create UG Umap is as follows

[0068]

[0069] Where Gaussian Blur represents a smoothing function that applies a Gaussian filter to the uncertainty map.

[0070] The first term in the equation highlights regions where either enhancement process has identified substantial blur, providing guidance for prioritizing these regions in subsequent enhancements. The second term in the equation provides an overall view that is consistent with the uncertainty of both methods while smoothing out extreme values ​​to prevent over-emphasizing differences.

[0071] With these advantages, the augmented image Xma is generated using the original image X1 and the first image X2 fused with the true label by utilizing a variant of linear interpolation.

[0072] Specifically, we multiply Umap and its inverse by X1 and X2 as follows:

[0073]

[0074] Where Pu represents the constraint probability, which is the average of the uncertainty probabilities of the foreground corresponding areas from Umap.

[0075] Use the segmentation model to segment the final Xma and obtain the segmentation result.

[0076] Use U-net as the final segmentation model, which accepts Xma as input;

[0077] During the training process, the decoder, encoder and parameters here are all learnable and the parameters are optimized together.

[0078] Calculate the loss (such as cross entropy loss) on X1, X2 and the final segmentation result respectively, and optimize the parameters three times.

[0079] Table 1 is a simulation experiment of the method of the present invention, using the public dataset ISIC2018. The evaluation index is Dice, which is used to evaluate the overall performance of segmentation. IOU is used to evaluate the accuracy of the model in object segmentation. The method of the present invention improves the performance of medical image segmentation.

[0080] Table 1 Comparison of the accuracy of this disclosure and other algorithms

[0081]

[0082] The present invention proposes a medical image segmentation method based on uncertainty knowledge modeling, which works similarly to unfolding RNN and has the advantage of modeling the temporal order relationship between different network layers. Specifically, the present invention develops an information interaction module that enables the current layer to effectively reuse and re-explore the historical information of the previous layer, allowing deep learning of more comprehensive features that contain low-level detail descriptions and high-level semantic abstractions. In addition, the present invention develops a global information fusion module to sensitively enhance the information extracted by the three-path encoder. The algorithm combines the advantages of local and global operations. Due to these innovative designs, a large number of experiments have been carried out on four challenging tasks, including skin lesions, polyps, brain tumors, and abdominal multi-organ segmentation, which consistently demonstrate that the proposed model has excellent segmentation performance and generalization ability.

[0083] Example 2

[0084] This embodiment provides a medical image segmentation system based on uncertainty knowledge modeling, including:

[0085] The data acquisition module is configured to: acquire the image to be segmented and its true segmentation label;

[0086] The encoder module is configured to: fuse the image to be segmented and the true segmentation label to obtain a first image, extract features from the first image and the original image respectively to obtain first image features and second image features, obtain hidden features based on the first image, the original image and a zero matrix of the same size as the image to be segmented, fuse the first image features and the second image features with the hidden features respectively and then split them to obtain third image features, fourth image features and new hidden features;

[0087] The decoder module is configured to: perform a decoding operation on the third image feature, the fourth image feature, and the new hidden feature to obtain a confidence of the third image feature and a confidence of the fourth image feature, perform Shannon entropy calculation on the two confidences to obtain two uncertainty maps focusing on different areas, perform a fusion operation on the two uncertainty maps, the third image feature, and the fourth image feature to obtain a fused feature map, and perform a segmentation operation on the fused feature map to obtain a segmentation result;

[0088] The model training module is configured to: define the loss function, optimize the model parameters, and obtain a trained medical image segmentation model.

[0089] It should be noted that the above modules correspond to the steps described in Example 1, and the examples and application scenarios implemented by the above modules and the corresponding steps are the same, but are not limited to the contents disclosed in the above Example 1. It should be noted that the above modules, as part of the system, can be executed in a computer system such as a set of computer executable instructions.

[0090] In further embodiments, there is also provided:

[0091] An electronic device includes a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the method described in Embodiment 1 is performed. For the sake of brevity, it will not be described in detail here.

[0092] It should be understood that in this embodiment, the processor may be a central processing unit CPU, and the processor may also be other general-purpose processors, digital signal processors DSP, application-specific integrated circuits ASIC, off-the-shelf programmable gate arrays FPGA or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0093] The memory may include a read-only memory and a random access memory, and provide instructions and data to the processor. A portion of the memory may also include a non-volatile random access memory. For example, the memory may also store information about the device type.

[0094] A computer-readable storage medium is used to store computer instructions, and when the computer instructions are executed by a processor, the method described in Example 1 is completed.

[0095] The method in Example 1 can be directly embodied as a hardware processor, or a combination of hardware and software modules in the processor. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware. To avoid repetition, it is not described in detail here.

[0096] A computer program product includes a computer program, and when the computer program is executed by a processor, the method described in embodiment 1 is implemented.

[0097] The present invention also provides at least one computer program product tangibly stored on a non-transitory computer-readable storage medium. The computer program product includes computer executable instructions, such as instructions included in a program module, which are executed in a device on a real or virtual processor of the target to perform the process / method as described above. Typically, program modules include routines, programs, libraries, objects, classes, components, data structures, etc. that perform specific tasks or implement specific abstract data types. In various embodiments, the functions of program modules can be combined or divided between program modules as needed. Machine executable instructions for program modules can be executed in local or distributed devices. In distributed devices, program modules can be located in local and remote storage media.

[0098] The computer program code for implementing the method of the present invention can be written in one or more programming languages. These computer program codes can be provided to the processor of a general-purpose computer, a special-purpose computer or other programmable data processing device, so that the program code, when executed by the computer or other programmable data processing device, causes the function / operation specified in the flow chart and / or block diagram to be implemented. The program code can be executed completely on a computer, partially on a computer, as an independent software package, partially on a computer and partially on a remote computer or completely on a remote computer or server.

[0099] In the context of the present invention, computer program codes or related data may be carried by any appropriate carrier to enable a device, apparatus or processor to perform the various processes and operations described above. Examples of carriers include signals, computer readable media, and the like. Examples of signals may include electrical, optical, radio, acoustic or other forms of propagation signals, such as carrier waves, infrared signals, and the like.

[0100] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with this embodiment can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0101] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. A medical image segmentation method based on uncertainty knowledge modeling, characterized in that: The following steps are involved: Get the image to be segmented and its true segmentation label; The image to be segmented and the true segmentation label are fused to obtain a first image, and features are extracted from the first image and the original image to obtain first image features and second image features, and hidden features are obtained based on the first image, the original image, and a zero matrix of the same size as the image to be segmented, and the first image features and the second image features are fused with the hidden features and then separated to obtain third image features, fourth image features, and new hidden features; The third image feature, the fourth image feature, and the new hidden feature are decoded to obtain the confidence of the third image feature and the confidence of the fourth image feature, the two confidences are calculated by Shannon entropy to obtain two uncertainty maps focusing on different areas, the two uncertainty maps, the third image feature, and the fourth image feature are fused to obtain a fused feature map, and the fused feature map is segmented to obtain a segmentation result; Define the loss function, optimize the model parameters, and obtain a trained medical image segmentation model.

2. A medical image segmentation method based on uncertainty knowledge modeling as claimed in claim 1, characterized in that: Based on the first image, the original image and a zero matrix of the same size as the image to be segmented, a hidden feature is obtained, specifically: the first image and the zero matrix and the original image and the zero matrix are concatenated respectively, the features of the spliced ​​image are extracted, a copy of the spliced ​​image feature map is connected to the spliced ​​image feature through a jump connection to generate a new image feature, and at the same time, another copy of the spliced ​​image feature is compressed through a shared 1×1 convolution to compress its feature channel, and then summed with the zero matrix, and finally a hidden feature is generated through a pre-activation unit.

3. A medical image segmentation system based on uncertainty knowledge modeling, characterized in that: include: The data acquisition module is configured to: acquire the image to be segmented and its true segmentation label; The encoder module is configured to: fuse the image to be segmented and the true segmentation label to obtain a first image, extract features from the first image and the original image respectively to obtain first image features and second image features, obtain hidden features based on the first image, the original image and a zero matrix of the same size as the image to be segmented, fuse the first image features and the second image features with the hidden features respectively and then split them to obtain third image features, fourth image features and new hidden features; The decoder module is configured to: perform a decoding operation on the third image feature, the fourth image feature, and the new hidden feature to obtain a confidence of the third image feature and a confidence of the fourth image feature, perform Shannon entropy calculation on the two confidences to obtain two uncertainty maps focusing on different areas, perform a fusion operation on the two uncertainty maps, the third image feature, and the fourth image feature to obtain a fused feature map, and perform a segmentation operation on the fused feature map to obtain a segmentation result; The model training module is configured to: define the loss function, optimize the model parameters, and obtain a trained medical image segmentation model.

4. A medical image segmentation system based on uncertainty knowledge modeling as claimed in claim 3, characterized in that: The encoder includes an encoder information interaction module and a global information fusion module, and the decoder includes a decoder information interaction module and an uncertainty fusion module.

5. A medical image segmentation system based on uncertainty knowledge modeling as claimed in claim 4, characterized in that: The encoder includes four encoder information interaction modules, each of which includes a residual module and an attention module. The global information fusion module is used to integrate image features and hidden features. The decoder includes four decoder information interaction modules, each of which includes a residual module and an attention module. The uncertainty fusion module is used to fuse the feature map with the uncertainty map to obtain a fused feature map.

6. A medical image segmentation system based on uncertainty knowledge modeling as claimed in claim 3, characterized in that: The path for processing image features is the image path, and the path for processing hidden features is the hidden state path. The number of feature channels on the image path changes with the change of the network stage, while the number of channels on the hidden state path remains unchanged.

7. A medical image segmentation system based on uncertainty knowledge modeling as claimed in claim 6, characterized in that: There is a 1×1 convolution with the same weights shared between layers at each stage in the hidden state path.

8. An electronic device, characterized in that: The method comprises a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the method according to any one of claims 1 to 2 is completed.

9. A computer-readable storage medium, characterized in that: Used to store computer instructions, which, when executed by a processor, complete the method described in any one of claims 1-2.

10. A computer program product, characterized in that The invention comprises a computer program, which, when executed by a processor, implements the method according to any one of claims 1 to 2.

Images