Medical image organ segmentation method, electronic equipment and storage medium

Abstract

The invention relates to the technical field of medical image processing, and particularly provides a medical image organ segmentation method, electronic equipment and a storage medium, and the method comprises the steps: obtaining a to-be-segmented medical image; inputting a to-be-segmented medical image into the trained organ segmentation model to obtain an organ segmentation result of the to-be-segmented medical image; wherein the organ segmentation model comprises an encoder and a decoder; the encoder comprises a plurality of encoding layers which are connected in sequence, and each encoding layer comprises a dynamic deformable convolution module and a down-sampling module. The decoder comprises a plurality of decoding layers connected in sequence, and each decoding layer comprises a dynamic deformable convolution module, an up-sampling module and an attention mechanism module; the coding layer and the corresponding decoding layer are connected through a jump connection module; the dynamic deformable convolution and the attention mechanism are combined, so that the network can better adapt to different input data, the features of the input data are carefully captured, and the organ segmentation precision and efficiency are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image processing, and in particular to a medical image organ segmentation method, electronic equipment and storage medium. Background Art

[0002] Organ segmentation is the basis of disease diagnosis, cancer treatment and radiotherapy planning. Accurate segmentation results can provide valuable information for clinical diagnosis and follow-up, such as organ size, position, boundary state, spatial relationship of multiple organs, etc. In the field of medical imaging, the abdomen contains multiple important organs, such as liver, spleen, kidney, pancreas, stomach, intestine, bladder, etc. The morphology, structure and position of these organs play a key role in disease diagnosis, treatment plan formulation and evaluation of treatment effects. Traditional organ segmentation mainly relies on doctors to perform manually, which is a very time-consuming and experience-dependent task. Doctors need to identify the boundaries of organs frame by frame on medical images (such as computed tomography and magnetic resonance imaging, etc.) and draw them. Manual organ segmentation is not only inefficient, but also the organ segmentation results between different doctors may be different, affecting the accuracy of diagnosis and treatment.

[0003] With the development of medical imaging technology and computer technology, automatic or semi-automatic organ segmentation methods have gradually become a research hotspot, including atlas-based methods, neural network-based methods, etc.; when there are only a small number of annotated images, atlas-based methods are usually used. However, this method relies on image registration technology and may lead to incorrect organ segmentation if the organ is occupied by a tumor; neural networks use computer algorithms to analyze medical images, automatically or assist doctors in identifying and outlining organs, and improve work efficiency and accuracy; however, for organs with large abdominal deformation, such as the stomach, intestines, bladder, etc., existing segmentation methods often find it difficult to accurately segment the boundaries of the organs, affecting the accuracy of diagnosis and treatment.

[0004] Accordingly, the art needs a new medical image organ segmentation solution to solve the above problems. Summary of the invention

[0005] In order to overcome the above defects, the present invention is proposed to solve or at least partially solve the technical problem that the existing organ segmentation method is difficult to accurately segment the boundaries of the organ and has poor segmentation accuracy.

[0006] In a first aspect, a method for segmenting organs in a medical image is provided, the method comprising:

[0007] Acquire a medical image to be segmented;

[0008] Inputting the medical image to be segmented into a trained organ segmentation model to obtain an organ segmentation result of the medical image to be segmented;

[0009] Among them, the organ segmentation model includes an encoder and a decoder; the encoder includes multiple encoding layers connected in sequence, and the encoding layer includes a dynamic deformable convolution module and a downsampling module; the decoder includes multiple decoding layers connected in sequence, and the decoding layer includes a dynamic deformable convolution module, an upsampling module and an attention mechanism module; the encoding layer and the corresponding decoding layer are connected through a jump connection module.

[0010] In a technical solution of the above-mentioned medical image organ segmentation method, the step of inputting the medical image to be segmented into a trained organ segmentation model to obtain the organ segmentation result of the medical image to be segmented includes:

[0011] Inputting the medical image to be segmented into the encoder module to obtain an encoded visual feature image of the medical image to be segmented;

[0012] Based on the jump connection module, feature fusion is performed on the encoded visual feature image, and based on the decoder module, an organ segmentation result of the medical image to be segmented is obtained.

[0013] In a technical solution of the above-mentioned medical image organ segmentation method, the step of inputting the medical image to be segmented into the encoder module to obtain the encoded visual feature image of the medical image to be segmented comprises:

[0014] The medical image to be segmented is processed step by step based on multiple coding layers to obtain a coded visual feature image; wherein the medical image to be segmented is sequentially input into the dynamic deformable convolution module and the downsampling module for convolution and dimension reduction respectively, and a coded feature image is output after each passing through the dynamic deformable convolution module and the downsampling module.

[0015] In a technical solution of the above-mentioned medical image organ segmentation method, the organ segmentation model further includes a 1×1 convolutional layer;

[0016] The step of performing feature fusion on the encoded visual feature image based on the skip connection module, and obtaining an organ segmentation result of the medical image to be segmented based on the decoder module, comprises:

[0017] The encoded visual feature image is processed step by step based on multiple decoding layers to obtain the organ segmentation result of the medical image to be segmented; wherein the encoded visual feature image is sequentially input into the dynamic deformable convolution module, the upsampling module and the attention mechanism module for convolution, dimension restoration and attention enhancement respectively, and a decoded feature image is output after each of the dynamic deformable convolution module, the upsampling module and the attention mechanism module; and based on the jump connection module, the encoded feature image of the corresponding layer is feature fused with the decoded feature image;

[0018] The decoded feature image finally outputted is processed based on a 1×1 convolutional layer to obtain an organ segmentation result of the medical image to be segmented.

[0019] In a technical solution of the above-mentioned medical image organ segmentation method, the attention mechanism module includes a global average pooling layer and a one-dimensional convolution layer;

[0020] Converting the decoded feature image into a one-dimensional feature vector based on the global average pooling layer;

[0021] Converting the one-dimensional feature vector into a channel attention vector based on the one-dimensional convolutional layer;

[0022] Multiply the channel attention vector by the decoded feature image to obtain a channel attention feature map.

[0023] In a technical solution of the above-mentioned medical image organ segmentation method, the dynamic deformable convolution module adopts a dynamic convolution kernel;

[0024] The method for obtaining the dynamic convolution kernel includes:

[0025] Preset an attention module of a dynamic convolution kernel, and preset k convolution kernels with the same scale and number of channels, wherein the attention module of the dynamic convolution kernel includes a global average pooling layer, a fully connected layer, and a normalization layer;

[0026] Input k convolution kernels into the global average pooling layer to obtain global spatial features, then obtain k dimensions through the fully connected layer, and input the k dimensions into the normalization layer to obtain k attention weights corresponding to the k convolution kernels;

[0027] The k convolution kernels are weighted based on the k attention weights to obtain the dynamic convolution kernel.

[0028] In a technical solution of the above-mentioned medical image organ segmentation method, the method further comprises training the organ segmentation model according to the following steps:

[0029] Label historical medical images to obtain real organ segmentation images;

[0030] Inputting the historical medical image into the organ segmentation model to be trained to obtain a predicted organ segmentation image;

[0031] Obtaining a loss function based on the real organ segmentation image and the predicted organ segmentation image;

[0032] The gradient of the parameters of each module in the organ segmentation model to be trained is obtained based on the loss function and the optimization algorithm, and the parameters of each module in the organ segmentation model to be trained are updated based on the gradient of the parameters of each module.

[0033] In a technical solution of the above-mentioned medical image organ segmentation method, after the medical image to be segmented is acquired, one or more preprocessing operations including normalization, cropping, and data enhancement are performed on the medical image to be segmented.

[0034] In a second aspect, an electronic device is provided, comprising at least one processor; and a memory communicatively connected to the at least one processor; wherein a computer program is stored in the memory, and when the computer program is executed by the at least one processor, the method described in any one of the technical solutions of the above-mentioned medical image organ segmentation method is implemented.

[0035] In a third aspect, a computer-readable storage medium is provided, wherein a plurality of program codes are stored therein, wherein the program codes are suitable for being loaded and run by a processor to execute the method described in any one of the technical solutions of the above-mentioned medical image organ segmentation method.

[0036] The above one or more technical solutions of the present invention have at least one or more of the following beneficial effects:

[0037] In the implementation of the technical solution of the medical image organ segmentation method provided by the present invention, a medical image to be segmented is obtained; the medical image to be segmented is input into a trained organ segmentation model to obtain an organ segmentation result of the medical image to be segmented; wherein the organ segmentation model includes an encoder and a decoder; the encoder includes a plurality of encoding layers connected in sequence, the encoding layer includes a dynamic deformable convolution module and a downsampling module; the decoder includes a plurality of decoding layers connected in sequence, the decoding layer includes a dynamic deformable convolution module, an upsampling module and an attention mechanism module; the encoding layer and the corresponding decoding layer are connected by a jump connection module; the present invention combines dynamic deformable convolution and attention mechanism, so that the network can better adapt to different input data, more finely capture the characteristics of the input data, and improve the accuracy and efficiency of organ segmentation. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] The disclosure of the present invention will become more easily understood with reference to the accompanying drawings. It is easy for those skilled in the art to understand that these drawings are only for illustrative purposes and are not intended to limit the scope of protection of the present invention. Among them:

[0039] Figure 1 is a schematic flow chart of main steps of a method for segmenting organs in medical images according to an embodiment of the present invention;

[0040] Figure 2 is a flowchart of main steps of a method for training an organ segmentation model according to an embodiment of the present invention;

[0041] Figure 3 is a schematic flow chart of main steps for obtaining organ segmentation results of medical images according to one embodiment of the present invention;

[0042] Figure 4 is a schematic diagram of a process of segmenting a medical image to be segmented based on an organ segmentation model according to an embodiment of the present invention;

[0043] Figure 5 1 is a schematic flow chart of main steps of a method for obtaining a dynamic convolution kernel according to an embodiment of the present invention;

[0044] Figure 6 It is a schematic diagram of the main structure of an electronic device according to an embodiment of the present invention.

[0045] Reference numerals:

[0046] 61: memory; 62: processor. DETAILED DESCRIPTION

[0047] Some embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the protection scope of the present invention.

[0048] In the description of the present invention, "module" and "processor" may include hardware, software or a combination of the two. A module may include hardware circuits, various suitable sensors, communication ports, and memories, and may also include software parts, such as program codes, or a combination of software and hardware. The processor may be a central processing unit, a microprocessor, an image processor, a digital signal processor, or any other suitable processor. The processor has data and / or signal processing functions. The processor may be implemented in software, hardware, or a combination of the two. Computer-readable storage media include any suitable medium that can store program codes, such as a magnetic disk, a hard disk, an optical disk, a flash memory, a read-only memory, a random access memory, and the like. The term "A and / or B" means all possible combinations of A and B, such as only A, only B, or A and B. The term "at least one A or B" or "at least one of A and B" has a similar meaning to "A and / or B", and may include only A, only B, or A and B. The singular terms "one" and "the" may also include plural forms.

[0049] The relevant user personal information that may be involved in the various embodiments of the present invention is strictly in accordance with the requirements of laws and regulations, following the principles of legality, legitimacy and necessity, based on the reasonable purposes of business scenarios, to process the personal information that users actively provide during the use of products / services or generated due to the use of products / services, as well as the personal information obtained with the user's authorization.

[0050] The user personal information processed by the present invention will vary depending on the specific product / service scenario, and shall be based on the specific scenario in which the user uses the product / service, and may involve the user's account information, device information, driving information, vehicle information or other related information. The present invention will treat the user's personal information and its processing with a high degree of diligence.

[0051] The present invention attaches great importance to the security of user personal information and has adopted reasonable and feasible security protection measures that meet industry standards to protect user information and prevent personal information from being accessed, disclosed, used, modified, damaged or lost without authorization.

[0052] See attached Figure 1 , Figure 1 FIG. 1 is a flow chart showing the main steps of a method for segmenting medical image organs according to an embodiment of the present invention. Figure 1 As shown, the medical image organ segmentation method in the embodiment of the present invention mainly includes the following steps S101 to S102.

[0053] Step S101: obtaining a medical image to be segmented;

[0054] In this embodiment, the medical images to be segmented include original medical images acquired by imaging technologies such as CT (Computed Tomography) and MR (Magnetic Resonance Imaging).

[0055] In this embodiment, after the medical image to be segmented is acquired, one or more preprocessing operations including normalization, cropping, and data enhancement are performed on the medical image to be segmented.

[0056] In one embodiment, after obtaining the medical image to be segmented, the medical image to be segmented is preprocessed, including preprocessing operations such as normalization, cropping, and data enhancement, which is conducive to the subsequent iterative training of the model. In the data enhancement, a random radiation transformation operation and a random noise addition operation are used to improve the segmentation accuracy and robustness of the model. It should be noted that the use of preprocessing operations including normalization, cropping, data enhancement, etc. for the medical image to be segmented and the use of random radiation transformation operations and random noise addition operations in data enhancement are only an example of an implementation method and do not constitute a limitation on the implementation methods of the present invention.

[0057] Step S102: inputting the medical image to be segmented into the trained organ segmentation model to obtain the organ segmentation result of the medical image to be segmented;

[0058] In this embodiment, the organ segmentation model includes an encoder and a decoder; the encoder includes multiple encoding layers connected in sequence, and the encoding layer includes a dynamic deformable convolution module and a downsampling module; the decoder includes multiple decoding layers connected in sequence, and the decoding layer includes a dynamic deformable convolution module, an upsampling module and an attention mechanism module; the encoding layer and the corresponding decoding layer are connected through a jump connection module.

[0059] In this embodiment, see the attached Figure 2 , Figure 2 FIG. 1 is a flow chart showing the main steps of a training method for an organ segmentation model according to an embodiment of the present invention. Figure 2 As shown, the training method of the organ segmentation model in the embodiment of the present invention mainly includes the following steps S201 to S205.

[0060] Step S201: annotating historical medical images to obtain real organ segmentation images;

[0061] In one embodiment, the real organ segmentation image includes labeling boxes of multiple organs, which are used to indicate the real positions of the organs; the historical medical image is input into professional medical imaging software for rough labeling, and calibration, verification and modification are performed based on the rough labeling to obtain the real organ segmentation image.

[0062] Optionally, 3D Slicer (3D printing slicer) software is used as professional medical imaging software for annotation. It should be noted that using 3D Slicer as professional medical imaging software for annotation is only an example of implementation and does not limit the implementation of the present invention. In actual applications, different professional medical imaging software can be selected for annotation.

[0063] Step S202: inputting the historical medical image into the organ segmentation model to be trained to obtain a predicted organ segmentation image;

[0064] In one embodiment, a historical medical image is input into an encoder module to obtain an encoded visual feature image of the historical medical image; based on a jump connection module, feature fusion is performed on the encoded visual feature image of the historical medical image, and based on a decoder module, a predicted organ segmentation image is obtained.

[0065] Step S203: obtaining a loss function based on the real organ segmentation image and the predicted organ segmentation image;

[0066] Optionally, a combined loss function of Dice loss function and cross entropy loss is selected as the loss function of the model; by calculating the combined loss function of Dice loss function and cross entropy loss, the difference between the real organ segmentation image and the predicted organ segmentation image is measured, and the smaller the value of the combined loss function, the closer the predicted organ segmentation image is to the real organ segmentation image, and the better the performance of the organ segmentation model. It should be noted that the combined loss function of Dice loss function and cross entropy loss is selected here to train the model is only an example of an implementation method, and does not constitute a limitation on the implementation method of the present invention. In actual applications, different selections can be made as needed, as long as the model can be trained.

[0067] Step S204: Obtain the gradient of each module parameter in the organ segmentation model to be trained based on the loss function and the optimization algorithm, and update the parameters of each module in the organ segmentation model to be trained based on the gradient of each module parameter.

[0068] Optionally, select Adam (adaptive moment estimation) optimization algorithm or Adadelta optimization algorithm as the optimization algorithm of the model to update the parameters of the model; after each iterative training, calculate the gradient of each module parameter during the current iterative training through the loss function, and update the parameters of each module in the organ segmentation model based on the gradient of each module parameter; starting from the output layer, propagate the calculated gradient layer by layer along the reverse direction of the organ segmentation model to be trained until the input layer; during the propagation process, calculate the gradient of each layer of each module according to the chain rule, and use it to update the parameters of the layer; the gradient represents the degree of influence of a small change in the parameter on the value of the loss function, that is, in which direction the parameter should be adjusted to reduce the value of the loss function. It should be noted that the selection of the Adam optimization algorithm or the Adadelta optimization algorithm to update the parameters of each module in the model here is only an example of an implementation method, and does not constitute a limitation on the implementation method of the present invention. In practical applications, different selections can be made as needed, as long as the parameters of each module in the model can be updated.

[0069] Step S205: inputting the predicted organ segmentation image into the organ segmentation model to be trained after updating parameters, so as to obtain an updated predicted organ segmentation image and a loss function;

[0070] Step S206: iteratively train the organ segmentation model to be trained using the updated predicted organ segmentation image and loss function until the organ segmentation model to be trained converges to obtain a trained organ segmentation model.

[0071] In one embodiment, after the parameters are updated, the predicted organ segmentation image obtained from the previous training is used as the input for the next iterative training to train the organ segmentation model to be trained; when the loss function value reaches a preset value and no longer decreases, it indicates that the organ segmentation model tends to be stable and the organ segmentation model converges, and the training is stopped to obtain a trained organ segmentation model.

[0072] In this embodiment, see the attached Figure 3 , Figure 3 FIG. 1 is a flow chart showing the main steps of obtaining organ segmentation results of a medical image according to an embodiment of the present invention. Figure 3 As shown, in the embodiment of the present invention, the medical image to be segmented is input into a trained organ segmentation model, and obtaining the organ segmentation result of the medical image to be segmented mainly includes the following steps S301 to S302.

[0073] Step S301: inputting the medical image to be segmented into an encoder module to obtain an encoded visual feature image of the medical image to be segmented;

[0074] Step S302: Based on the skip connection module, feature fusion is performed on the encoded visual feature image, and based on the decoder module, an organ segmentation result of the medical image to be segmented is obtained.

[0075] In this embodiment, see the attached Figure 4 , Figure 4 FIG. 1 is a flow chart of segmenting a medical image to be segmented based on an organ segmentation model according to an embodiment of the present invention. Figure 4 As shown, the medical image to be segmented is input into the encoder module to obtain the encoded visual feature image of the medical image to be segmented, including: the medical image to be segmented is processed step by step based on multiple encoding layers to obtain the encoded visual feature image; wherein the medical image to be segmented is sequentially input into the dynamic deformable convolution module and the downsampling module for convolution and dimensionality reduction respectively, and an encoded feature image is output after each dynamic deformable convolution module and downsampling module.

[0076] In one embodiment, local features at different positions of an image are detected through a dynamic deformable convolution module, and a maximum pooling operation is performed in a downsampling module to downsample the local features and reduce the data dimension, so that the size of the feature image is gradually reduced and key information is retained; the encoder is used to extract features of the medical image to be segmented, capture edge features, texture features, shape features, noise distribution and other information of the image, and obtain an encoded visual feature image of the medical image to be segmented.

[0077] In one embodiment, a skip connection is used between the encoder and the decoder to help the model to splice image features of the same layer and the same size, so as to ensure that the semantic information in the encoding process is supplemented in the process of decoding and restoring the image information; based on the skip connection module, the encoded visual feature image is feature fused, and based on the decoder module, the organ segmentation result of the medical image to be segmented is obtained, including:

[0078] The encoded visual feature image is processed step by step based on multiple decoding layers to obtain the organ segmentation result of the medical image to be segmented; wherein, the encoded visual feature image is sequentially input into the dynamic deformable convolution module, the upsampling module and the attention mechanism module for convolution, dimension restoration and attention enhancement respectively, and a decoded feature image is output after each dynamic deformable convolution module, the upsampling module and the attention mechanism module; so that the size of the feature image is gradually restored, and finally an output image with a size close to that of the input image is generated; and based on the jump connection module, the encoded feature image of the corresponding layer is feature fused with the decoded feature image; the decoder is used to remap the low-dimensional feature image back to the spatial dimension of the encoder feature, extract higher-level features, improve the quality and accuracy of image generation, and finally obtain the organ segmentation result of the medical image to be segmented.

[0079] Optionally, an ECA attention mechanism module is selected in the attention mechanism module to improve the performance and generalization ability of the model. It should be noted that the type of attention mechanism module selected here is only an exemplary description and does not constitute a limitation on the implementation of the present invention. In practical applications, different selections can be made as needed, as long as the channel attention feature map can be obtained.

[0080] Optionally, the attention mechanism module is set in the last two decoding layers in the decoder. It should be noted that the position and number of the attention mechanism modules in this embodiment are only exemplary and not limiting. In practical applications of the embodiments of the present invention, the position and number of the attention mechanism modules can be selected according to actual conditions.

[0081] In one implementation, the attention mechanism module includes a global average pooling layer and a one-dimensional convolutional layer;

[0082] The decoded feature image is converted into a one-dimensional feature vector based on the global average pooling layer;

[0083] Based on the one-dimensional convolution layer, the one-dimensional feature vector is converted into a channel attention vector; usually a smaller convolution kernel (such as 1x1 convolution) and an activation function are used to generate the channel attention vector;

[0084] Multiply the channel attention vector with the decoded feature image to obtain the channel attention feature map.

[0085] The attention mechanism module effectively utilizes the correlation between channels to enhance the key channel information in the decoded feature image through the combination of global average pooling layer and one-dimensional convolution layer. This mechanism helps to improve the performance of tasks such as medical image segmentation because it can focus on the image features that are most critical to the segmentation results.

[0086] In one embodiment, the organ segmentation model further includes a 1×1 convolution layer; the convolution kernel size of the 1×1 convolution layer is adaptively determined according to the nonlinear mapping of the channel size, and no dimensionality reduction operation is required;

[0087] The method also includes:

[0088] The final output decoded feature image is processed based on a 1×1 convolutional layer to obtain the organ segmentation result of the medical image to be segmented.

[0089] In this embodiment, the dynamic deformable convolution module adopts a dynamic convolution kernel; the principle of dynamic convolution is to adaptively adjust the convolution parameters according to the input image; the static convolution kernel performs the same processing on different inputs, while the dynamic convolution kernel uses different convolution kernels to select appropriate convolution parameters to extract features according to the different inputs; different from the traditional static convolution (a single convolution kernel per layer), the dynamic convolution allows the network to dynamically generate convolution kernels according to the input data at runtime; this feature enables the network to use the convolution kernel that best suits the current data for feature extraction according to the different input data; dynamic convolution can not only dynamically adjust the convolution kernel between different input data, but also dynamically adjust the convolution kernel at different layers or different positions of the same input data; dynamic convolution increases the complexity of the model by dynamically aggregating multiple convolution kernels without increasing the depth or width of the network; it can capture the features of the input data more finely, improve the accuracy and efficiency of feature extraction, and help improve the representation ability and generalization performance of the model while maintaining the computational efficiency of the model.

[0090] In one implementation, dynamic convolution dynamically aggregates multiple parallel convolution kernels by fusion attention mechanism; firstly, an attention operation is performed on the input and then the weight of each convolution kernel is obtained, and the learned weights are superimposed on different convolution kernels to realize dynamic selection of convolution kernels; see Appendix for details. Figure 5 , Figure 5 FIG. 1 is a flow chart of the main steps of a method for obtaining a dynamic convolution kernel according to an embodiment of the present invention. Figure 5 As shown, the method for obtaining a dynamic convolution kernel in an embodiment of the present invention mainly includes the following steps S501 to S503.

[0091] Step S501: Preset the attention module of the dynamic convolution kernel, and preset k convolution kernels with the same scale and number of channels, and the k convolution kernels share the same kernel size and input / output dimensions. The attention module of the dynamic convolution kernel includes a global average pooling layer (GlobalAvgPooling), a fully connected layer (Fully Connected Layer) and a normalization layer;

[0092] Step S502: input k convolution kernels into the global average pooling layer to obtain global spatial features, then obtain k dimensions through the fully connected layer, and input the k dimensions into the normalization layer to obtain k attention weights corresponding to the k convolution kernels;

[0093] Step S503: weight the k convolution kernels based on the k attention weights to obtain a dynamic convolution kernel.

[0094] Dynamic convolution that dynamically superimposes multiple convolution kernels based on attention not only significantly improves the expressive power, but also has very little additional computational cost, making it more friendly to efficient networks and can be easily integrated into existing network architectures.

[0095] It should be pointed out that although the various steps in the above embodiments are described in a specific order, those skilled in the art can understand that in order to achieve the effect of the present invention, different steps do not have to be executed in such an order, they can be executed simultaneously (in parallel) or in other orders. These adjusted schemes are equivalent to the technical schemes described in the present invention, and therefore will also fall within the scope of protection of the present invention.

[0096] It is understood by those skilled in the art that the present invention implements all or part of the processes in the method of the above embodiment, and can also be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable storage medium may include: any entity or device, medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory, random access memory, electric carrier signal, telecommunication signal and software distribution medium, etc. that can carry the computer program code.

[0097] Another aspect of the present invention provides a computer-readable storage medium.

[0098] In an embodiment of a computer-readable storage medium according to the present invention, the computer-readable storage medium may be configured to store a program for executing the medical image organ segmentation method of the above method embodiment, and the program may be loaded and run by a processor to implement the above medical image organ segmentation method. For ease of explanation, only the parts related to the embodiment of the present invention are shown. For specific technical details not disclosed, please refer to the method part of the embodiment of the present invention. The computer-readable storage medium may be a storage device formed by various electronic devices. Optionally, the computer-readable storage medium in the embodiment of the present invention is a non-temporary computer-readable storage medium.

[0099] Another aspect of the present invention provides an electronic device.

[0100] In an embodiment of an electronic device according to the present invention, the electronic device may include at least one processor; and a memory connected to the at least one processor in communication; wherein a computer program is stored in the memory, and when the computer program is executed by the at least one processor, the method described in any of the above embodiments is implemented. Figure 6 , Figure 6 FIG. 6 exemplarily shows that the memory 61 and the processor 62 are communicatively connected via a bus.

[0101] In some embodiments of the present invention, the electronic device described in the present invention may be, but is not limited to, a mobile phone, a tablet computer, a desktop, a laptop, a handheld computer, a notebook computer, a vehicle-mounted device, an ultra-mobile personal computer (UMPC), a netbook, a personal digital assistant (PDA), an augmented reality (AR) or a virtual reality (VR) device, etc., and the embodiments of the present invention are not limited to this.)

[0102] So far, the technical solution of the present invention has been described in conjunction with an embodiment shown in the accompanying drawings, but it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.

Claims

1. A method for segmenting organs in medical images, characterized in that: include: Acquire a medical image to be segmented; Inputting the medical image to be segmented into a trained organ segmentation model to obtain an organ segmentation result of the medical image to be segmented; Among them, the organ segmentation model includes an encoder and a decoder; the encoder includes multiple encoding layers connected in sequence, and the encoding layer includes a dynamic deformable convolution module and a downsampling module; the decoder includes multiple decoding layers connected in sequence, and the decoding layer includes a dynamic deformable convolution module, an upsampling module and an attention mechanism module; the encoding layer and the corresponding decoding layer are connected through a jump connection module.

2. The medical image organ segmentation method according to claim 1, characterized in that: The step of inputting the medical image to be segmented into a trained organ segmentation model to obtain an organ segmentation result of the medical image to be segmented comprises: Inputting the medical image to be segmented into the encoder module to obtain an encoded visual feature image of the medical image to be segmented; Based on the jump connection module, feature fusion is performed on the encoded visual feature image, and based on the decoder module, an organ segmentation result of the medical image to be segmented is obtained.

3. The medical image organ segmentation method according to claim 2, characterized in that: The step of inputting the medical image to be segmented into the encoder module to obtain an encoded visual feature image of the medical image to be segmented comprises: The medical image to be segmented is processed step by step based on multiple coding layers to obtain a coded visual feature image; wherein the medical image to be segmented is sequentially input into the dynamic deformable convolution module and the downsampling module for convolution and dimension reduction respectively, and a coded feature image is output after each passing through the dynamic deformable convolution module and the downsampling module.

4. The medical image organ segmentation method according to claim 3, characterized in that: The organ segmentation model also includes a 1×1 convolutional layer; The step of performing feature fusion on the encoded visual feature image based on the skip connection module, and obtaining an organ segmentation result of the medical image to be segmented based on the decoder module, comprises: The encoded visual feature image is processed step by step based on multiple decoding layers to obtain the organ segmentation result of the medical image to be segmented; wherein the encoded visual feature image is sequentially input into the dynamic deformable convolution module, the upsampling module and the attention mechanism module for convolution, dimension restoration and attention enhancement respectively, and a decoded feature image is output after each of the dynamic deformable convolution module, the upsampling module and the attention mechanism module; and based on the jump connection module, the encoded feature image of the corresponding layer is feature fused with the decoded feature image; The decoded feature image finally outputted is processed based on a 1×1 convolutional layer to obtain an organ segmentation result of the medical image to be segmented.

5. The medical image organ segmentation method according to claim 4, characterized in that: The attention mechanism module includes a global average pooling layer and a one-dimensional convolution layer; Converting the decoded feature image into a one-dimensional feature vector based on the global average pooling layer; Converting the one-dimensional feature vector into a channel attention vector based on the one-dimensional convolutional layer; Multiply the channel attention vector by the decoded feature image to obtain a channel attention feature map.

6. The medical image organ segmentation method according to claim 1, characterized in that: The dynamic deformable convolution module adopts a dynamic convolution kernel; The method for obtaining the dynamic convolution kernel includes: Preset an attention module of a dynamic convolution kernel, and preset k convolution kernels with the same scale and number of channels, wherein the attention module of the dynamic convolution kernel includes a global average pooling layer, a fully connected layer, and a normalization layer; Input k convolution kernels into the global average pooling layer to obtain global spatial features, then obtain k dimensions through the fully connected layer, and input the k dimensions into the normalization layer to obtain k attention weights corresponding to the k convolution kernels; The k convolution kernels are weighted based on the k attention weights to obtain the dynamic convolution kernel.

7. The medical image organ segmentation method according to claim 1, characterized in that: The method further comprises training the organ segmentation model according to the following steps: Label historical medical images to obtain real organ segmentation images; Inputting the historical medical image into the organ segmentation model to be trained to obtain a predicted organ segmentation image; Obtaining a loss function based on the real organ segmentation image and the predicted organ segmentation image; The gradient of the parameters of each module in the organ segmentation model to be trained is obtained based on the loss function and the optimization algorithm, and the parameters of each module in the organ segmentation model to be trained are updated based on the gradient of the parameters of each module.

8. The method for medical image organ segmentation according to any one of claims 1 to 7, characterized in that: The method further comprises: After the medical image to be segmented is acquired, one or more preprocessing operations including normalization, cropping, and data enhancement are performed on the medical image to be segmented.

9. An electronic device, characterized in that: include: at least one processor; and, a memory communicatively coupled to the at least one processor; Wherein, a computer program is stored in the memory, and when the computer program is executed by the at least one processor, the medical image organ segmentation method according to any one of claims 1 to 8 is implemented.

10. A computer-readable storage medium storing a plurality of program codes, characterized in that: The program code is suitable for being loaded and run by a processor to execute the medical image organ segmentation method according to any one of claims 1 to 8.

Images