Image segmentation method based on deep learning and computer program product

Abstract

The invention discloses an image segmentation method based on deep learning and a computer program product, and relates to the technical field of image processing, and the method comprises the steps: carrying out the feature extraction of a target medical image through a target image encoder, and obtaining a multi-scale feature; the additional prompt information is input into a prompt encoder, and prompt features are obtained through information conversion; based on the multi-scale features and the prompt features, image segmentation is carried out on the target medical image through a target mask decoder to obtain a target segmentation result, and a target image encoder and the target mask decoder optimize the segmentation performance by introducing distance loss. Through multi-scale feature extraction, detail information on different scales can be captured, and the calculation cost is reduced while the accuracy of image segmentation is improved. Information input of the segmentation model can be enriched by introducing additional prompt information, boundary accuracy is emphasized by introducing distance loss, and the problem that the small target data segmentation effect is poor is solved.

Description

Technical Field

[0001] The present application relates to the field of image processing technology, and in particular to an image segmentation method and a computer program product based on deep learning. Background Art

[0002] Compared with natural images, medical images usually contain complex anatomical structures, with blurred boundaries and variable shapes between different organs and tissues, which brings great difficulties to the segmentation task.

[0003] At present, deep learning technology has been widely used in image segmentation tasks. However, when the overall image area is small, it is easy to have uneven attention distribution and ignore small targets. This makes the model unable to give accurate segmentation results when dealing with small targets or areas with unclear boundaries.

[0004] Therefore, how to improve the accuracy of image segmentation when dealing with small objects or areas with unclear boundaries is a problem that needs to be solved urgently. Summary of the invention

[0005] The main purpose of this application is to provide an image segmentation method and a computer program product based on deep learning, aiming to solve the technical problem of low image segmentation accuracy when processing small targets or areas with unclear boundaries.

[0006] To achieve the above objectives, the present application proposes an image segmentation method based on deep learning, the method comprising:

[0007] Extract features of target medical images through target image encoder to obtain multi-scale features;

[0008] Input the additional prompt information into the prompt encoder, and obtain the prompt features through information conversion;

[0009] Based on the multi-scale features and the prompt features, the target medical image is segmented by a target mask decoder to obtain a target segmentation result, wherein the target image encoder and the target mask decoder optimize the segmentation performance by introducing a distance loss.

[0010] In one embodiment, the step of extracting features from the target medical image by using a target image encoder to obtain multi-scale features includes:

[0011] Extract features of the target medical image through the residual visual Mamba model to obtain the initial features of the target medical image;

[0012] Based on the initial features, the features extracted from the target medical image are integrated through a state space model to obtain multi-scale features of the target medical image.

[0013] In one embodiment, the step of extracting features of the target medical image by using the residual visual Mamba model to obtain initial features of the target medical image includes:

[0014] Input the target medical image into the initial convolutional layer for feature extraction to obtain shallow features;

[0015] Recursively processing the shallow features through a residual visual Mamba model to obtain local features;

[0016] By introducing deep separable convolution into the residual visual Mamba model, the local features are optimized and the initial features of the target medical image are obtained.

[0017] In one embodiment, the step of integrating the features extracted from the target medical image through a state space model based on the initial features to obtain multi-scale features of the target medical image includes:

[0018] Input the target medical image into the state space model for feature extraction to obtain global features;

[0019] The initial features and the global features are integrated at multiple levels to obtain multi-scale features.

[0020] In one embodiment, before the step of extracting features from the target medical image by using the target image encoder to obtain multi-scale features, the following steps are included:

[0021] Freezing the prompt encoder, wherein the freezing is used to stop training the prompt encoder;

[0022] The image encoder and the mask decoder are jointly trained by optimizing the loss function to obtain the target image encoder and the target mask decoder.

[0023] In one embodiment, the step of jointly training the image encoder and the mask decoder by optimizing the loss function to obtain the target image encoder and the target mask decoder comprises:

[0024] Based on the predicted point set and the real labeled point set, determining the distance loss according to the activation function;

[0025] Based on the segmentation loss, the cross entropy loss and the distance loss, the image encoder and the mask decoder are jointly trained to obtain the target image encoder and the target mask decoder.

[0026] In addition, to achieve the above objectives, the present application also proposes an image segmentation device based on deep learning, and the image segmentation device based on deep learning includes:

[0027] A feature extraction module is used to extract features of a target medical image through a target image encoder to obtain multi-scale features;

[0028] An information conversion module, used for inputting the additional prompt information into the prompt encoder, and obtaining prompt features through information conversion;

[0029] An image segmentation module is used to perform image segmentation on the target medical image through a target mask decoder based on the multi-scale features and the prompt features to obtain a target segmentation result, wherein the target image encoder and the target mask decoder optimize the segmentation performance by introducing a distance loss.

[0030] In addition, to achieve the above-mentioned objectives, the present application also proposes a deep learning-based image segmentation device, the device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the deep learning-based image segmentation method as described above.

[0031] In addition, to achieve the above-mentioned purpose, the present application also proposes a storage medium, which is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the deep learning-based image segmentation method described above are implemented.

[0032] In addition, to achieve the above-mentioned purpose, the present application also provides a computer program product, which includes a computer program, and when the computer program is executed by a processor, it implements the steps of the deep learning-based image segmentation method as described above.

[0033] One or more technical solutions proposed in this application have at least the following technical effects:

[0034] The target image encoder is used to extract features from the target medical image to obtain multi-scale features; the additional prompt information is input into the prompt encoder to obtain prompt features through information conversion; based on the multi-scale features and prompt features, the target medical image is segmented through the target mask decoder to obtain the target segmentation result, wherein the target image encoder and the target mask decoder optimize the segmentation performance by introducing distance loss. The target image encoder is used to extract multi-scale features from medical images, which can capture detailed information at different scales, improve the accuracy of image segmentation, and help reduce the model's dependence on high-resolution input images, thereby reducing computational costs. The introduction of additional prompt information and its conversion into prompt features through the prompt encoder can enrich the information input of the segmentation model, further improving the accuracy and robustness of the segmentation. The distance loss can emphasize the accuracy of the boundary. By introducing the distance loss, the performance of the model at the edge is ensured, solving the problem of poor segmentation of small target data. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0036] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0037] Figure 1 This is a flowchart of the first embodiment of the deep learning-based image segmentation method of the present application;

[0038] Figure 2 This is a flowchart of the second embodiment of the deep learning-based image segmentation method of the present application;

[0039] Figure 3 This is a schematic diagram of a target image encoder in an embodiment of the present application;

[0040] Figure 4 This is a schematic diagram of RVM+ in an embodiment of the present application;

[0041] Figure 5 This is a flowchart of the third embodiment of the image segmentation method based on deep learning of the present application;

[0042] Figure 6 is a schematic diagram of an image encoder (before optimization) in an embodiment of the present application;

[0043] Figure 7 This is a schematic diagram of the module structure of the image segmentation device based on deep learning in an embodiment of the present application;

[0044] Figure 8 This is a schematic diagram of the device structure of the hardware operating environment involved in the deep learning-based image segmentation method in the embodiment of the present application.

[0045] The purpose, features and advantages of this application will be further described in conjunction with the embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0046] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of the present application and are not used to limit the present application.

[0047] In order to better understand the technical solution of the present application, a detailed description will be given below in conjunction with the accompanying drawings and specific implementation methods.

[0048] Since medical images usually contain complex anatomical structures, the boundaries between different organs and tissues are blurred and the shapes are varied, which brings great difficulties to the segmentation task. When the overall image area is small, it is easy to have uneven attention distribution. The attention mechanism of the model may tend to focus on larger or prominent areas, thereby ignoring the details of small targets. This uneven attention distribution will make the model unable to give accurate segmentation results when dealing with small targets or areas with unclear boundaries. Due to the lack of small target data, data imbalance problems are prone to occur in medical image segmentation tasks, and it is difficult for the model to effectively learn on limited data. In addition, image segmentation is generally based on large-scale pre-trained models. Such models have a large number of parameters, a time complexity of O(n2), and high computational complexity. Especially when processing 3D medical images, the amount of calculation increases exponentially. It is difficult for a single image processor (Graphics Processing Unit, GPU) to meet the needs. Obviously, more memory and hardware resources need to be added, and the computational complexity and memory consumption are high.

[0049] The present application provides a solution, which extracts features from a target medical image through a target image encoder to obtain multi-scale features; inputs additional prompt information into the prompt encoder to obtain prompt features through information conversion; based on the multi-scale features and prompt features, the target medical image is segmented through a target mask decoder to obtain a target segmentation result, wherein the target image encoder and the target mask decoder optimize the segmentation performance by introducing distance loss. By extracting multi-scale features from medical images through a target image encoder, it is possible to capture detailed information at different scales, improve the accuracy of image segmentation, and also help reduce the model's dependence on high-resolution input images, thereby reducing computational costs. Introducing additional prompt information and converting it into prompt features through a prompt encoder can enrich the information input of the segmentation model, further improving the accuracy and robustness of the segmentation. Distance loss can emphasize the accuracy of the boundary, and by introducing distance loss, the performance of the model at the edge is ensured, solving the problem of poor segmentation of small target data.

[0050] Based on this, the present application embodiment provides an image segmentation method based on deep learning. Figure 1 , Figure 1 This is a flowchart of the first embodiment of the deep learning-based image segmentation method of the present application.

[0051] In this embodiment, the method described includes steps S10 to S30:

[0052] Step S10, extracting features from the target medical image through a target image encoder to obtain multi-scale features.

[0053] It should be noted that the target image encoder can be a pre-trained deep learning model module for extracting features with semantic information from the input target medical image. Multi-scale features can be image features extracted at different resolutions or levels. Exemplarily, a visual backbone network based on a convolutional neural network can be constructed. Since lesions may appear at different spatial scales, multi-scale features are particularly important in medical images.

[0054] Step S20: input the additional prompt information into a prompt encoder, and obtain prompt features through information conversion.

[0055] It should be noted that the additional prompt information can be contextual information provided through external input to assist in the processing of the target medical image. The prompt encoder can encode the additional prompt information to generate a feature representation that matches the target medical image. The prompt feature is a feature representation generated by the prompt encoder. Exemplarily, the prompt feature can provide the user with information such as clicks, boundaries, points or other annotations.

[0056] Step S30, based on the multi-scale features and the hint features, the target medical image is segmented by a target mask decoder to obtain a target segmentation result.

[0057] It should be noted that the target medical image can be input into the target segmentation model, and the target segmentation model can include a target image encoder, a prompt encoder, and a target mask decoder, wherein the target image encoder and the target mask decoder optimize the segmentation performance by introducing a distance loss, and the target mask decoder can be an optimized mask decoder for mapping the features extracted by the image encoder to the target segmentation result (mask). It can be understood that the original image encoder and the original mask decoder can be optimized by introducing a distance loss optimization to obtain the target image encoder and the target mask decoder. Exemplarily, the multi-scale features and the prompt features can be remapped back to the spatial resolution of the original image, thereby restoring the segmentation result.

[0058] In this embodiment, multi-scale feature extraction of medical images is performed through the target image encoder, which can capture detailed information at different scales, improve the precision and accuracy of the segmentation results, and at the same time, help reduce the model's dependence on high-resolution input images, thereby reducing computational costs. Introducing additional prompt information and converting it into prompt features through the prompt encoder can enrich the information input of the segmentation model and further improve the accuracy and robustness of the segmentation. Distance loss can emphasize the accuracy of the boundary. By introducing distance loss, the performance of the model at the edge is ensured, solving the problem of poor segmentation effect of small target data.

[0059] Reference Figure 2 , Figure 2This is a flow chart of the second embodiment of the image segmentation method based on deep learning of this application. Figure 1 The first embodiment shown here proposes a second embodiment of the image segmentation method based on deep learning of the present application.

[0060] In the second embodiment, the step S10 includes:

[0061] Step S101, extracting features of a target medical image through a residual visual Mamba model to obtain initial features of the target medical image.

[0062] It should be noted that most of the model parameters are concentrated in the image encoder, accounting for about 80% of the total parameters. The huge amount of parameters has a significant impact on the training and reasoning process. Therefore, the transformer structure in the image encoder can be changed, and the target image encoder can be set to the residual vision mamba model (RVM). The residual vision mamba model combines the residual network, which reduces the overall size of the model and improves the training speed while slightly improving the segmentation performance. Through the residual vision mamba model, the initial features of the image can be extracted from the target medical image.

[0063] Step S102 : Based on the initial features, the features extracted from the target medical image are integrated through a state space model to obtain multi-scale features of the target medical image.

[0064] It should be noted that the state-space model (SSM) has the ability to model long-distance dependencies and can integrate multiple feature layers to improve the semantic consistency of feature representation through dynamic modeling or context association. Multi-scale features can be understood as the result of integrating initial features of different resolutions, receptive fields, and semantic levels through SSM.

[0065] In this embodiment, the residual visual mamba model improves the original Transformer module into a lighter residual network structure by optimizing the encoder structure, greatly reducing the overall parameter amount of the model, and more efficiently learning complex features through residual connections, while alleviating the gradient vanishing problem in deep networks, ensuring high-quality initial feature extraction. The state space model is good at modeling long-distance dependencies and can effectively integrate information from different feature layers to ensure the consistency of contextual information in multi-scale features. Multi-scale features integrate initial features at different resolutions, receptive fields, and semantic levels, allowing the model to focus on detailed information in the image at the same time, improving the fineness and accuracy of the segmentation results.

[0066] In order to make the embodiments of the present application clearer, the following implementation methods are proposed based on the above-mentioned second embodiment.

[0067] In one embodiment, based on the above-mentioned second embodiment, the step S101 includes: inputting the target medical image into the initial convolution layer for feature extraction to obtain shallow features; recursively processing the shallow features through the residual visual mamba model to obtain local features; and optimizing the local features by introducing deep separable convolution into the residual visual mamba model to obtain the initial features of the target medical image.

[0068] It is understandable that models with a large number of processing parameters need to focus more on extracting local spatial information, especially for normalized feature maps. Under the premise of maintaining a lightweight design, feature extraction needs to be further refined. For example, for the RVM block, we add a depthwise separable convolution (DWConv.) after the normalization layer (LayerNorm) so that it can focus more on extracting local spatial information, especially for normalized feature maps, which improves the network's sensitivity to local details and overall performance. The optimized RVM model is defined as RVM+, and its composition is as follows Figure 3 shown.

[0069] For example, Figure 4 As shown in the figure, the input image P(3,1024,1024) first passes through a 7×7 large kernel convolution layer intial Conv (initialization convolution layer) to perform preliminary shallow feature extraction into (32, 512, 512), and then uses 4 RVM+ blocks to gradually extract deeper features. After downsampling, the feature map becomes (64, 256, 256); 8 RVM+ blocks are used to extract features again, and the final feature map after downsampling is (128, 128, 128).

[0070] In this implementation, the residual visual mamba model is used to recursively process shallow features to further extract local features and refine the local spatial detail information in the image. The introduction of deep separable convolution (DWConv.) effectively reduces the computational complexity of the convolution operation, allowing more focus on extracting local spatial information, especially for normalized feature maps. While maintaining a lightweight design, feature extraction is further refined to improve the network's sensitivity to local details and overall performance.

[0071] In one implementation, based on the above second embodiment, step S102 includes: inputting the target medical image into the state space model for feature extraction to obtain global features; and integrating the initial features and the global features at multiple levels to obtain multi-scale features.

[0072] It should be noted that the state space model can also dynamically model the global context information, so that the output features contain not only local details but also global semantic information, and the state space model can convert the global semantic information into global features. Exemplarily, the initial features and global features can be integrated through a fusion strategy (such as weighted summation, concatenation or attention mechanism) to form multi-scale features.

[0073] In this implementation, global features capture the overall structure and contextual semantic information of the target medical image, while local features focus on detail information (such as boundaries, textures, and small lesion areas). By integrating the initial features and global features at multiple levels, the target segmentation results can focus on local details while utilizing global semantics, significantly enhancing the feature expression capability while reducing redundant calculations.

[0074] Reference Figure 5 , Figure 5 This is a flowchart of the third embodiment of the image segmentation method based on deep learning in this application. Figure 2 The second embodiment shown proposes a third embodiment of the image segmentation method based on deep learning of the present application.

[0075] In the third embodiment, before step S10, the following steps are included:

[0076] Step S010, freezing the prompt encoder.

[0077] It should be noted that freezing is used to stop the training of the prompt encoder, for example, Figure 6 As shown, freezing the prompt encoder can be understood as stopping the parameter update of the prompt encoder, thereby avoiding the influence of new parameters on the prompt encoder.

[0078] Step S020, jointly training the image encoder and the mask decoder by optimizing the loss function to obtain a target image encoder and a target mask decoder.

[0079] For example, the loss function can combine Dice loss and cross entropy loss. Dice loss can enhance the model's ability to recognize small objects and key areas, while cross entropy loss provides detailed pixel-level classification information, which helps to improve the overall segmentation quality. The difference between the predicted mask and the true label can be measured by the optimized loss function. The loss value is back-propagated to optimize the parameters of the image encoder and mask decoder, so that its performance is gradually improved, and the target image encoder and target mask decoder are obtained.

[0080] In this embodiment, by using the strategy of freezing the hint encoder and retraining the image encoder and mask decoder, the performance of the model in the medical image segmentation task is optimized, while avoiding overfitting of the hint encoder. Retraining the image encoder can better adapt to the medical image segmentation task and enable the model to better extract medical impact features; retraining the mask decoder converts the features extracted by the image encoder into the final segmentation mask to improve the final segmentation accuracy; freezing the hint decoder ensures that the hint mechanism remains stable and is not excessively disturbed by the training data or target task, and effectively processes the hint information during the reasoning process.

[0081] In one implementation, based on the third embodiment above, step S020 includes: determining the distance loss according to the activation function based on the predicted point set and the real annotation point set; and jointly training the image encoder and the mask decoder based on the segmentation loss, the cross entropy loss, and the distance loss to obtain the target image encoder and the target mask decoder.

[0082] Exemplarily, the prediction point set can be a prediction result generated by the model, which usually represents the coordinate points or boundary points of the target area. The real annotation point set can be a target area point set annotated by a predefined algorithm, which represents the real target area. The activation function can be used to map or smooth the distance value, the distance loss can be used to measure the geometric distance difference between the prediction point set and the real annotation point set, the segmentation loss can measure the difference between the predicted segmentation result and the real segmentation result, and the cross entropy loss can be used to measure the difference between the predicted probability distribution and the real distribution in the classification task. By sharing the loss function (segmentation loss, cross entropy loss and distance loss), the parameters of the image encoder and the mask decoder are updated at the same time, so that the target segmentation model achieves the best overall performance.

[0083] For example, a bounding box hint with 0-20 pixels of random perturbation can be generated from the real annotated mask to simulate the selection error caused by manually selecting the region of interest. In order to improve the segmentation accuracy of small targets and data with unclear boundaries, a smooth Hausdorff loss function is defined using the Softmax function based on the Hausdorff distance. The definition of L2 distance can be expressed as formula (1):

[0084] d(p,t)=||pt||2 (1)

[0085] In formula (1), p represents any point in the predicted point set P, t represents any point in the true labeled point set T, and d(p, t) represents the L2 distance. For the predicted point set P and the true labeled point set T, the smoothed Hausdorff distance formula is expressed as formula (2):

[0086]

[0087] In formula (2), represents the smoothed Hausdorff distance, β represents the adjustment parameter for adjusting the smoothness of Softmax, and Softmax is a function used to smooth the distance value, which can be defined as formula (3):

[0088]

[0089] The definition formula of the smoothed Hausdorff loss function L can be expressed as formula (4):

[0090]

[0091] The gradient of the loss function can be expressed as formula (5):

[0092]

[0093] Among them, the gradient of L2 distance with respect to point p is expressed as formula (6):

[0094]

[0095] Formula (6) represents the differential form of the smoothed Hausdorff loss function, which provides the gradient of the loss function with respect to the predicted point, facilitating back propagation during the optimization process.

[0096] In this embodiment, by calculating the maximum distance between the predicted boundary and the true boundary, the Hausdorff loss ensures the performance of the model at the edge, especially when dealing with complex structures (such as blood vessels or organ contours). The three loss functions are combined together so that the model can focus on global segmentation performance, pixel-level accuracy and boundary clarity at the same time.

[0097] It should be noted that the above examples are only used to understand the present application and do not constitute a limitation on the image segmentation method based on deep learning of the present application. More forms of simple transformations based on this technical concept are all within the scope of protection of the present application.

[0098] This application also provides an image segmentation device based on deep learning, please refer to Figure 7 , the image segmentation device based on deep learning includes:

[0099] A feature extraction module 10 is used to extract features from a target medical image through a target image encoder to obtain multi-scale features;

[0100] An information conversion module 20, used to input the additional prompt information into the prompt encoder, and obtain prompt features through information conversion;

[0101] The image segmentation module 30 is used to perform image segmentation on the target medical image based on the multi-scale features and the prompt features through a target mask decoder to obtain a target segmentation result, wherein the target image encoder and the target mask decoder optimize the segmentation performance by introducing a distance loss.

[0102] The deep learning-based image segmentation device provided by the present application adopts the deep learning-based image segmentation method in the above-mentioned embodiment, which can solve the technical problem of low accuracy of image segmentation when processing small objects or areas with unclear boundaries. Compared with the prior art, the beneficial effects of the deep learning-based image segmentation device provided by the present application are the same as the beneficial effects of the deep learning-based image segmentation method provided by the above-mentioned embodiment, and the other technical features of the deep learning-based image segmentation device are the same as the features disclosed in the above-mentioned embodiment method, which will not be repeated here.

[0103] The present application provides an image segmentation device based on deep learning, and the image segmentation device based on deep learning includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the image segmentation method based on deep learning in the above-mentioned embodiment one.

[0104] Reference below Figure 8 , which shows a schematic diagram of the structure of a deep learning-based image segmentation device suitable for implementing the embodiment of the present application. The deep learning-based image segmentation device in the embodiment of the present application may include but is not limited to mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 8 The deep learning-based image segmentation device shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0105] like Figure 8As shown, the image segmentation device based on deep learning may include a processing device 1001 (such as a central processing unit, a graphics processor, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM: Read Only Memory) 1002 or a program loaded from a storage device 1003 to a random access memory (RAM: Random Access Memory) 1004. In RAM1004, various programs and data required for the operation of the image segmentation device based on deep learning are also stored. The processing device 1001, ROM1002 and RAM1004 are connected to each other via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to the I / O interface 1006: input devices 1007 including, for example, a touch screen, a touchpad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; output devices 1008 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; storage devices 1003 including, for example, a magnetic tape, a hard disk, etc.; and communication devices 1009. The communication device 1009 can allow the deep learning-based image segmentation device to communicate with other devices wirelessly or by wire to exchange data. Although Figure 8 A deep learning-based image segmentation device with various systems is shown, but it should be understood that it is not required to implement or have all the systems shown. More or fewer systems may be implemented or have instead.

[0106] In particular, according to the embodiments disclosed in the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication device, or installed from a storage device 1003, or installed from a ROM 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the method of the embodiment disclosed in the present application are executed.

[0107] The deep learning-based image segmentation device provided by the present application adopts the deep learning-based image segmentation method in the above-mentioned embodiment, which can solve the technical problem of low accuracy of image segmentation when processing small targets or areas with unclear boundaries. Compared with the prior art, the beneficial effects of the deep learning-based image segmentation device provided by the present application are the same as the beneficial effects of the deep learning-based image segmentation method provided by the above-mentioned embodiment, and the other technical features of the deep learning-based image segmentation device are the same as the features disclosed in the method of the previous embodiment, which will not be repeated here.

[0108] It should be understood that the various parts disclosed in this application can be implemented by hardware, software, firmware or a combination thereof. In the description of the above embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0109] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0110] The present application provides a computer-readable storage medium having computer-readable program instructions (i.e., computer programs) stored thereon, and the computer-readable program instructions are used to execute the deep learning-based image segmentation method in the above-mentioned embodiment.

[0111] The computer-readable storage medium provided in the present application may be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, systems or devices, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, system or device. The program code contained on the computer-readable storage medium may be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination of the above.

[0112] The above-mentioned computer-readable storage medium may be included in the image segmentation device based on deep learning; or it may exist independently without being assembled into the image segmentation device based on deep learning.

[0113] The above-mentioned computer-readable storage medium carries one or more programs. When the above-mentioned one or more programs are executed by an image segmentation device based on deep learning, the image segmentation device based on deep learning: extracts features of a target medical image through a target image encoder to obtain multi-scale features; inputs additional prompt information into the prompt encoder to obtain prompt features through information conversion; based on the multi-scale features and the prompt features, performs image segmentation on the target medical image through a target mask decoder to obtain a target segmentation result, wherein the target image encoder and the target mask decoder optimize the segmentation performance by introducing distance loss.

[0114] Computer program code for performing the operations of the present application may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0115] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present application. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0116] The modules involved in the embodiments described in this application may be implemented by software or hardware, wherein the name of the module does not constitute a limitation on the unit itself in some cases.

[0117] The readable storage medium provided in the present application is a computer-readable storage medium, which stores computer-readable program instructions (i.e., computer programs) for executing the above-mentioned image segmentation method based on deep learning, and can solve the technical problem of low accuracy of image segmentation when processing small targets or areas with unclear boundaries. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in the present application are the same as the beneficial effects of the image segmentation method based on deep learning provided in the above-mentioned embodiments, and will not be repeated here.

[0118] The present application also provides a computer program product, including a computer program, which, when executed by a processor, implements the steps of the above-mentioned deep learning-based image segmentation method.

[0119] The computer program product provided by the present application can solve the technical problem of low accuracy of image segmentation when processing small objects or areas with unclear boundaries. Compared with the prior art, the beneficial effects of the computer program product provided by the present application are the same as the beneficial effects of the image segmentation method based on deep learning provided by the above embodiment, and will not be repeated here.

[0120] The above descriptions are only some embodiments of the present application, and are not intended to limit the patent scope of the present application. All equivalent structural changes made using the contents of the present application specification and drawings under the technical concept of the present application, or direct / indirect applications in other related technical fields are included in the patent protection scope of the present application.

Claims

1. A deep learning-based image segmentation method, characterized in that: The method includes: Extract features of target medical images through target image encoder to obtain multi-scale features; Input the additional prompt information into the prompt encoder, and obtain the prompt features through information conversion; Based on the multi-scale features and the prompt features, the target medical image is segmented by a target mask decoder to obtain a target segmentation result, wherein the target image encoder and the target mask decoder optimize the segmentation performance by introducing a distance loss.

2. The method according to claim 1, characterized in that The step of extracting features from the target medical image by using a target image encoder to obtain multi-scale features includes: Extract features of the target medical image through the residual visual Mamba model to obtain the initial features of the target medical image; Based on the initial features, the features extracted from the target medical image are integrated through a state space model to obtain multi-scale features of the target medical image.

3. The method according to claim 2, characterized in that The step of extracting features of the target medical image through the residual visual Mamba model to obtain initial features of the target medical image includes: Input the target medical image into the initial convolutional layer for feature extraction to obtain shallow features; Recursively processing the shallow features through a residual visual Mamba model to obtain local features; By introducing deep separable convolution into the residual visual Mamba model, the local features are optimized and the initial features of the target medical image are obtained.

4. The method according to claim 2, characterized in that The step of integrating the features extracted from the target medical image through the state space model based on the initial features to obtain the multi-scale features of the target medical image includes: Input the target medical image into the state space model for feature extraction to obtain global features; The initial features and the global features are integrated at multiple levels to obtain multi-scale features.

5. The method according to any one of claims 1 to 4, characterized in that Before the step of extracting features from the target medical image by the target image encoder to obtain multi-scale features, the method includes: Freezing the prompt encoder, wherein the freezing is used to stop training the prompt encoder; The image encoder and the mask decoder are jointly trained by optimizing the loss function to obtain the target image encoder and the target mask decoder.

6. The method according to claim 5, characterized in that The step of jointly training the image encoder and the mask decoder by optimizing the loss function to obtain the target image encoder and the target mask decoder comprises: Based on the predicted point set and the real labeled point set, determining the distance loss according to the activation function; Based on the segmentation loss, the cross entropy loss and the distance loss, the image encoder and the mask decoder are jointly trained to obtain the target image encoder and the target mask decoder.

7. An image segmentation device based on deep learning, characterized in that: The device comprises: A feature extraction module is used to extract features of a target medical image through a target image encoder to obtain multi-scale features; An information conversion module, used for inputting the additional prompt information into the prompt encoder, and obtaining prompt features through information conversion; An image segmentation module is used to perform image segmentation on the target medical image through a target mask decoder based on the multi-scale features and the prompt features to obtain a target segmentation result, wherein the target image encoder and the target mask decoder optimize the segmentation performance by introducing a distance loss.

8. An image segmentation device based on deep learning, characterized in that: The device comprises: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the deep learning-based image segmentation method according to any one of claims 1 to 6.

9. A storage medium, characterized in that: The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the deep learning-based image segmentation method according to any one of claims 1 to 6 are implemented.

10. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the steps of the deep learning-based image segmentation method according to any one of claims 1 to 6 are implemented.

Images