Image segmentation method, device, equipment, storage medium and program product

Abstract

The application relates to an image segmentation method, device, equipment, storage medium and program product. The method comprises the following steps: acquiring an input interactive text and a medical image; the interactive text is used for indicating a target segmentation task to be performed on the medical image; a target segmentation subnetwork matched with the target segmentation task is determined according to the interactive text; an image segmentation network is determined according to a preset common segmentation network and the target segmentation subnetwork; the common segmentation network is used for segmenting related tasks in the medical image and other images; and the medical image is segmented by using the image segmentation network to determine a target segmentation result of the medical image. The method can realize segmentation of various medical images.

Description

Technical Field

[0001] This application relates to the field of medical image processing technology, and in particular to an image segmentation method, apparatus, device, storage medium, and program product. Background Technology

[0002] With the continuous development of medical imaging technology and neural network technology, the application of combining neural network technology and medical imaging technology for patient image analysis is becoming increasingly widespread. Currently, when using medical imaging to analyze patient images, a segmentation model based on neural networks can be used first to segment the medical images to extract lesion regions, and then further image analysis can be performed on the lesion regions.

[0003] Currently, with the continuous development of language vision technology, segmentation models based on language vision can be trained for different segmentation objects during segmentation, thereby achieving image segmentation of the object.

[0004] However, the aforementioned techniques are difficult to apply to segmentation tasks for various medical images. Summary of the Invention

[0005] Therefore, it is necessary to provide an image segmentation method, apparatus, device, storage medium, and program product that can be applied to various medical image segmentation tasks to address the above-mentioned technical problems.

[0006] In a first aspect, this application provides an image segmentation method, which includes:

[0007] The system acquires input interactive text and medical images; the interactive text is used to indicate the target segmentation task to be performed on the medical images.

[0008] Determine the target segmentation subnetwork that matches the target segmentation task based on the interactive text;

[0009] The image segmentation network is determined based on the preset common segmentation network and the target segmentation subnetwork; the aforementioned common segmentation network is used to segment medical images and related tasks in other images.

[0010] Medical images are segmented using an image segmentation network to determine the target segmentation result.

[0011] In one embodiment, the above-mentioned determination of the target segmentation sub-network matching the target segmentation task based on interactive text includes:

[0012] The interactive text is encoded to determine the encoded text; the encoded text includes the target segmentation task.

[0013] The encoded text is processed through multiple preset routing modules to select a segmentation subnetwork and determine the target segmentation subnetwork that matches the target segmentation task; each routing module includes at least one candidate segmentation subnetwork.

[0014] In one embodiment, the above-mentioned process of selecting a segmented sub-network from multiple preset routing modules to determine the target segmentation sub-network that matches the target segmentation task includes:

[0015] The encoded text is input into each routing module for path selection processing to determine the candidate segmented subnetworks selected in each routing module;

[0016] The candidate segmented subnetworks selected from each routing module will be determined as the target segmented subnetworks.

[0017] In one embodiment, determining the candidate segmented subnetworks selected from each routing module as the target segmented subnetwork includes:

[0018] The candidate segmented subnetworks selected from each routing module are connected according to the connection relationship between the corresponding routing modules to obtain the target segmented subnetwork.

[0019] In one embodiment, the aforementioned common segmentation network includes multiple downsampling modules, each of which is used to perform downsampling processing on medical images at different resolutions;

[0020] The aforementioned target segmentation subnetwork includes multiple segmentation subnetworks, each used to perform upsampling processing on medical images at different resolutions.

[0021] In one embodiment, determining the image segmentation network based on a preset common segmentation network and a target segmentation sub-network includes:

[0022] Each downsampling module of the common segmentation network is connected to the corresponding segmentation subnetwork in the target segmentation subnetwork to obtain the image segmentation network.

[0023] In one embodiment, the above-described encoding process for interactive text to determine the encoded text includes:

[0024] The interactive text is encoded using a language model to determine the encoded text.

[0025] In one embodiment, the above-mentioned use of an image segmentation network to segment medical images and determine the target segmentation result of the medical images includes:

[0026] An image segmentation network is used to segment medical images to determine the initial segmentation results.

[0027] If the initial segmentation result does not meet the segmentation requirements, a new interactive text is obtained; the new interactive text is the text set for the local segmentation result or the overall segmentation result that does not meet the segmentation requirements in the initial segmentation result.

[0028] Return to the steps described above for determining the target segmentation subnetwork that matches the target segmentation task based on the interactive text, until the new initial segmentation result meets the segmentation requirements and the target segmentation result is obtained.

[0029] Secondly, this application also provides an image segmentation apparatus, which includes:

[0030] The acquisition module is used to acquire the input interactive text and medical images; the interactive text is used to indicate the target segmentation task to be performed on the medical images.

[0031] The segmentation subnetwork determination module is used to determine the target segmentation subnetwork that matches the target segmentation task based on the interactive text.

[0032] The image segmentation network determination module is used to determine the image segmentation network based on a preset common segmentation network and a target segmentation subnetwork; the aforementioned common segmentation network is used to segment medical images and related tasks in other images.

[0033] The segmentation module is used to segment medical images using an image segmentation network and determine the target segmentation result of the medical images.

[0034] Thirdly, this application also provides a computer device. The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to perform the following steps:

[0035] The system acquires input interactive text and medical images; the interactive text is used to indicate the target segmentation task to be performed on the medical images.

[0036] Determine the target segmentation subnetwork that matches the target segmentation task based on the interactive text;

[0037] The image segmentation network is determined based on the preset common segmentation network and the target segmentation subnetwork; the aforementioned common segmentation network is used to segment medical images and related tasks in other images.

[0038] Medical images are segmented using an image segmentation network to determine the target segmentation result.

[0039] Fourthly, this application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program thereon, which, when executed by a processor, performs the following steps:

[0040] The system acquires input interactive text and medical images; the interactive text is used to indicate the target segmentation task to be performed on the medical images.

[0041] Determine the target segmentation subnetwork that matches the target segmentation task based on the interactive text;

[0042] The image segmentation network is determined based on the preset common segmentation network and the target segmentation subnetwork; the aforementioned common segmentation network is used to segment medical images and related tasks in other images.

[0043] Medical images are segmented using an image segmentation network to determine the target segmentation result.

[0044] Fifthly, this application also provides a computer program product. The computer program product includes a computer program that, when executed by a processor, performs the following steps:

[0045] The system acquires input interactive text and medical images; the interactive text is used to indicate the target segmentation task to be performed on the medical images.

[0046] Determine the target segmentation subnetwork that matches the target segmentation task based on the interactive text;

[0047] The image segmentation network is determined based on the preset common segmentation network and the target segmentation subnetwork; the aforementioned common segmentation network is used to segment medical images and related tasks in other images.

[0048] Image segmentation networks are used to segment medical images and determine the target segmentation result. The aforementioned image segmentation method, apparatus, device, storage medium, and program product acquire medical images and interactive text indicating the target segmentation task to be performed on the medical images. Then, based on the interactive text, a target segmentation sub-network matching the target segmentation task is determined. After determining an image segmentation network based on a preset common segmentation network and the target segmentation sub-network, the image segmentation network is used to segment the medical images to obtain the target segmentation result. The common segmentation network is used to segment related tasks in the medical images and other images. In this method, since different segmentation sub-networks can be determined based on the interactive text of different segmentation tasks, and these different segmentation sub-networks can be combined with the common segmentation network to obtain different image segmentation networks for segmenting medical images, it is possible to segment medical images with different or multiple segmentation tasks. Furthermore, the image segmentation network also incorporates a common segmentation network for segmenting related tasks in various types of images, thus improving the generalization performance of the image segmentation network. Attached Figure Description

[0049] Figure 1 This is an internal structural diagram of a computer device in one embodiment;

[0050] Figure 2 This is a flowchart illustrating an image segmentation method in one embodiment;

[0051] Figure 3 This is a flowchart illustrating the image segmentation method in another embodiment;

[0052] Figure 4 This is a flowchart illustrating the image segmentation method in another embodiment;

[0053] Figure 5 Here is a structural example diagram of the routing module in another embodiment;

[0054] Figure 6 This is an example diagram of the architecture of an image segmentation network in another embodiment;

[0055] Figure 7 This is a flowchart illustrating the image segmentation method in another embodiment;

[0056] Figure 8 This is a schematic diagram illustrating the specific process of the image segmentation method in another embodiment;

[0057] Figure 9 This is a structural block diagram of an image segmentation device in one embodiment. Detailed Implementation

[0058] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0059] Multimodal learning refers to using artificial intelligence methods to process and understand multi-source modal information. This multimodal information can include images, image modalities, language, discrete instruction sets, impact reports, etc. It has wide applications in image-text retrieval, visual question answering, visual-language navigation, and multimodal dialogue. Currently, common visual models typically adopt a single-task approach, with each task trained from scratch. Tasks cannot learn from each other, and the limited data size of a single task leads to a series of bias problems. While similar tasks can mutually improve performance, in practical applications, performance often depends on the task data distribution, resulting in poor generalization. In recent years, big data-based pre-training techniques have learned more general information by using large amounts of data and can easily transfer it to downstream tasks. Essentially, different tasks learn from each other's knowledge. Deep learning has witnessed the rise of language models, such as BERT and GPT. For visual problems involving language, there is also some work, such as CLIP (Contrastive Language-Image Pretraining), which jointly trains language and image information. On large-scale image-text datasets, contrastive loss is used to train image and text encoders to match the results of the image and text encoders. However, this approach can only be applied to image classification or image retrieval tasks and cannot produce image segmentation results requiring output masks. Therefore, none of the above techniques can achieve segmentation tasks for various types of medical images.

[0060] The image segmentation method provided in this application can be applied to computer devices, which can be servers or terminals. Taking a terminal as an example, its internal structure diagram can be as follows: Figure 1 As shown, the computer device includes a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, NFC (Near Field Communication), or other technologies. When the computer program is executed by the processor, it implements an image segmentation method. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the computer device casing, or an external keyboard, touchpad, or mouse.

[0061] Those skilled in the art will understand that Figure 1 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0062] In one embodiment, such as Figure 2 As shown, an image segmentation method is provided, which can be applied to... Figure 1 Taking a computer device as an example, the method may include the following steps:

[0063] S202, Obtain the input interactive text and medical image; the interactive text is used to indicate the target segmentation task to be performed on the medical image.

[0064] The medical images can be images of the site or region to be tested, and can be two-dimensional or three-dimensional images. The acquisition methods for medical images can be medical images obtained by scanning the site or region to be tested and then reconstructing the image, or medical images that have been pre-obtained and stored from the cloud or server, or other acquisition methods, which are not specifically limited here.

[0065] The interactive text input can be entered by the user on the display interface of the computer device. This interactive text can be entered by the user when they need to segment medical images. It can indicate the target segmentation task to be performed on the medical images. The target segmentation task can be, for example, segmenting the part or region to be measured in the medical image, or segmenting a sub-region within the part or region to be measured, or other forms of segmentation task.

[0066] For interactive text input by the user, it can be input in the form of voice, which is then recognized and converted into text to obtain the interactive text; alternatively, the user can directly input text in the form of text; or other input methods can be used to obtain interactive text, without specific limitations. The input interactive text can indicate the target segmentation task to be performed on the medical image. For example, it can be the name or identifier of the specific location or region to be segmented, or a feature of the location or region to be segmented, or a task for further fine segmentation, or other situations. Taking lung segmentation as an example, the input interactive text could be "left lung," "lung lobe," "precise segmentation," "filling gap in the pulmonary artery," etc.

[0067] S204, determine the target segmentation subnetwork that matches the target segmentation task based on the interactive text.

[0068] In this step, different segmentation subnetworks can be pre-set for different segmentation tasks, so that different segmentation subnetworks can perform different segmentation tasks, thereby achieving accurate segmentation of different segmentation targets.

[0069] The different segmentation subnetworks here can be segmentation subnetworks set up separately for unrelated segmentation tasks in various medical images. Generally, each segmentation subnetwork can be a segmentation subnetwork with completely different structures, or it can be a segmentation subnetwork with some of the same structure.

[0070] After obtaining the interactive text input by the user, the target segmentation task that the user wants to perform can be obtained from the interactive text, such as segmenting the left lung. In this way, a suitable target segmentation subnetwork can be matched according to the target segmentation task.

[0071] Generally, the higher the accuracy requirement of a segmentation task, the more complex the structure of the segmentation subnetwork needs to be. For segmentation tasks with lower accuracy requirements, a simpler structure can be used for the segmentation subnetwork. This allows for matching different segmentation subnetworks to the accuracy requirements of different segmentation tasks, thus meeting the needs of various segmentation tasks while ensuring that segmentation network resources are not wasted excessively.

[0072] It should be noted that different segmentation tasks correspond to different sub-networks, and different target segmentation sub-networks can be selected for texts with different segmentation tasks. For texts with similarity, the segmentation tasks they include are also similar, so the corresponding target segmentation sub-networks are generally likely to be the same or similar; for example, "kidney" and "kidney tumor" are essentially similar texts, and when segmenting tasks such as "kidney" or "kidney tumor" in the text, it is very likely that a similar segmentation sub-network will be selected.

[0073] S206, determine the image segmentation network based on the preset common segmentation network and target segmentation sub-network; the aforementioned common segmentation network is used to segment medical images and related tasks in other images.

[0074] The preset common segmentation network can be a task that segments related tasks in various medical images, such as feature extraction, feature map sampling (e.g., downsampling and / or upsampling), convolution processing, etc. For these tasks that are related in each segmentation task, they can be tasks that need to be performed regardless of the segmentation task, and can be referred to as common segmentation tasks. Then, these common segmentation tasks can be set as common segmentation networks. After obtaining the target segmentation subnetwork that matches the target segmentation task, the common segmentation network and the determined target segmentation subnetwork can be combined to form an image segmentation network. This image segmentation network can be a segmentation network specifically for segmenting the target segmentation network.

[0075] To combine a common segmentation network and a defined target segmentation subnetwork to form an image segmentation network, the common segmentation network and the target segmentation subnetwork can be combined in the order in which the segmentation network performs the target segmentation task to obtain the image segmentation network.

[0076] The above-mentioned segmentation tasks or common segmentation tasks in each segmented image are represented by a common segmentation network. In this way, different segmentation tasks can share the first common segmentation network, thereby reducing the cost of setting up a corresponding segmentation network for each segmentation task and improving the versatility of the determined image segmentation network.

[0077] S208 uses an image segmentation network to segment medical images and determines the target segmentation result of the medical images.

[0078] In this step, after obtaining the image segmentation network for the target segmentation task, this network can be used to perform the target segmentation task on the medical image, ultimately obtaining the target segmentation result corresponding to the target segmentation task. For example, if the target segmentation task in the input interactive text is "left lung" and the input medical image is a chest CT image, then the final target segmentation result is a segmentation mask for the left lung. As another example, the target segmentation task can also be a feature of the target to be segmented. For instance, if the target segmentation task in the input interactive text is "filling defects in the pulmonary artery" and the input medical image is a chest CT image, then the final target segmentation result is the segmentation result for pulmonary embolism.

[0079] As described above, by matching appropriate segmentation subnetworks to different segmentation tasks and combining each segmentation subnetwork with a common segmentation network that segments related tasks in various medical images, image segmentation networks corresponding to different segmentation tasks can be obtained, improving the versatility of the final image segmentation network. At the same time, different segmentation tasks can also learn from each other, thereby promoting the performance improvement of the image segmentation network through different segmentation tasks, thus improving the segmentation performance of the image segmentation network.

[0080] It should be noted that the method of this application embodiment can be applied to the segmentation of medical images, such as the segmentation of multiple organ structures and various regions containing lesions in an image using a multi-scan application. Taking lung lesion segmentation as an example, it can achieve the segmentation function of multiple tasks involving organs and lesions. In addition, in this embodiment, by fusing instruction sets and other modal information such as text (e.g., the aforementioned interactive input text), the name or specific features of the organ or lesion to be segmented can be input, and the corresponding segmentation results can be obtained quickly and accurately.

[0081] In the aforementioned image segmentation method, medical images and interactive text indicating the target segmentation task to be performed on the medical images are acquired. Then, a target segmentation sub-network matching the target segmentation task is determined based on the interactive text. Finally, an image segmentation network is determined based on a preset common segmentation network and the target segmentation sub-network. The image segmentation network is then used to segment the medical images to obtain the target segmentation result. The common segmentation network is used to segment related tasks in the medical images and other images. This method allows for the segmentation of medical images with different segmentation tasks or multiple segmentation tasks by combining different segmentation sub-networks with the common segmentation network to obtain different image segmentation networks for segmenting the medical images. Furthermore, the image segmentation network incorporates a common segmentation network for segmenting related tasks in various images, thus improving the generalization performance of the image segmentation network.

[0082] The above embodiments mention that the target segmentation subnetwork corresponding to the target segmentation task can be determined through interactive text input by the user. The following embodiments illustrate a possible implementation of this process.

[0083] In another embodiment, another image segmentation method is provided, based on the above embodiments, such as... Figure 3 As shown, the above S204 may include the following steps:

[0084] S302, Encode the interactive text to determine the encoded text; the encoded text includes the target segmentation task.

[0085] In this step, after obtaining the interactive text, since the interactive text may not be in the computer's encoding language, it needs to be encoded to conform to the computer's encoding language form.

[0086] As an optional implementation, a language model can be used to encode the interactive text to determine the encoded text. The language model can be a pre-trained model; the specific training process is not described here. In short, a trained language model can perform text understanding, analysis, encoding, and other processing, ultimately generating encoded text. The language model here could be, for example, a Transformer model.

[0087] In this step, after obtaining the interactive text input by the user, the interactive text can be input into the language model for encoding processing. The interactive text is encoded to obtain the encoded text, which is denoted as the encoded text. The encoded text here can consist of one or more text vectors, and it can include the target segmentation task to be performed on the medical image.

[0088] S304, the encoded text is processed by selecting a segmentation subnetwork in multiple preset routing modules to determine the target segmentation subnetwork that matches the target segmentation task; each routing module includes at least one candidate segmentation subnetwork.

[0089] Each routing module includes at least one candidate segmentation subnetwork. These candidate segmentation subnetworks can be the same or different across modules. If a routing module includes multiple candidate segmentation subnetworks, these subnetworks can be different. For example, they can have different network structures, such as using different types of convolutional blocks, using the same type of convolutional blocks but different numbers, or using different types and numbers of convolutional blocks, or other variations. In short, each routing module can include multiple different candidate segmentation subnetworks, or it can include only one.

[0090] In addition, for each of the above routing modules, the candidate segmentation subnetworks included, different routing modules and their different combinations of candidate segmentation subnetworks generally correspond to different segmentation tasks. Usually, the simpler the segmentation task, the simpler the segmentation network structure.

[0091] After obtaining the encoded text, it can be matched with candidate segmentation subnetworks in multiple preset routing modules to perform segmentation tasks. For example, the encoded text can be matched with each routing module simultaneously, or the encoded text can be first input into the first routing module for matching, and then the result from the first routing module can be combined with the encoded text for matching in the second routing module, and so on, until all routing modules are matched. Finally, the target segmentation subnetwork matching the target segmentation task in the encoded text is obtained. This target segmentation subnetwork can consist of candidate segmentation subnetworks from one routing module or candidate segmentation subnetworks from multiple routing modules.

[0092] The above-mentioned process of selecting segmentation subnetworks in multiple preset routing modules can be determined by the correspondence saved when the image segmentation network is pre-trained. This correspondence can include different segmentation tasks and their corresponding network paths, and the network paths include the candidate segmentation subnetworks selected in each routing module.

[0093] The training process of an image segmentation network can be as follows: First, obtain samples and their segmentation masks, and obtain the sample input text. Then, input the sample input text into each routing module of the initial image segmentation network. A softmax layer is used for network path selection. Segmentation is performed using the selected path to obtain the segmentation result corresponding to the sample input text. Next, the loss of the segmentation result and the corresponding segmentation mask under the selected path is calculated, and the loss is backpropagated to train the weights of the initial image segmentation network and the weights of each routing module, ultimately obtaining the trained image segmentation network.

[0094] The above correspondence is obtained by associating each sample input text with the corresponding optimal network path during the training process. The optimal network path here is the candidate segmentation sub-network selected from each routing module that has the best segmentation effect on the corresponding sample input text.

[0095] In this embodiment, after obtaining the interactive text, it can be encoded to obtain encoded text that includes the target segmentation task. Then, the encoded text undergoes sub-network matching processing within multiple preset routing modules to obtain the target segmentation sub-network that matches the target segmentation task. Through encoding and candidate sub-network matching within multiple routing modules, the corresponding target segmentation sub-network can be matched quickly and accurately, improving the efficiency and accuracy of image segmentation. Furthermore, encoding the input interactive text to obtain encoded text facilitates faster execution of subsequent segmentation sub-network matching processes by the computer, further improving the efficiency of image segmentation.

[0096] The above embodiments mention that a target segmented subnetwork can be obtained by matching in multiple routing modules. The following embodiments illustrate a possible implementation of this process.

[0097] In another embodiment, another image segmentation method is provided, based on the above embodiments, such as... Figure 4 As shown, the above S304 may include the following steps:

[0098] S402, input the encoded text into each routing module for path selection processing, and determine the candidate segmented subnetwork selected in each routing module.

[0099] In this step, the routing module refers to the module capable of selecting network paths; see [link / reference]. Figure 5 As shown, each routing module (Router) can include at least one network path, and each network path corresponds to a candidate segmented subnetwork. The candidate segmented subnetworks on different network paths can be different segmented subnetworks. For details, please refer to the different candidate segmented subnetworks in S304 above.

[0100] The interactive text can be encoded first. After obtaining the encoded text containing the target segmentation task, the encoded text can be input into each routing module for matching. Each routing module will then match suitable candidate segmentation subnetworks. For example, see... Figure 6 The diagram shows the architecture of the image segmentation network. The adaptive routing module is the routing module, and the Prompt is the encoded text. The Prompt can be input into each adaptive routing module for network path matching, and finally the candidate segmentation subnetworks matched or selected in each adaptive routing module are obtained.

[0101] Specifically, when matching in each routing module, candidate segmentation subnetworks can be selected using statistical, sparse, or statistical network paths. For example, different network paths corresponding to different segmentation tasks can be pre-set. After obtaining the encoded text of different segmentation tasks, the corresponding network paths can be matched in each routing module, thus obtaining the candidate segmentation subnetworks matched in each routing module.

[0102] S404 determines the candidate segmented subnet selected from each routing module as the target segmented subnet.

[0103] In this step, after selecting or matching suitable candidate segmentation subnetworks in each routing module, the selected candidate segmentation subnetworks can be combined to obtain the target segmentation subnetwork corresponding to the target segmentation task. Generally, each routing module will select a corresponding candidate segmentation subnetwork here.

[0104] Specifically, when combining the candidate segmented subnetworks, as an optional embodiment, the candidate segmented subnetworks selected in each routing module can be connected according to the connection relationship between the corresponding routing modules to obtain the target segmented subnetwork.

[0105] When combining candidate segmented subnetworks, the connection relationships between routing modules can be obtained in advance. Then, according to these connections, the candidate segmented subnetworks selected by each routing module can be connected to obtain the target segmented subnetwork. For example, see here. Figure 6 The image segmentation network architecture diagram shown illustrates that the candidate segmentation subnetwork selected in the bottom routing module has its output connected to the input of the candidate segmentation subnetwork selected in the routing module of the previous layer. Then, the output of the candidate segmentation subnetwork selected in the routing module of the previous layer is connected to the input of the candidate segmentation subnetwork selected in the routing module of the layer above that, and so on. This process can be repeated to connect the candidate segmentation subnetworks selected in each routing module to obtain the target segmentation subnetwork.

[0106] It should be noted that different segmentation tasks have different requirements for the functionality and performance of the network model. For example, lung field and heart segmentation can be completed with a simple network model structure, while more intricate structures like lung vessels require a relatively complex network model structure. Furthermore, different segmentation tasks also have different requirements for the accuracy and performance of the network model. This application's embodiments introduce a multi-routing module selection mechanism, where the complexity of the network model varies depending on the network path. Each routing module can include different candidate segmentation sub-networks, and each candidate segmentation sub-network can be a network path. The number of parameters, task functions, or achievable segmentation accuracy in each network path can be different. This allows segmentation sub-networks with different parameter numbers, task functions, and segmentation accuracies to be combined with a common segmentation network to form a general network model. Training this general network model simplifies the training of the image segmentation network.

[0107] In this embodiment, encoded text is input into each routing module for path selection to obtain candidate segmentation subnetworks selected in each module. These candidate subnetworks are then combined to determine the target segmentation subnetwork. By selecting candidate subnetworks corresponding to the encoded text in each routing module, the efficiency of determining the final target segmentation subnetwork is improved. Furthermore, connecting the selected candidate subnetworks according to the connection relationships between the corresponding routing modules ensures a fast and accurate acquisition of the target segmentation subnetwork. Additionally, providing more optional segmentation subnetworks improves the segmentation performance of the network model.

[0108] The above embodiments mention that a common segmentation network and a target segmentation task determined for different segmentation tasks can be combined to obtain a corresponding image segmentation network. The following embodiments will describe part of the structure of the common segmentation network and the target segmentation sub-network.

[0109] In another embodiment, based on the above embodiments, the common segmentation network includes multiple downsampling modules, each used to perform downsampling processing on the medical image at different resolutions; the target segmentation sub-network includes multiple segmentation sub-networks, each used to perform upsampling processing on the medical image at different resolutions.

[0110] See also Figure 6 The image segmentation network architecture shown is composed of multiple Transformer Layers and multiple Conv Blocks, which together form a common segmentation network. The Transformer Layers, which can be called downsampling modules, are mainly used to downsample the input medical images at different resolutions. That is, the resolution of the downsampling process performed by each Transformer Layer on the medical image can be different. The Conv Blocks are mainly used to extract features from the medical image or from the downsampled medical image to obtain downsampled feature maps.

[0111] Furthermore, the candidate segmentation sub-networks selected in each routing module (i.e., the adaptive routing in the diagram) can perform upsampling processing on the downsampled feature maps output by each Conv Block in the common segmentation network at corresponding resolutions, obtaining upsampled feature maps of different resolutions. The input of each routing module can include the upsampling result from the previous layer and the downsampling result from the current layer (for example, connection C in the diagram can combine the downsampling result from the current layer and the upsampling result from the previous layer), allowing subsequent network modules to incorporate the inputs of earlier network modules. The more features incorporated, the more accurate the final segmentation result.

[0112] It should be noted that the downsampling resolution and upsampling resolution of each layer can generally be the same. For example, if the figure includes 5 Transformer Layers and the resolution of the input medical image is 512*512, then after 5 downsampling processes, the obtained resolutions from top to bottom are: 512*512, 256*256, 128*128, 64*64, and 32*32. The corresponding upsampling resolutions from bottom to top are: 32*32, 64*64, 128*128, 256*256, and 512*512.

[0113] Having described the architecture of the common segmentation network and the target segmentation subnetwork above, the common segmentation network and the target segmentation subnetwork can be combined to obtain the target segmentation subnetwork. As an optional embodiment, S206 above may include:

[0114] Each downsampling module of the common segmentation network is connected to the corresponding segmentation subnetwork in the target segmentation subnetwork to obtain the image segmentation network.

[0115] In this step, the aforementioned common segmentation network can be used as an encoder, and the target segmentation sub-network can be used as a decoder; the two are used in combination. That is, after obtaining the various downsampling modules and convolutional modules of the common segmentation network, they can be used as follows: Figure 6 The method shown adaptively connects each downsampling module and convolutional module in the common segmentation network with the corresponding routing module. That is, the outputs of each downsampling module and convolutional module in the common segmentation network are used as the inputs of the corresponding routing modules. Combined with the outputs of the previous layers, a combined image segmentation network is finally obtained.

[0116] Additionally, it should be noted that each of the above routing modules is similar to a multiplexer with multiple network paths. After obtaining and encoding the input interactive text, the appropriate candidate segmentation subnetwork can be selected from each routing module through the encoded text. The selected candidate segmentation subnetwork can be considered as closing the corresponding switch in its corresponding routing module. The path of the selected candidate segmentation subnetwork can be connected or energized with the common segmentation network and the routing modules of the preceding and following layers to obtain the image segmentation network for the target segmentation task.

[0117] In this embodiment, the common segmentation network includes multiple downsampling modules that perform downsampling processing on medical images at different resolutions, and the target segmentation subnetwork includes multiple segmentation subnetworks that upsample medical images. This clarifies the functions of the common segmentation network and the target segmentation subnetwork, resulting in a more accurate image segmentation network. Furthermore, connecting the downsampling modules in the common segmentation network with the corresponding segmentation subnetworks in the target segmentation subnetwork yields the image segmentation network itself. This allows for the rapid construction of an image segmentation network tailored to a specific segmentation task, thereby improving image segmentation efficiency.

[0118] In actual image segmentation, the image segmentation result obtained in one segmentation may not meet the user's needs well. Therefore, it is necessary to perform multiple segmentations. The following examples illustrate how to perform multiple segmentations.

[0119] In another embodiment, another image segmentation method is provided, based on the above embodiments, such as... Figure 7 As shown, the above S208 may include the following steps:

[0120] S502 uses an image segmentation network to segment medical images and determines the initial segmentation results of the medical images.

[0121] In this step, after obtaining the encoded text of the interactive text, suitable candidate segmentation subnetworks can be selected from each routing module through the interactive text. These subnetworks are then connected to a common segmentation network to obtain an image segmentation network for the target segmentation task within the encoded text. The medical image can then be input into this image segmentation network for segmentation processing to obtain the segmentation result. The first segmentation result obtained here can be recorded as the initial segmentation result.

[0122] S504, if the initial segmentation result does not meet the segmentation requirements, then obtain new interactive text; the aforementioned new interactive text is text set for the local segmentation result or the overall segmentation result that does not meet the segmentation requirements in the initial segmentation result.

[0123] In this step, the segmentation requirements can be set according to the actual segmentation situation, such as the specific segmentation accuracy that the segmentation results need to achieve.

[0124] After obtaining the initial segmentation results, it can be determined manually based on experience whether the initial segmentation results meet the segmentation requirements, or it can be determined automatically. If the initial segmentation results do not meet the segmentation requirements, the medical images need to be re-segmented.

[0125] When re-segmenting medical images, the specific segmentation task to be performed can be determined by the interactive text re-entered by the user. That is, if the initial segmentation result is determined not to meet the segmentation requirements, the user can re-enter new interactive text. This new interactive text can include the requirements for re-segmenting the medical image, such as text set for local segmentation results or overall segmentation results that do not meet the segmentation requirements in the initial segmentation result. For example, the user can enter new interactive text such as "precise segmentation" or "re-segmentation".

[0126] S506, return to the above steps of determining the target segmentation subnetwork that matches the target segmentation task based on the interactive text, until the new initial segmentation result meets the segmentation requirements and the target segmentation result is obtained.

[0127] In this step, after obtaining the new interactive text input by the user, the process can return to steps S204-S208. That is, the new interactive text can be input into each routing module for matching or selection, obtaining the selected candidate segmentation subnetworks, and then obtaining a new target segmentation subnetwork. This new target segmentation subnetwork is then combined with the common segmentation network to obtain a new image segmentation network. The medical image is then re-input into this new image segmentation network for segmentation processing, yielding a new segmentation result. Afterward, it can be determined whether the new segmentation result meets the segmentation requirements. If it does, this result is taken as the final target segmentation result. If it does not meet the requirements, new interactive text is obtained again, and steps S204-S208 are repeated to finally obtain the target segmentation result.

[0128] In this embodiment, the medical image is segmented using an image segmentation network to obtain an initial segmentation result. If the initial segmentation result does not meet the segmentation requirements, new interactive text is obtained, and the steps of determining the target segmentation sub-network and the image segmentation network based on the interactive text and segmenting the medical image are repeated until the segmentation requirements are met and the target segmentation result is obtained. Here, the interactive text can be repeatedly input and the segmentation result can be continuously adjusted, making it easier to achieve fine segmentation of medical images and improve the user experience.

[0129] The following detailed embodiment illustrates the technical solution of this application. Based on the above embodiment, as follows... Figure 8 As shown, the method may include the following steps:

[0130] S1, Obtain the input interactive text and medical image; the interactive text is used to indicate the target segmentation task to be performed on the medical image;

[0131] S2, Encode the interactive text to determine the encoded text; the encoded text includes the target segmentation task.

[0132] S3, input the encoded text into each routing module for path selection processing, and determine the candidate segmented sub-networks selected in each routing module;

[0133] S4 connects the candidate segmented subnets selected in each routing module according to the connection relationship between the corresponding routing modules to obtain the target segmented subnet;

[0134] S5, connect each downsampling module of the common segmentation network to the corresponding segmentation subnetwork in the target segmentation subnetwork to obtain the image segmentation network;

[0135] S6. Use an image segmentation network to segment the medical image and determine the initial segmentation result of the medical image;

[0136] S7: Determine whether the initial segmentation result meets the segmentation requirements. If not, return to execute S1-S7; otherwise, execute S8.

[0137] S8, take the initial segmentation result as the target segmentation result.

[0138] As described above, the image segmentation network in this embodiment comprises two parts: a common segmentation network, which mainly segments or processes related tasks in various medical images; and an adaptively selected target segmentation subnetwork. Specifically, the candidate segmentation subnetwork selected by each target segmentation subnetwork is determined by the input interactive text. Different interactive texts represent different segmentation targets, allowing different network paths to be selected in each routing module based on different interactive texts, thus achieving different segmentation tasks. This allows related segmentation tasks to share more parameters while minimizing interference between unrelated segmentation tasks, thereby obtaining image segmentation networks tailored to different segmentation tasks and improving the segmentation performance of the image segmentation network.

[0139] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0140] Based on the same inventive concept, this application also provides an image segmentation apparatus for implementing the image segmentation method described above. The solution provided by this apparatus is similar to the implementation described in the above method; therefore, the specific limitations in one or more image segmentation apparatus embodiments provided below can be found in the limitations of the image segmentation method described above, and will not be repeated here.

[0141] In one embodiment, such as Figure 9 As shown, an image segmentation apparatus is provided, comprising: an acquisition module, a segmentation sub-network determination module, an image segmentation network determination module, and a segmentation module, wherein:

[0142] The acquisition module is used to acquire the input interactive text and medical images; the interactive text is used to indicate the target segmentation task to be performed on the medical images.

[0143] The segmentation subnetwork determination module is used to determine the target segmentation subnetwork that matches the target segmentation task based on the interactive text.

[0144] The image segmentation network determination module is used to determine the image segmentation network based on a preset common segmentation network and a target segmentation subnetwork; the aforementioned common segmentation network is used to segment medical images and related tasks in other images.

[0145] The segmentation module is used to segment medical images using an image segmentation network and determine the target segmentation result of the medical images.

[0146] In another embodiment, another image segmentation apparatus is provided. Based on the above embodiments, the segmentation sub-network determination module may include:

[0147] The encoding unit is used to encode the interactive text and determine the encoded text; the encoded text includes the target segmentation task.

[0148] The selection unit is used to select the segmentation subnetwork from multiple preset routing modules for the encoded text, and to determine the target segmentation subnetwork that matches the target segmentation task; each routing module includes at least one candidate segmentation subnetwork.

[0149] Optionally, the aforementioned encoding unit is specifically used to encode the interactive text using a language model to determine the encoded text.

[0150] In another embodiment, another image segmentation apparatus is provided. Based on the above embodiments, the selection unit may include:

[0151] The selection sub-unit is used to input encoded text into each routing module for path selection processing and to determine the candidate segmented sub-networks selected in each routing module.

[0152] The subnetwork determination sub-unit is used to determine the candidate segmented subnetwork selected from each routing module as the target segmented subnetwork.

[0153] Optionally, the aforementioned sub-network determining sub-unit is specifically used to connect the candidate segmented sub-networks selected in each routing module according to the connection relationship between the corresponding routing modules to obtain the target segmented sub-network.

[0154] In another embodiment, based on the above embodiments, the common segmentation network includes multiple downsampling modules, each used to perform downsampling processing on the medical image at different resolutions; the target segmentation sub-network includes multiple segmentation sub-networks, each used to perform upsampling processing on the medical image at different resolutions.

[0155] In another embodiment, another image segmentation apparatus is provided. Based on the above embodiments, the image segmentation network determination module is specifically used to connect each downsampling module of the common segmentation network to the corresponding segmentation subnetwork in the target segmentation subnetwork to obtain the image segmentation network.

[0156] In another embodiment, another image segmentation apparatus is provided. Based on the above embodiments, the segmentation module may include:

[0157] The segmentation unit is used to segment medical images using an image segmentation network and determine the initial segmentation result of the medical images;

[0158] The acquisition unit is used to acquire new interactive text if the initial segmentation result does not meet the segmentation requirements; the new interactive text is the text set for the local segmentation result or the overall segmentation result that does not meet the segmentation requirements in the initial segmentation result.

[0159] The return unit is used to return to the steps of determining the target segmentation subnetwork that matches the target segmentation task based on the interactive text, until the new initial segmentation result meets the segmentation requirements and the target segmentation result is obtained.

[0160] Each module in the aforementioned image segmentation device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the operations corresponding to each module.

[0161] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0162] The process involves acquiring input interactive text and medical images; the interactive text indicates the target segmentation task to be performed on the medical images; determining a target segmentation subnetwork that matches the target segmentation task based on the interactive text; determining an image segmentation network based on a preset common segmentation network and the target segmentation subnetwork; the common segmentation network is used to segment the medical images for tasks related to other images; and segmenting the medical images using the image segmentation network to determine the target segmentation result of the medical images.

[0163] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0164] The interactive text is encoded to determine the encoded text; the encoded text includes the target segmentation task; the encoded text is processed by selecting a segmentation subnetwork in multiple preset routing modules to determine the target segmentation subnetwork that matches the target segmentation task; each routing module includes at least one candidate segmentation subnetwork.

[0165] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0166] The encoded text is input into each routing module for path selection processing to determine the candidate segmented subnetworks selected in each routing module; the candidate segmented subnetworks selected in each routing module are then determined as the target segmented subnetwork.

[0167] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0168] The candidate segmented subnetworks selected from each routing module are connected according to the connection relationship between the corresponding routing modules to obtain the target segmented subnetwork.

[0169] In one embodiment, the aforementioned public segmentation network includes multiple downsampling modules, each used to perform downsampling processing on the medical image at different resolutions; the aforementioned target segmentation sub-network includes multiple segmentation sub-networks, each used to perform upsampling processing on the medical image at different resolutions.

[0170] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0171] Each downsampling module of the common segmentation network is connected to the corresponding segmentation subnetwork in the target segmentation subnetwork to obtain the image segmentation network.

[0172] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0173] The interactive text is encoded using a language model to determine the encoded text.

[0174] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0175] The medical image is segmented using an image segmentation network to determine the initial segmentation result. If the initial segmentation result does not meet the segmentation requirements, a new interactive text is obtained. The new interactive text is set for the local segmentation result or the overall segmentation result that does not meet the segmentation requirements in the initial segmentation result. The process returns to the above steps of determining the target segmentation subnetwork that matches the target segmentation task based on the interactive text, until the new initial segmentation result meets the segmentation requirements and the target segmentation result is obtained.

[0176] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor:

[0177] The process involves acquiring input interactive text and medical images; the interactive text indicates the target segmentation task to be performed on the medical images; determining a target segmentation subnetwork that matches the target segmentation task based on the interactive text; determining an image segmentation network based on a preset common segmentation network and the target segmentation subnetwork; the common segmentation network is used to segment the medical images for tasks related to other images; and segmenting the medical images using the image segmentation network to determine the target segmentation result of the medical images.

[0178] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0179] The interactive text is encoded to determine the encoded text; the encoded text includes the target segmentation task; the encoded text is processed by selecting a segmentation subnetwork in multiple preset routing modules to determine the target segmentation subnetwork that matches the target segmentation task; each routing module includes at least one candidate segmentation subnetwork.

[0180] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0181] The encoded text is input into each routing module for path selection processing to determine the candidate segmented subnetworks selected in each routing module; the candidate segmented subnetworks selected in each routing module are then determined as the target segmented subnetwork.

[0182] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0183] The candidate segmented subnetworks selected from each routing module are connected according to the connection relationship between the corresponding routing modules to obtain the target segmented subnetwork.

[0184] In one embodiment, the aforementioned public segmentation network includes multiple downsampling modules, each used to perform downsampling processing on the medical image at different resolutions; the aforementioned target segmentation sub-network includes multiple segmentation sub-networks, each used to perform upsampling processing on the medical image at different resolutions.

[0185] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0186] Each downsampling module of the common segmentation network is connected to the corresponding segmentation subnetwork in the target segmentation subnetwork to obtain the image segmentation network.

[0187] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0188] The interactive text is encoded using a language model to determine the encoded text.

[0189] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0190] The medical image is segmented using an image segmentation network to determine the initial segmentation result. If the initial segmentation result does not meet the segmentation requirements, a new interactive text is obtained. The new interactive text is set for the local segmentation result or the overall segmentation result that does not meet the segmentation requirements in the initial segmentation result. The process returns to the above steps of determining the target segmentation subnetwork that matches the target segmentation task based on the interactive text, until the new initial segmentation result meets the segmentation requirements and the target segmentation result is obtained.

[0191] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, performs the following steps:

[0192] The process involves acquiring input interactive text and medical images; the interactive text indicates the target segmentation task to be performed on the medical images; determining a target segmentation subnetwork that matches the target segmentation task based on the interactive text; determining an image segmentation network based on a preset common segmentation network and the target segmentation subnetwork; the common segmentation network is used to segment the medical images for tasks related to other images; and segmenting the medical images using the image segmentation network to determine the target segmentation result of the medical images.

[0193] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0194] The interactive text is encoded to determine the encoded text; the encoded text includes the target segmentation task; the encoded text is processed by selecting a segmentation subnetwork in multiple preset routing modules to determine the target segmentation subnetwork that matches the target segmentation task; each routing module includes at least one candidate segmentation subnetwork.

[0195] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0196] The encoded text is input into each routing module for path selection processing to determine the candidate segmented subnetworks selected in each routing module; the candidate segmented subnetworks selected in each routing module are then determined as the target segmented subnetwork.

[0197] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0198] The candidate segmented subnetworks selected from each routing module are connected according to the connection relationship between the corresponding routing modules to obtain the target segmented subnetwork.

[0199] In one embodiment, the aforementioned public segmentation network includes multiple downsampling modules, each used to perform downsampling processing on the medical image at different resolutions; the aforementioned target segmentation sub-network includes multiple segmentation sub-networks, each used to perform upsampling processing on the medical image at different resolutions.

[0200] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0201] Each downsampling module of the common segmentation network is connected to the corresponding segmentation subnetwork in the target segmentation subnetwork to obtain the image segmentation network.

[0202] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0203] The interactive text is encoded using a language model to determine the encoded text.

[0204] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0205] The medical image is segmented using an image segmentation network to determine the initial segmentation result. If the initial segmentation result does not meet the segmentation requirements, a new interactive text is obtained. The new interactive text is set for the local segmentation result or the overall segmentation result that does not meet the segmentation requirements in the initial segmentation result. The process returns to the above steps of determining the target segmentation subnetwork that matches the target segmentation task based on the interactive text, until the new initial segmentation result meets the segmentation requirements and the target segmentation result is obtained.

[0206] It should be noted that all data involved in this application (including but not limited to data used for analysis, data stored, data displayed, etc.) are data that have been fully authorized by all parties.

[0207] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0208] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0209] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. An image segmentation method, characterized in that, The method includes: The system acquires input interactive text and medical images; the interactive text is used to indicate the target segmentation task to be performed on the medical images; the interactive text is input by the user. The interactive text is encoded to obtain encoded text that includes the target segmentation task; The encoded text is input into each routing module for path selection processing to obtain candidate segmented subnetworks selected in each routing module. The candidate segmented subnetworks are then connected according to the connection relationship between the corresponding routing modules to obtain the target segmented subnetwork, which includes multiple segmented subnetworks. An image segmentation network is determined based on a preset common segmentation network and the target segmentation sub-network; the common segmentation network is used to segment the medical image and other related tasks; the image segmentation network is obtained by combining the common segmentation network and the target segmentation sub-network in the order in which the segmentation network performs the target segmentation task. The medical image is segmented using the image segmentation network to determine the target segmentation result of the medical image.

2. The method according to claim 1, characterized in that, The encoded text includes the target segmentation task; each of the routing modules includes at least one candidate segmentation sub-network.

3. The method according to claim 1 or 2, characterized in that, The common segmentation network includes multiple downsampling modules, each of which is used to perform downsampling processing on the medical image at different resolutions; Each of the segmentation sub-networks is used to perform upsampling processing on the medical images at different resolutions.

4. The method according to claim 1, characterized in that, The interactive text can be obtained by recognizing and converting the user's voice input into text.

5. The method according to claim 1, characterized in that, The step of segmenting the medical image using the image segmentation network to determine the target segmentation result of the medical image includes: The medical image is segmented using the image segmentation network to determine the initial segmentation result of the medical image; If the initial segmentation result does not meet the segmentation requirements, a new interactive text is obtained; the new interactive text is text set for the local segmentation result or the overall segmentation result in the initial segmentation result that does not meet the segmentation requirements. The process returns to the step of encoding the interactive text to obtain encoded text including the target segmentation task. The encoded text is then input into each routing module for path selection processing to obtain candidate segmentation subnetworks selected in each routing module. The candidate segmentation subnetworks are then connected according to the connection relationship between the corresponding routing modules to obtain the target segmentation subnetwork. This process continues until a new initial segmentation result meets the segmentation requirements, and the target segmentation result is obtained.

6. An image segmentation apparatus, characterized in that, The device includes: An acquisition module is used to acquire input interactive text and medical images; the interactive text is used to indicate the target segmentation task to be performed on the medical images; the interactive text is input by the user; The segmentation subnetwork determination module is used to encode the interactive text to obtain encoded text including the target segmentation task; input the encoded text into each routing module for path selection processing to obtain candidate segmentation subnetworks selected in each routing module; connect each candidate segmentation subnetwork according to the connection relationship between the corresponding routing modules to obtain the target segmentation subnetwork, which includes multiple segmentation subnetworks; The image segmentation network determination module is used to determine an image segmentation network based on a preset common segmentation network and the target segmentation sub-network; the common segmentation network is used to segment the medical image and related tasks in other images; the image segmentation network is obtained by combining the common segmentation network and the target segmentation sub-network according to the order in which the segmentation network performs the target segmentation task. The segmentation module is used to segment the medical image using the image segmentation network and determine the target segmentation result of the medical image.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 5.

9. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 5.

Images