Target area delineation methods, model training methods, devices and equipment

Abstract

The present application discloses a target area delineation method, a model training method, an apparatus and equipment, the method comprising obtaining medical image data of a target area to be delineated; calling a trained image segmentation model, inputting medical image data to output corresponding target area delineation results and implant needle segmentation results; wherein the image segmentation model comprises a target area delineation network for outputting target area delineation results and an implant needle segmentation network for outputting implant needle segmentation results, the target area delineation network and the implant needle segmentation network share at least some parameters; the target area delineation results are corrected based on the set dose plan data and the implant needle segmentation results. The present application achieves fully automatic segmentation of the target area and the implant needle simultaneously by constructing a multi-task network model including a target area delineation network and an interpolation needle segmentation network; thereby improving the segmentation capability of the image segmentation model as a whole; the target area delineation results can be corrected based on the dose plan data and the implant needle segmentation results.

Description

Technical Field

[0001] The present application belongs to the field of image processing technology, and in particular relates to a target area delineation method, a model training method, a device and equipment. Background Art

[0002] Cervical cancer is the fourth most common cancer worldwide, with an estimated 528,000 new cases and 266,000 deaths annually. Incidence and mortality rates increase with age. Interstitial brachytherapy (ISBT) has emerged as a localized treatment approach, demonstrating superior targeted coverage for patients with paracervical cancer compared with intracavitary radiotherapy (ICRT) alone. Image-guided brachytherapy (IGBT) has been introduced to better localize the applicator.

[0003] IGBT uses an implant needle, which acts as an applicator, to directly insert the radiation source into the tumor, placing the radiation source directly inside or near the tumor, thereby concentrating the dose distribution and improving the treatment effect. However, in IGBT, clinical target area delineation is a time-consuming and labor-intensive process. In this process, radiotherapists often use the spatial position information provided by the three-dimensional model of the implant needle applicator to achieve accurate delineation. However, if the implant needle planning scheme is used, it is easy to have errors with the actual situation. When using interpolation needle implantation and reconstruction, errors in implant needle implantation are not easy to detect. Deep learning tools are good at handling complex structural segmentation problems and have the potential to significantly simplify clinical workflows.

[0004] However, current deep learning tools for brachytherapy target delineation still use a general workflow, without optimizing the network architecture for the specific anatomical information in this field. Furthermore, the 3D reconstruction of implant needles is not tied to the target delineation task, forgoing important learning features and resulting in poor performance of existing models. Furthermore, when reconstructing implant needles in 3D, doctors often need to open multiple software programs simultaneously, making it impossible to intuitively visualize the discrepancy between the interpolated needle predictions and theoretical results, thus impacting the efficiency and accuracy of target planning. Summary of the Invention

[0005] This application provides a target area delineation method, model training method, device and equipment, aiming to solve the problems of poor performance of existing models and low efficiency and accuracy in planning target areas.

[0006] In a first aspect, the present application provides a target area delineation method, the method comprising:

[0007] Acquiring medical image data of the target area to be delineated;

[0008] Calling a trained image segmentation model, inputting the medical image data to output corresponding target area delineation results and implant needle segmentation results; wherein the image segmentation model includes a target area delineation network for outputting target area delineation results and an implant needle segmentation network for outputting implant needle segmentation results, and the target area delineation network and the implant needle segmentation network share at least some parameters;

[0009] The target area delineation result is corrected based on the set dose plan data and the implant needle segmentation result.

[0010] In a second aspect, the present application provides an image segmentation model training method, the method comprising:

[0011] Acquire a medical sample image; the medical sample image includes a medical image of a true result of target area delineation and a true result of interpolation needle segmentation;

[0012] Inputting the medical sample image into the image segmentation model to be trained, so as to output the target area delineation prediction result through the first network and the interpolation needle segmentation prediction result through the second network;

[0013] The parameters of the image segmentation model are optimized based on the comparison results of the target area delineation actual results and the target area delineation prediction results, as well as the comparison results of the interpolation needle segmentation actual results and the interpolation needle segmentation prediction results, wherein the first network and the second network share at least some parameters.

[0014] In a third aspect, the present application provides a target area delineation device, comprising:

[0015] An image acquisition module is used to acquire target medical image data of the target area to be delineated;

[0016] a model calling module, configured to call a trained image segmentation model, input the medical image data, and output corresponding target area delineation results and implant needle segmentation results; wherein the image segmentation model includes a target area delineation network for outputting target area delineation results and an implant needle segmentation network for outputting implant needle segmentation results, and the target area delineation network and the implant needle segmentation network share at least some parameters;

[0017] A correction module is used to correct the target area delineation result based on the set dose plan data and the implant needle segmentation result.

[0018] In a fourth aspect, the present application provides an image segmentation model training device, the device comprising:

[0019] A sample acquisition module is used to acquire medical sample images; the medical sample images include medical images of the target area delineation true results and the interpolation needle segmentation true results;

[0020] A model training module is used to input the medical sample image into an image segmentation model to be trained, wherein the image segmentation model includes a first network and a second network, so as to output a target area delineation prediction result through the first network and an interpolation needle segmentation prediction result through the second network; based on the comparison results of the actual target area delineation result and the target area delineation prediction result, and the comparison results of the actual interpolation needle segmentation result and the interpolation needle segmentation prediction result, the parameters of the image segmentation model are optimized, wherein the target area delineation network and the interpolation needle segmentation network share at least some parameters.

[0021] In a fifth aspect, the present application provides an electronic device comprising a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the target area delineation method or the image segmentation model training method is implemented.

[0022] In a sixth aspect, a storage medium stores computer-readable instructions thereon, wherein the computer-readable instructions are executed by one or more processors to implement the target area delineation method as described above.

[0023] In the seventh aspect, a computer program product includes computer-readable instructions, which are stored in a storage medium. One or more processors of an electronic device read the computer-readable instructions from the storage medium, load and execute the computer-readable instructions, so that the electronic device implements the target area delineation method as described above.

[0024] Compared with the existing technology, the present application realizes the fully automatic segmentation of the target area and the implant needle at the same time by building a multi-task network model including a target area delineation network and an interpolation needle segmentation network; since the target area delineation network and the interpolation needle segmentation network are trained in a shared parameter manner, they can guide each other to improve their respective prediction performance without affecting the prediction ability of the two networks themselves, thereby improving the segmentation ability of the image segmentation model as a whole; the target area delineation results can be corrected based on the dose planning data and the implant needle segmentation results. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0026] Figure 1 A schematic diagram of the main process of the target area delineation method provided in an embodiment of the present application;

[0027] Figure 2 for Figure 1 A schematic diagram of a sub-process of the illustrated embodiment;

[0028] Figure 3 for Figure 1 A schematic diagram of a sub-process of the illustrated embodiment;

[0029] Figure 4 for Figure 1 A schematic diagram of a sub-process of the illustrated embodiment;

[0030] Figure 5 A schematic diagram of the main process of the image segmentation model training method provided in an embodiment of the present application;

[0031] Figure 6 A structural block diagram of a target area delineation device provided in an embodiment of the present application;

[0032] Figure 7 A structural block diagram of the image segmentation model training device provided in an embodiment of the present application;

[0033] Figure 8 This is a structural block diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0034] The following will be combined with the accompanying drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are part of the embodiments of this application, not all of them. Based on the embodiments of this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0035] It will be understood that when used in this specification and the appended claims, the terms “comprises” and “comprising” indicate the presence of described features, integers, steps, operations, elements and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof.

[0036] See also Figures 1 to 4 , Figure 1 The main flow diagram of a target area delineation method provided in an embodiment of the present application is shown. An embodiment of a target area delineation method of the present application includes the following steps S100 to S300:

[0037] S100: Acquire medical image data of a target area to be delineated.

[0038] In this embodiment, clinical cancer image data for target delineation is first acquired. Specifically, the cancer image data includes a number of cancer images and various related information, such as modality information, dimensional information, sequence UID, volume data, etc.

[0039] It is understood that modality information generally refers to the type or source of the image, such as CT (computed tomography), MRI (magnetic resonance imaging), X-ray, ultrasound, etc. Different modalities can provide different information about the patient's anatomical structure and pathological conditions.

[0040] In medical imaging, dimensional information typically refers to the spatial and temporal dimensions of an image. The spatial dimension can be two-dimensional (2D) or three-dimensional (3D), representing the planar or three-dimensional information within the image. The temporal dimension refers to the temporal changes within an image sequence, such as dynamic MRI or cardiac cine.

[0041] A sequence UID, or Unique Identifier, is an identifier used to uniquely identify an image sequence in medical images. UID is a key concept in the DICOM (Digital Imaging and Communications in Medicine) standard, ensuring interoperability and uniqueness of images across different systems and devices. Each DICOM image sequence has a unique UID that distinguishes one image sequence from another.

[0042] Volume data refers to three-dimensional medical image data, typically composed of a series of stacked two-dimensional images to form a three-dimensional image of anatomical structures. Volume data can contain a variety of information, such as pixel intensity, tissue type, and spatial location, and is used to analyze and visualize a patient's internal structure and pathological conditions.

[0043] S200. Call the trained image segmentation model and input medical image data to output the corresponding target area delineation results and implant needle segmentation results; wherein the image segmentation model includes a target area delineation network for outputting the target area delineation results and an implant needle segmentation network for outputting the implant needle segmentation results, and the target area delineation network and the implant needle segmentation network share at least some parameters.

[0044] In this embodiment, the acquired target clinical cancer image data is input into a preset image segmentation model to simultaneously output the corresponding target area delineation results and implant needle segmentation results. Specifically, the image segmentation model is a pre-trained model that is specifically used to identify and segment specific areas or objects from medical images. In this embodiment, the model is designed to handle two tasks simultaneously: target area delineation and implant needle segmentation. The target area delineation network is responsible for identifying and delineating tumors or areas requiring treatment (i.e., target areas) from medical images. The implant needle segmentation network is responsible for identifying and segmenting the implant needle portion in the image. The two networks each undertake different tasks, but share at least some parameters. Sharing parameters means that the target area delineation network and the implant needle segmentation network are interrelated. Sharing parameters can reduce the overall number of parameters of the model, thereby helping to prevent overfitting and improving the generalization ability of the model. Furthermore, the medical image data is output through the image segmentation model to include target area delineation images and interpolated needle segmentation images of various medical image data.

[0045] By building a multi-task deep learning model including a target area delineation network and an interpolation needle segmentation network, fully automatic segmentation of the target area and the implant needle can be achieved simultaneously; because the target area delineation network and the interpolation needle segmentation network are trained using shared parameters, they can guide each other to improve their respective prediction performance without affecting the prediction capabilities of the two networks themselves.

[0046] S300: Modify the target area delineation result based on the set dose plan data and the implant needle segmentation result.

[0047] In an exemplary embodiment, Figure 2 As shown, step S300 specifically includes steps S310 to S330:

[0048] S310: Acquire the dosage plan data.

[0049] Dose planning data in radiotherapy refers to the radiation dose distribution information determined for a specific patient's radiotherapy treatment plan. Dose planning data can be obtained through a combination of patient imaging data, target volume delineation, dose planning software, and dose calculation and optimization algorithms.

[0050] S320. Calculate the interpolation needle theoretical placement result according to the dose plan data.

[0051] Interpolated needle placement results can be calculated based on dose planning data. Using dose planning data, interpolated needle placement results employ mathematical models and algorithms to calculate dose values ​​at different locations within the target area. These calculations help radiotherapists understand the dose distribution under different placement schemes and select the optimal placement strategy.

[0052] S330 , correcting the target area delineation result by comparing the interpolation needle theoretical placement result with the interpolation needle segmentation result.

[0053] By comparing the actual interpolation needle segmentation results predicted by the image segmentation model with the calculated interpolation needle theoretical placement results, the target area delineation results obtained by the image segmentation model can be better corrected in accordance with the actual situation.

[0054] In an exemplary embodiment, Figure 3 As shown, step S330 may further include steps S331 to S333:

[0055] S331, constructing a three-dimensional model of the target area delineation results, the implantation needle segmentation results, and the interpolation needle theoretical placement results.

[0056] In this embodiment, a corresponding three-dimensional model can be established based on several target area delineation images and interpolation needle segmentation images output by the image segmentation model. Independent three-dimensional models can be established for the target area delineation results, the implantation needle segmentation results, and the interpolation needle theoretical placement results, respectively. Alternatively, a three-dimensional model can be established that includes the above-mentioned target area delineation results, implantation needle segmentation results, and interpolation needle theoretical placement results.

[0057] S332, displaying a display interface, wherein the display interface displays a three-dimensional image of the three-dimensional model;

[0058] In this embodiment, a three-dimensional model of the target area delineation results, the implant needle segmentation results, and the interpolation needle theoretical placement results is displayed through a display interface. The displayed image can be a three-dimensional stereogram or a cross-sectional view from a certain perspective, such as: a cross-section, a sagittal plane, and a coronal plane. Among them, the implant needle segmentation results and the ideal needle placement data are distinguished by different colors.

[0059] S333: In response to the first trigger operation, the target area delineation result is corrected through the three-dimensional image, and the corrected image is displayed through the display interface.

[0060] The three-dimensional image obtained in step S320 is used to visualize both the implant needle segmentation results and the theoretical interpolation needle placement results. This allows for intuitive comparison of the two results, and allows for further correction of the target area delineation output by the model through appropriate operations. For example, the above steps can be performed by clicking on an electronic device with a display screen. Furthermore, the specific operation method can be designed based on the actual use scenario and requirements.

[0061] The three-dimensional model is reconstructed by the obtained target area delineation results, implant needle segmentation results and interpolation needle theoretical placement results, and the model is intuitively observed through a visual display interface, and the implant needle segmentation results and interpolation needle theoretical placement results are distinguished; that is, when the implant needle segmentation results and interpolation needle theoretical placement results are compared and observed at the same time, the target area delineation results predicted by the model can be better corrected to improve work efficiency.

[0062] In an exemplary embodiment, Figure 4 As shown, step S100 may further include steps S110 to S120:

[0063] S110, performing image reconstruction on the acquired medical image data to generate a planar image of a set viewing angle;

[0064] S120: Displaying a display interface, where the display interface displays a plan view with a set viewing angle.

[0065] The acquired volume data of the target medical image data contains a large number of image slices. Based on the volume data and other information in the medical image data, the target medical image is planar reconstructed according to the set viewing angle to obtain a plan view under the set viewing angle, and the reconstructed image is displayed through a visual interface. In a specific embodiment, the set viewing angle image may include a cross-section, a sagittal plane, and a coronal plane. Among them, the cross-section represents a plane parallel to the horizontal plane, which can divide the human body into upper and lower parts; the sagittal plane represents a plane perpendicular to the horizontal plane, which can divide the human body into left and right parts; the coronal plane represents a plane perpendicular to the horizontal plane, which can divide the human body into front and back parts. Further, the plan view of the set viewing angle obtained in step S110 is input into the image segmentation model, and the corresponding target area delineation result can be obtained, and the target area delineation result is displayed in the plan view of the set viewing angle through a visual display interface. The volume data contains multi-dimensional information of the patient.

[0066] Multiplanar reconstruction allows doctors to view coronal, sagittal, and transverse images simultaneously on a computer screen, providing a comprehensive understanding of the location, size, and shape of the lesion.

[0067] In addition, a three-dimensional model with target area delineation results can be further constructed through the target area delineation results at various perspectives.

[0068] In an exemplary embodiment, Figure 4 As shown, step S120 may further include step S130:

[0069] S130 , in response to the second triggering operation, editing one of the plan views.

[0070] Based on the visual display, the displayed images can be edited through corresponding operations. For example, in the three perspectives of cross-section, sagittal, and coronal images, the image can be rotated, scaled, and moved, and the target position in one image can be selected and automatically positioned in the images of other perspectives.

[0071] The embodiment of the present application also provides a training method for an image segmentation model, wherein the image segmentation model includes a first network and a second network, such as Figure 5 As shown, the training method of the image segmentation model includes steps S10 to S30:

[0072] S10. Obtain a medical sample image; the medical sample image includes a medical image with a true result of target area delineation and a true result of interpolation needle segmentation.

[0073] In this embodiment, the label of the target area delineation result is the target area manually delineated by the doctor, and the label of the implant needle segmentation result is obtained by threshold segmentation of the CT image.

[0074] In certain embodiments, after acquiring a medical sample image, the sample image may be preprocessed. Preprocessing methods include, but are not limited to, cropping and scaling, random rotation, flipping, color change, and normalization of the sample image. The purpose of preprocessing is to eliminate irrelevant information in the sample image, restore useful real information, enhance the detectability of relevant information, and simplify the data, thereby improving the reliability of feature extraction, image segmentation, matching, and recognition.

[0075] S20. Input the medical sample image into the image segmentation model to be trained, so as to output the target area delineation prediction result through the first network and output the interpolation needle segmentation prediction result through the second network.

[0076] In this embodiment, the image segmentation model includes a first network for processing the target delineation task and a second network for the interpolation needle segmentation task. The first network is used to output a segmentation probability map for the target area, and the second network is used to output a segmentation probability map for the interpolation needle. The probability of each pixel in the segmentation map corresponds to two classification results: foreground and background, thereby achieving image segmentation. By establishing an image segmentation model based on multi-task deep learning, the target delineation and interpolation needle segmentation tasks are simultaneously completed.

[0077] S30. Optimizing the parameters of the image segmentation model based on the comparison results between the actual target delineation result and the predicted target delineation result, and the comparison results between the actual interpolation needle segmentation result and the predicted interpolation needle segmentation result, wherein the first network and the second network share at least some parameters.

[0078] The target area delineation actual result predicted by the model based on the above steps is compared with the target area delineation prediction result and the actual sample label result, and the predicted result is constrained by the corresponding loss function; and the above steps S10-S20 are repeated to minimize the loss function to optimize the parameters of the image segmentation model to achieve the purpose of training. Furthermore, the first network and the second network share some parameters to achieve the purpose of training the two networks at the same time, so that without affecting the prediction ability of the two networks themselves, they can guide each other to improve their respective prediction performance. In a specific embodiment, the target area delineation actual result and the target area delineation prediction result are constrained by the first loss function L ctv Constraints are performed, and the interpolation needle segmentation real results and the interpolation needle segmentation prediction results are compared through the second loss function L needles Constraints are performed, where the first loss function and the second loss function are shown in the following formulas (1) and (2):

[0079]

[0080]

[0081] Among them, L ctv Represents the first loss function, L needles Represents the second loss function, X, Y, Z represent the pixel coordinates of the input image, u represents the predicted probability, and v represents the true result.

[0082] In an exemplary embodiment, the first and second networks of the image segmentation model can adopt an encoder-decoder-based structure, that is, the first network includes a first encoder and a first decoder, and the second network includes a second encoder and a second decoder; for example, a U-net neural network, which is completely implemented by convolution, upsampling, and downsampling, and incorporates the characteristics of skip connections. The encoder is responsible for extracting the features of the input image layer by layer through multiple upsampling layers, while the decoder uses this feature information from multiple downsampling layers to restore the segmented image. For example, it includes three upsampling layers, each corresponding to a downsampling layer, and the scales of each sampling layer are 64*64*24 (length*width*number of channels), 16*16*12, and 4*4*6. That is, the target area delineation network and the pin segmentation network each include an encoder and a decoder, wherein the encoder parameters of the two models are shared through proportional mixing.

[0083] The following are embodiments of the device of the present application, which can be used to perform the target area delineation method involved in the present application. For details not disclosed in the embodiment of the device of the present application, please refer to the method embodiment of the target area delineation method involved in the present application.

[0084] like Figure 6As shown, an embodiment of the present application also provides a target area delineation device 150, which includes an image acquisition module 151, a model calling module 152 and a correction module 153; wherein, the image acquisition module 151 is used to acquire target medical image data of the target area to be delineated; the model calling module 152 is used to call a trained image segmentation model, input medical image data to output corresponding target area delineation results and implantation needle segmentation results; wherein, the image segmentation model includes a target area delineation network for outputting target area delineation results and an implantation needle segmentation network for outputting implantation needle segmentation results, and the target area delineation network and the implantation needle segmentation network share at least some parameters; the correction module 153 is used to correct the target area delineation result based on the set dose plan data and the implantation needle segmentation result.

[0085] In an exemplary embodiment, the correction module 153 also includes an acquisition submodule, a calculation submodule and a correction submodule. The acquisition submodule is used to acquire the dose plan data; the calculation submodule is used to calculate the interpolation needle theoretical placement result based on the dose plan data; the correction submodule corrects the target area delineation result by comparing the interpolation needle theoretical placement result with the interpolation needle segmentation result.

[0086] In an exemplary embodiment, the comparison submodule includes a construction submodule, a first response submodule and a display submodule. The construction submodule is used to construct a three-dimensional model of the target area delineation results, the insertion needle segmentation results and the interpolation needle theoretical placement results; the display submodule is used to display a display interface, which displays a three-dimensional image of the three-dimensional model and a corrected image. The first response submodule is used to respond to a first trigger operation to correct the target area delineation results.

[0087] In an exemplary embodiment, the target area delineation device 150 also includes a plane construction module and a second response submodule. The plane construction module is used to reconstruct the acquired medical image data to generate a plan view of a set perspective. The display interface also displays the plan view of the set perspective. The second response submodule is used to respond to a second trigger operation to edit one of the plan views.

[0088] like Figure 7As shown, an embodiment of the present application also provides an image segmentation model training device 160, which includes a sample acquisition module 161 and a model training module 162; wherein, the sample acquisition module 161 is used to acquire medical sample images; the medical sample images include medical images of the true results of target area delineation and the true results of interpolation needle segmentation; the model training module 162 is used to input the medical sample images into the image segmentation model to be trained, so as to output the target area delineation prediction results through the first network, and output the interpolation needle segmentation prediction results through the second network; based on the comparison results of the true results of target area delineation and the target area delineation prediction results and the comparison results of the true results of interpolation needle segmentation and the interpolation needle segmentation prediction results, the parameters of the image segmentation model are optimized, wherein the first network and the second network share at least some parameters.

[0089] like Figure 8 As shown, an embodiment of the present application further provides an electronic device 140, comprising a processor 141, a memory 142, a non-volatile memory 144, and a computer program 1441 stored in the non-volatile memory 144 and executable on the processor. When the processor 141 executes the computer program 1441, any embodiment of the target area delineation method and any embodiment of the image segmentation model training method described above are implemented. Specifically, the electronic device 140 further comprises an input / output interface 145 and an input / output device 146 connected thereto. The processor 141, the memory 142, the non-volatile memory 144, and the input / output interface 145 are connected via an internal bus 143.

[0090] It should be understood that in the embodiments of the present application, the processor may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0091] The present application also proposes a storage medium. In an exemplary embodiment, the storage medium stores at least one instruction, at least one program, code set or instruction set. When the at least one instruction, at least one program, code set or instruction set is executed by a processor of a computer device, it implements the target area delineation method of any of the above embodiments and the image segmentation model training method of any of the above embodiments.

[0092] Compared with related technologies, this application achieves fully automatic segmentation of the target area and the implant needle by building a multi-task network model including a target area delineation network and an interpolation needle segmentation network. Since the target area delineation network and the interpolation needle segmentation network are trained in a shared parameter manner, they can guide each other to improve their respective prediction performance without affecting the prediction capabilities of the two networks themselves, thereby improving the segmentation capabilities of the image segmentation model as a whole. The target area delineation results can be corrected based on the dose plan data and the implant needle segmentation results. Furthermore, a three-dimensional model can be constructed that includes the target area delineation results, the interpolation needle segmentation results, and the interpolation needle theoretical placement results. By visualizing the three-dimensional model, the difference between the interpolation needle prediction results and the theoretical results can be more intuitively displayed, thereby improving the efficiency and accuracy of the doctor's target area delineation.

[0093] Those skilled in the art will appreciate that all or part of the steps in the method of the above-described embodiment can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. The computer program is executed by at least one processor in the computer system to implement the steps in the method of the above-described embodiment.

[0094] It is understood that the present application is described by way of some embodiments, and it is understood by those skilled in the art that various changes or equivalent substitutions may be made to these features and embodiments without departing from the spirit and scope of the present application. In addition, under the guidance of the present application, these features and embodiments may be modified to adapt to specific circumstances and materials without departing from the spirit and scope of the present application. Therefore, the present application is not limited to the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of the present application are within the scope of protection of the present application.

Claims

1. A target area delineation method, characterized in that: The method comprises: Acquiring medical image data of the target area to be delineated; Calling a trained image segmentation model, inputting the medical image data to output corresponding target area delineation results and implant needle segmentation results; wherein the image segmentation model includes a target area delineation network for outputting target area delineation results and an implant needle segmentation network for outputting implant needle segmentation results, and the target area delineation network and the implant needle segmentation network share at least some parameters; Correcting the target area delineation result based on the set dose plan data and the implant needle segmentation result; The parameters of the target delineation network and the implant needle segmentation network are determined during the training process based on the comparison results between the target delineation results and the target delineation prediction results, as well as the comparison results between the interpolation needle segmentation results and the interpolation needle segmentation prediction results. The target area delineation actual result and the target area delineation predicted result are compared through the first loss function L ctv Constraints are performed, and the interpolation needle segmentation real results and the interpolation needle segmentation prediction results are compared through the second loss function L needles Constraints are performed, where the first loss function and the second loss function are shown in the following formulas: Among them, L ctv Represents the first loss function, L needles Represents the second loss function, X, Y, Z represent the pixel coordinates of the input image, u represents the predicted probability, and v represents the true result.

2. The method according to claim 1, characterized in that The modifying of the target area delineation result based on the set dose plan data and the implant needle segmentation result includes: obtaining the dosage plan data; Calculating interpolation needle theoretical placement results based on the dose plan data; The target area delineation result is corrected by comparing the interpolation needle theoretical placement result with the interpolation needle segmentation result.

3. The method according to claim 2, characterized in that The step of comparing the interpolation needle theoretical placement result with the interpolation needle segmentation result to correct the target area delineation result includes: Constructing a three-dimensional model of the target area delineation results, the implant needle segmentation results, and the interpolation needle theoretical placement results; displaying a display interface, wherein the display interface displays an image of the three-dimensional model; In response to the first trigger operation, the target area delineation result is corrected through the image, and the corrected image is displayed through the display interface.

4. The method according to claim 1, wherein After obtaining the medical image data of the target area to be delineated, the method further includes: Performing image reconstruction on the acquired medical image data to generate a planar image of a set viewing angle; A display interface is displayed, wherein the display interface displays the plan view of the set viewing angle.

5. The method according to claim 4, characterized in that After the display interface displays the plan view at the set viewing angle, the method further includes: In response to the second triggering operation, one of the plan views is edited, and the edited image is displayed through the display interface.

6. A method for training an image segmentation model, characterized in that: The image segmentation model includes a first network and a second network, and the method includes: Acquire a medical sample image; the medical sample image includes a medical image of a true result of target area delineation and a true result of interpolation needle segmentation; Inputting the medical sample image into the image segmentation model to be trained, so as to output the target area delineation prediction result through the first network and the interpolation needle segmentation prediction result through the second network; Optimizing parameters of the image segmentation model based on a comparison between the actual target delineation result and the predicted target delineation result, and a comparison between the actual interpolation needle segmentation result and the predicted interpolation needle segmentation result, wherein the first network and the second network share at least some parameters; Among them, the target area delineation actual result and the target area delineation prediction result are compared through the first loss function L ctv Constraints are performed, and the interpolation needle segmentation real results and the interpolation needle segmentation prediction results are compared through the second loss function L needles Constraints are performed, where the first loss function and the second loss function are shown in the following formulas: Among them, L ctv Represents the first loss function, L needles Represents the second loss function, X, Y, Z represent the pixel coordinates of the input image, u represents the predicted probability, and v represents the true result.

7. The method according to claim 6, characterized in that The first network includes a first encoder and a first decoder, and the second network includes a second encoder and a second decoder, wherein the first encoder and the second encoder share some or all parameters.

8. A target area delineation device, characterized in that: The device comprises: An image acquisition module is used to acquire target medical image data of the target area to be delineated; a model calling module, configured to call a trained image segmentation model, input the medical image data, and output corresponding target area delineation results and implant needle segmentation results; wherein the image segmentation model includes a target area delineation network for outputting target area delineation results and an implant needle segmentation network for outputting implant needle segmentation results, and the target area delineation network and the implant needle segmentation network share at least some parameters; A correction module, configured to correct the target area delineation result based on the set dose plan data and the implant needle segmentation result; The parameters of the target delineation network and the implant needle segmentation network are determined during the training process based on the comparison results between the target delineation results and the target delineation prediction results, as well as the comparison results between the interpolation needle segmentation results and the interpolation needle segmentation prediction results. The target area delineation actual result and the target area delineation predicted result are compared through the first loss function L ctv Constraints are performed, and the interpolation needle segmentation real results and the interpolation needle segmentation prediction results are compared through the second loss function L needles Constraints are performed, where the first loss function and the second loss function are shown in the following formulas: Among them, L ctv Represents the first loss function, L needles Represents the second loss function, X, Y, Z represent the pixel coordinates of the input image, u represents the predicted probability, and v represents the true result.

9. An image segmentation model training device, characterized in that: The device comprises: A sample acquisition module is used to acquire medical sample images; the medical sample images include medical images of the target area delineation true results and the interpolation needle segmentation true results; A model training module is configured to input the medical sample image into an image segmentation model to be trained, the image segmentation model comprising a first network and a second network, wherein the first network outputs a target delineation prediction result, and the second network outputs an interpolated needle segmentation prediction result; and optimize the parameters of the image segmentation model based on a comparison between the actual target delineation result and the predicted target delineation result, and a comparison between the actual interpolated needle segmentation result and the predicted interpolated needle segmentation result, wherein the target delineation network and the interpolated needle segmentation network share at least some parameters. Among them, the target area delineation actual result and the target area delineation prediction result are compared through the first loss function L ctv Constraints are performed, and the interpolation needle segmentation real results and the interpolation needle segmentation prediction results are compared through the second loss function L needles Constraints are performed, where the first loss function and the second loss function are shown in the following formulas: Among them, L ctv Represents the first loss function, L needles Represents the second loss function, X, Y, Z represent the pixel coordinates of the input image, u represents the predicted probability, and v represents the true result.

10. An electronic device, characterized in that: The electronic device includes a memory and a processor, the memory stores a computer program, and the processor implements the method according to any one of claims 1 to 7 when executing the computer program.

Images