Adaptive radiation delivery determination method, training method of a trigger analysis model, and storage medium

Abstract

This application relates to an adaptive radiodelivery determination method, a training method for a trigger analysis model, and a storage medium. The adaptive radiodelivery determination method includes: acquiring a current fraction image of the target object corresponding to the current radiodelivery fraction, and acquiring an initial image used by the target object when determining the initial radiodelivery plan; determining a trigger analysis result based on the current fraction image, the initial image, and the trigger analysis model; and determining whether to execute adaptive radiodelivery based on the trigger analysis result. By employing the adaptive radiodelivery determination method, an adaptive radiodelivery trigger judgment can be achieved through a machine learning model. This not only improves the efficiency of adaptive radiodelivery trigger judgment but also enhances the consistency of trigger judgment, avoiding different trigger judgment results generated when different users make trigger judgments based on different clinical experiences. This improves the accuracy and rationality of adaptive trigger judgment.

Description

Technical Field

[0001] This application relates to the field of radiographic delivery technology, and in particular to an adaptive radiographic delivery determination method, a training method for a trigger analysis model, and a storage medium. Background Technology

[0002] With the development of radiotherapy technology, image-guided radiotherapy (IGRT) and adaptive radiotherapy (ART) have become two important means of radiotherapy. Among them, adaptive radiotherapy planning (adaptive planning) is a radiotherapy method that readjusts the plan according to changes in tumor location, size, or disease progression during the patient's radiotherapy. It is a relatively complex and time-consuming process.

[0003] Traditionally, the method for determining whether an adaptive plan is triggered is for the user to analyze IGRT images at the treatment site based on clinical experience to determine whether the adaptive plan has been triggered.

[0004] However, traditional methods for determining adaptive plan triggering suffer from low efficiency. Summary of the Invention

[0005] Based on this, it is necessary to provide an adaptive radiation delivery determination method, a trigger analysis model training method, an apparatus, a computer device, a computer-readable storage medium, and a computer program product that can improve the judgment efficiency and rationality of adaptive plan triggering in order to address the above-mentioned technical problems.

[0006] In a first aspect, this application provides an adaptive radiation delivery determination method, the method comprising:

[0007] Obtain the current fraction image of the target object corresponding to the current radiographic delivery fraction, and obtain the initial image of the target object used when determining the initial radiographic delivery plan;

[0008] Based on the current segmented images, the initial image, and the trigger analysis model, determine the trigger analysis results;

[0009] Whether to perform adaptive radio delivery is determined based on the trigger analysis results.

[0010] In one embodiment, the trigger analysis results include trigger concern parameters and / or adaptive radiation delivery trigger determination information, wherein the trigger concern parameters include benefit information for one or more dose indices.

[0011] In one embodiment, the trigger analysis model includes a first trigger analysis model; determining the trigger analysis result based on the current segmented image, the initial image, and the trigger analysis model includes:

[0012] The current segmented image and the initial image are registered to obtain the registered image;

[0013] The region of interest in the registered image is delineated to obtain a delineated image, which includes the delineation result of the region of interest.

[0014] The drawn image is input into the first trigger analysis model, and the trigger analysis result is output.

[0015] In one embodiment, the first trigger analysis model includes a first feature extraction network and a trigger analysis network. The drawn image is input into the first trigger analysis model, and the trigger analysis result is output, including:

[0016] The outlined image and the initial image are input into the first feature extraction network for feature extraction to obtain target feature information;

[0017] The target feature information is input into the trigger analysis network, and the trigger analysis result is output. The initial image includes the original delineation result of the region of interest.

[0018] In one embodiment, the target feature information includes at least one of morphological difference feature information based on the region of interest, image difference feature information, and registration difference feature information.

[0019] In one embodiment, the trigger analysis model includes a second trigger analysis model; determining the trigger analysis result based on the current segmented image, the initial image, and the trigger analysis model includes:

[0020] Based on the original delineation result in the initial image, the region of interest in the current segmented image is delineated for the first time to obtain the first delineated image.

[0021] Perform a second delineation operation on the region of interest in the current segmented image to obtain the second delineated image;

[0022] The first and second outlined images are input into the second trigger analysis model, and the trigger analysis results are output.

[0023] In one embodiment, a first delineation operation is performed on the region of interest in the current segmented image based on the original delineation result in the initial image to obtain a first delineated image, including:

[0024] The original drawing result in the initial image is rigidly copied to the current segmented image to obtain the first drawing image containing the original drawing result.

[0025] In one embodiment, the second trigger analysis model includes a second feature extraction network and a decision network. The first outlined image and the second outlined image are input into the second trigger analysis model, and the trigger analysis result is output, including:

[0026] The first outlined image is input into the second feature extraction network for feature extraction to obtain the first feature information;

[0027] The second outlined image is input into the second feature extraction network for feature extraction to obtain the second feature information;

[0028] The first and second feature information are input into the decision network, and the trigger analysis results are output.

[0029] In one embodiment, the trigger analysis results include trigger interest parameters, and the method further includes:

[0030] The benefit information of each dose index in the triggering attention parameters is weighted and summed to obtain the comprehensive benefit information;

[0031] Based on comprehensive benefit information and preset benefit information thresholds, adaptive radioactive delivery triggering judgment information is determined.

[0032] Secondly, this application also provides a method for training a trigger analysis model, which is used for adaptive radiation delivery determination. This training method includes:

[0033] Acquire training sample data; the training sample data includes the initial sample images used when determining the initial radiographic delivery plan for different sample objects, the fractional sample images of the sample objects at different radiographic delivery fractions, and the corresponding sample trigger analysis results;

[0034] The initial trigger analysis network is trained based on the training sample data to obtain the trigger analysis model; the trigger analysis model is used to output the trigger analysis results to determine whether to perform adaptive radiodelivery.

[0035] In one embodiment, the sample-triggered analysis results include sample-triggered attention parameters, which include benefit information labels for one or more dose indicators; the method further includes:

[0036] For each fractional sample image, the difference in dose indices is obtained after performing image-guided radiodelivery planning and adaptive planning for the corresponding radiodelivery fractions of the fractional sample images.

[0037] The differences between each dose index are normalized to obtain benefit information labels for one or more dose indices corresponding to the sample images.

[0038] Thirdly, this application also provides an adaptive radiation delivery determination device, the device comprising:

[0039] The acquisition module is used to acquire the current fraction image of the target object corresponding to the current radiographic delivery fraction, and to acquire the initial image used by the target object when determining the initial radiographic delivery plan;

[0040] The first determining module is used to determine the trigger analysis result based on the current segmented image, the initial image, and the trigger analysis model;

[0041] The second determination module is used to determine whether to perform adaptive radio delivery based on the trigger analysis results.

[0042] Fourthly, this application also provides a training device for a trigger analysis model used for adaptive radiation delivery determination, the training device comprising:

[0043] The first acquisition module is used to acquire training sample data; the training sample data includes the initial sample images used when determining the initial radiographic delivery plan for different sample objects, the sample images of the sample objects at different radiographic delivery fractions, and the corresponding sample trigger analysis results.

[0044] The training module is used to train the initial trigger analysis network based on training sample data to obtain the trigger analysis model; the trigger analysis model is used to output the trigger analysis results to determine whether to perform adaptive radiodelivery.

[0045] Fifthly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the adaptive radiodelivery determination method in the first aspect and / or the training method of the trigger analysis model in the second aspect.

[0046] In a sixth aspect, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the adaptive radiometric delivery determination method in the first aspect and / or the training method of the trigger analysis model in the second aspect.

[0047] In a seventh aspect, this application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the adaptive radio delivery determination method in the first aspect and / or the training method of the trigger analysis model in the second aspect.

[0048] The aforementioned adaptive radiodelivery determination method, trigger analysis model training method, apparatus, computer device, storage medium, and computer program product involve the computer device acquiring the current fraction image of the target object corresponding to the current radiodelivery fraction, and acquiring the initial image used by the target object when determining the initial radiodelivery plan. Then, based on the current fraction image, the initial image, and the trigger analysis model, a trigger analysis result is determined, and based on the trigger analysis result, it is determined whether to execute adaptive radiodelivery. In other words, the method proposed in this application, through a machine learning model, can achieve automatic trigger judgment for adaptive radiodelivery. Compared to manual trigger judgment analysis based on experience, this not only improves the efficiency of adaptive radiodelivery trigger judgment but also improves the consistency of trigger judgment, avoiding different trigger judgment results generated when different users make trigger judgments based on different clinical experiences, thereby improving the accuracy and rationality of adaptive trigger judgment. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1 This is a diagram illustrating the application environment of the adaptive radiometric delivery determination method in one embodiment.

[0051] Figure 2 This is a flowchart illustrating an adaptive radiation delivery determination method in one embodiment;

[0052] Figure 3 This is a flowchart illustrating the adaptive radiation delivery determination method in another embodiment;

[0053] Figure 4 This is a flowchart illustrating the adaptive radiation delivery determination method in another embodiment;

[0054] Figure 5 This is a flowchart illustrating the adaptive radiation delivery determination method in another embodiment;

[0055] Figure 6 This is a flowchart illustrating the adaptive radiation delivery determination method in another embodiment;

[0056] Figure 7 This is a flowchart illustrating a method for triggering the training of an analysis model in one embodiment;

[0057] Figure 8(a) is a schematic diagram of the first application of the first trigger analysis model in one embodiment;

[0058] Figure 8(b) is a schematic diagram of a second application of the first trigger analysis model in one embodiment;

[0059] Figure 9(a) is a schematic diagram of the application of the second trigger analysis model in one embodiment;

[0060] Figure 9(b) is a schematic diagram of the training of the second trigger analysis model in one embodiment;

[0061] Figure 10 This is a structural block diagram of an adaptive radiation delivery determination device in one embodiment;

[0062] Figure 11 This is a structural block diagram of a training device that triggers the analysis model in one embodiment;

[0063] Figure 12 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0064] 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.

[0065] With the development of radiotherapy technology, image-guided radiotherapy (IGRT) and adaptive radiotherapy (ART) have become two important methods of radiotherapy. Among them, adaptive radiotherapy planning (adaptive planning) is a method of readjusting the plan according to changes in tumor location, size, or disease progression during the patient's radiotherapy. Its main advantages are that it can effectively alleviate off-target problems caused by tumor regression and movement during treatment, reduce the radiation dose to normal tissues around the tumor, effectively reduce organ protection side effects, and improve the patient's treatment effect.

[0066] However, adaptive planning is a complex and time-consuming process, requiring simultaneous participation and cooperation from users, physicists, and technicians. This necessitates the rapid recalculation of patient data, plan optimization, and dose calculation, placing significant demands on clinical resources. Compared to traditional image-guided radiotherapy, it consumes far more medical resources. Furthermore, during treatment, factors such as minimal tumor regression in some fractions and minimal target movement can result in minimal benefits from adaptive radiotherapy, leading to unnecessary waste of medical resources. Therefore, maximizing patient benefit while rationally utilizing clinical resources has become a crucial issue.

[0067] Traditionally, adaptive planning triggering is determined by the user at the treatment site, based on clinical experience and the registered images, to decide whether an adaptive radiotherapy plan needs to be executed. However, this traditional method is inefficient, and the triggering results obtained by different users vary, leading to poor consistency in triggering decisions.

[0068] Based on this, this application proposes an adaptive radiotherapy delivery determination method. By introducing a machine learning algorithm and considering changes in the target area and organs at risk between radiotherapy sessions, it automatically determines whether online adaptive radiotherapy needs to be triggered and automatically obtains triggering attention parameters for adaptive triggering determination. This provides users with more specific reference information to determine whether adaptive plan adjustments are necessary, thereby improving the rationality and interpretability of triggering determination. This method can shorten the triggering determination time for online adaptive planning and solve the problem of poor consistency in triggering determination.

[0069] The adaptive radiometric delivery determination method provided in this application embodiment can be applied to, for example... Figure 1 In the application environment shown, computer device 102 can be a terminal, a server, or a radiographic delivery device, etc.; the terminal can be, but is not limited to, various personal computers, laptops, smartphones, tablets, and portable wearable devices, such as smartwatches, smart bracelets, and head-mounted devices. The server can be implemented using a standalone server or a server cluster composed of multiple servers.

[0070] In one exemplary embodiment, such as Figure 2 As shown, an adaptive radiation delivery determination method is provided, which is applied to... Figure 1 The following steps are used as an example of computer equipment, including steps 202 to 206. Wherein:

[0071] Step 202: Obtain the current fraction image of the target object corresponding to the current radiographic delivery fraction, and obtain the initial image used by the target object when determining the initial radiographic delivery plan.

[0072] The current fractional image is the fractional scan image of the target object obtained by scanning the target object with a scanning device before the current radiotherapy fractional treatment; for example, when performing the second treatment (the second radiotherapy fractional treatment), the scan image of the target object is acquired as the fractional scan image of the second radiotherapy fractional treatment.

[0073] For example, when scanning a target object, various imaging methods such as computed tomography (CT), magnetic resonance imaging (MRI), and electronic portal imaging device (EPID) can be used to scan the target object, thereby obtaining the initial image and / or the current fractional image of the target object.

[0074] Typically, when it is determined that a target object will be radioactively delivered, an initial radioactive delivery plan can be developed. This involves scanning the target object to obtain a scanned image, known as the initial image. Based on this initial image, an initial radioactive delivery plan can be formulated for the target object, including the number of radioactive deliveries and the radioactive delivery parameters for each delivery, such as machine parameters and dose parameters. Radioactive delivery can include, but is not limited to, radioactive delivery itself, radioactive delivery simulation (such as using phantoms for verification, simulated radioactive delivery, etc.), and radiation processing.

[0075] For example, for each radiodelivery fraction, before treatment in the current fraction, an image of the target object in the current fraction is acquired, along with an initial image of the target object. This allows for determination of whether adaptive radiodelivery should be triggered in the current fraction based on the differences between the current and initial images. If the differences between the current and initial images are small, adaptive radiodelivery is not required; image-guided radiodelivery can be performed based on a pre-defined radiodelivery plan. If the differences are large, adaptive radiodelivery needs to be triggered, and radiodelivery is performed on the target object based on the adjusted plan. It should be noted that adaptive radiodelivery triggering is not required for every radiodelivery. This not only saves resources but also improves efficiency, reduces user workload, and increases the number of patients treated or plan verifications that a single radiodelivery device can perform per day.

[0076] For example, when acquiring the initial image of the target object, the computer device can retrieve it from local storage. For instance, if the computer device has already acquired the initial image of the target object during the adaptive triggering judgment of the previous radiographic delivery, then the computer device can save the initial image of the target object for this time so that the computer device can directly and quickly retrieve the initial image of the target object from local storage next time, thereby improving data acquisition efficiency and thus improving the triggering efficiency of adaptive radiographic delivery. Of course, the computer device can also retrieve it from a preset database based on the identifier of the target object. The preset database may include, but is not limited to, a Picture Archiving and Communication System (PACS) server.

[0077] Step 204: Determine the trigger analysis result based on the current segmented image, the initial image, and the trigger analysis model.

[0078] Step 206: Determine whether to perform adaptive radiodelivery based on the trigger analysis results.

[0079] For example, the trigger analysis results may include trigger concern parameters and / or adaptive radiation delivery trigger judgment information, and the trigger concern parameters may include benefit information for one or more dose indicators.

[0080] For example, this trigger analysis model can output trigger concern parameters related to adaptive radiodelivery triggering, i.e., output benefit information for one or more dose indicators. This benefit information can be used to characterize the dosimetric benefit of adaptive radiodelivery adjustment in the current radiodelivery fraction compared to an image-guided radiodelivery plan. As an example, the benefit information may include the degree or amount of dose indicator benefit from performing adaptive radiotherapy in the current fraction. For instance, compared to conventional image-guided radiotherapy (IGRT), if adaptive radiotherapy is performed in the current fraction, the radiation dose received by the irradiated target area is greater (e.g., 30% greater or a specific difference), and / or the radiation dose received by normal tissues or organs surrounding the un-irradiated target area is less (e.g., less than 50% or a specific difference), etc.

[0081] Optionally, the trigger analysis model can also output adaptive radiodelivery triggering decision information, i.e., whether to trigger adaptive radiodelivery or not. For example, the adaptive radiodelivery triggering decision information output by the trigger analysis model can be determined based on triggering parameters of interest, such as using a triggering parameter model as input to the trigger analysis model or its sub-model, or as an intermediate parameter of the trigger analysis model or its sub-model.

[0082] For example, the triggering attention parameter can be an intermediate parameter obtained by the trigger analysis model during the process of outputting adaptive radiodelivery triggering judgment information. That is, after the trigger analysis model analyzes and processes the current fractionated image and the initial image, it can obtain benefit information of one or more dose indicators; then, it determines the adaptive radiodelivery triggering judgment information based on the benefit information of one or more dose indicators.

[0083] However, this is not the only limitation. The trigger analysis model can also directly output adaptive radiographic delivery trigger judgment information based on the current segment image and the initial image. For example, the trigger analysis model can also output only adaptive radiographic delivery trigger judgment information, which can be used to prompt the user that the target object is more suitable for adaptive radiographic delivery in the current radiographic delivery segment, and the effect is better than image-guided radiographic delivery. As an example, this trigger analysis model can be trained based on historical images and historical adaptive radiographic delivery trigger judgment results.

[0084] As an example, the trigger analysis model can be various machine learning models, such as neural network models, decision tree models, and so on.

[0085] For example, the trigger attention parameters and adaptive radiometric delivery trigger judgment information can also be obtained by the trigger analysis model after performing different analysis and processing on the current fractional image and the initial image; for example, the trigger analysis model can obtain the trigger judgment information by performing a first analysis and processing on the current fractional image and the initial image, or it can obtain the adaptive radiometric delivery trigger attention parameters by performing a second analysis and processing on the current fractional image and the initial image.

[0086] For example, the adaptive radiotherapy delivery triggering determination information may include whether the current radiotherapy delivery segment triggers adaptive radiotherapy delivery, or whether the next segment or multiple subsequent segments of the current radiotherapy delivery segment trigger adaptive radiotherapy delivery. For example, the current radiotherapy delivery segment may not trigger adaptive radiotherapy delivery, but which subsequent segment needs to trigger adaptive radiotherapy delivery, or, in the entire treatment process, none of the subsequent segments of the current segment need to trigger adaptive radiotherapy delivery, etc.

[0087] In the aforementioned adaptive radiodelivery determination method, the computer device acquires the current fraction image of the target object corresponding to the current radiodelivery fraction, and also acquires the initial image used by the target object when determining the initial radiodelivery plan. Then, based on the current fraction image, the initial image, and the trigger analysis model, the trigger analysis result is determined, and based on the trigger analysis result, it is determined whether to execute adaptive radiodelivery. In other words, the method proposed in this application, through a machine learning model, can achieve automatic trigger judgment for adaptive radiodelivery. Compared to manual trigger judgment analysis based on experience, this not only improves the efficiency of adaptive radiodelivery trigger judgment but also improves the consistency of trigger judgment, avoiding different trigger judgment results generated when different users make trigger judgments based on different clinical experiences. This improves the accuracy and rationality of adaptive trigger judgment.

[0088] Furthermore, when the trigger analysis results include trigger concern parameters and / or adaptive radiodelivery trigger judgment information, and the trigger concern parameters include benefit information for one or more dose indices, this method can not only provide users with more reasonable and valuable trigger concern parameters, but also provide users with adaptive radiodelivery trigger judgment information, i.e., whether adaptive radiodelivery is triggered. In other words, the trigger analysis results obtained by this method can provide users with a reference basis for trigger judgment and result verification, and can also provide specific reference information for the adjustment of adaptive radiodelivery in the current or subsequent fractions, thus comprehensively improving the accuracy of adaptive trigger judgment, as well as the rationality and interpretability of adaptive triggering.

[0089] In an exemplary embodiment, adaptive radiometric delivery triggering determination information can be obtained through the above trigger analysis model, such as... Figure 3 As shown, a method for obtaining adaptive radiographic delivery trigger judgment information is provided. For example, the trigger analysis model may include a first trigger analysis model, which is used to output adaptive radiographic delivery trigger judgment information; based on this, step 204 may include steps 302 to 306. Wherein:

[0090] Step 302: Perform image registration between the current segmented image and the initial image to obtain the registered image.

[0091] For example, when performing image registration, the current segment image and the initial image can be rigidly registered. Since the target object may have different positions and postures in different radiographic delivery segments, in order to ensure that the position and structure of the tissue in the two images are consistent, the current segment image and the initial image can be rigidly registered, including but not limited to registration operations such as rotation and translation. That is, the current segment image can be rotated and translated to make the adjusted registered image basically consistent with the initial image.

[0092] Step 304: Delineate the region of interest in the registered image to obtain a delineated image, which includes the delineation result of the region of interest.

[0093] The registered image is the image obtained after processing the current scanned image by rotating, translating, or performing other operations.

[0094] For example, for a registered image, the region of interest (ROI) in the registered image can be delineated to obtain a delineated image containing the delineated ROI. The ROI may include the target area and / or organs at risk surrounding the target area. The target area is the lesion area requiring radiotherapy, i.e., the area where radiation is concentrated. The organs at risk surrounding the target area are healthy organs that may be irradiated during radiotherapy. During radiotherapy, radiation exposure to these organs should be minimized, i.e., healthy organs other than the lesion, to reduce adverse effects on healthy organs during radiotherapy. Optionally, a certain area surrounding the target area can be designated as the ROI, which may include, but is not limited to, specific sites, organs, or tissues of interest to which the user is of particular interest.

[0095] For example, when delineating the region of interest, manual delineation by the user or automatic delineation can be supported. When delineating automatically, an automatic segmentation model can be used to segment the region of interest in the registered image, and then the region of interest in the registered image can be delineated based on the segmentation result to obtain a delineated image containing the delineation result of the region of interest.

[0096] For example, deformation registration can also be used to delineate the region of interest (ROI) in the registered image based on the original delineation result in the original scanned image, thereby obtaining a delineated image containing the delineated ROI. When formulating the initial radiographic delivery plan, the user delineates the ROI in the initial image; that is, the initial image can include the original delineation result corresponding to the ROI. For instance, a computer device can use a deformation registration algorithm to perform deformation registration on the initial image and the registered image, thereby obtaining deformation information, such as a deformation field, between the two images. Then, based on this deformation information, the original delineation result in the initial image is registered to the registered image, thereby obtaining a delineated image containing the delineated ROI.

[0097] It should be noted that the method of outlining the region of interest is not limited to the outlining methods described above; for example, it can be a combination of one or more of the above outlining methods, or other outlining methods or other outlining methods combined with the above outlining methods, etc.

[0098] Step 306: Input the outlined image into the first trigger analysis model and output the trigger analysis result.

[0099] The trigger analysis result may include adaptive radiographic delivery trigger judgment information; optionally, the first trigger analysis model may determine the adaptive radiographic delivery trigger judgment information based on intermediate parameters obtained during the analysis process, such as trigger attention parameters; or it may determine the adaptive radiographic delivery trigger judgment information based on other intermediate parameters, or it may obtain the adaptive radiographic delivery trigger judgment information based on the feature information of the delineated image obtained by the first trigger analysis model during the processing.

[0100] For example, the registration image obtained in step 302 above can be the adjusted current segmented image or the fused image of the adjusted current segmented image and the initial image; correspondingly, the delineation image obtained in step 304 above can be the delineation image including the current delineation result of the region of interest after delineating the region of interest on the adjusted current segmented image, or the delineation image including the current delineation result of the region of interest and the original delineation result after delineating the region of interest on the fused image.

[0101] For example, when the delineated image is a delineated image including the current delineation result, the original delineation result, the current segmented image, and the initial image, the computer device can input the delineated image into the first trigger analysis model for analysis and processing. Based on the feature differences between the current delineation result and the original delineation result, and / or the feature differences between the current segmented image and the initial image, adaptive radiometric delivery trigger judgment information is obtained. After the analysis and processing of the first trigger analysis model, a trigger analysis result of whether adaptive radiometric delivery is triggered or not can be output.

[0102] For example, when the outlined image is an outlined image that includes the current outlined result and the current segmented image, the computer device can input the outlined image and the initial image including the original outlined result into the first trigger analysis model for analysis and processing, thereby outputting a trigger analysis result of triggering or not triggering based on the feature differences between the current outlined result and the original outlined result, and / or the feature differences between the current segmented image and the initial image.

[0103] For example, the first trigger analysis model can be any type of machine learning model, including but not limited to random forest, decision tree, support vector machine, regression, deep learning and other models; the output of the first trigger analysis model can include two methods: regression and classification. When the regression method is used, the probability value of whether adaptive radio delivery is triggered can be output. By comparing the output probability value with the preset probability threshold, the trigger analysis result in the form of classification can be obtained, that is, whether adaptive radio delivery is triggered or not.

[0104] For example, when the first trigger analysis model is a deep learning model, the outlined image, or the outlined image and the initial image including the original outlined result, can be input into the deep learning model for feature learning and trigger judgment, and finally output adaptive radial delivery trigger judgment information. The deep learning method can automatically find and learn features and complete the trigger judgment at the same time, thereby improving the processing efficiency of the first trigger analysis model.

[0105] In this embodiment, the trigger analysis model may include a first trigger analysis model, which can be used to output a trigger analysis result indicating whether adaptive radiodelivery is triggered, i.e., adaptive radiodelivery trigger judgment information. When using the first trigger analysis model for trigger judgment, the computer device can first perform image registration between the current fractionated image and the initial image to obtain a registered image; then, the region of interest in the registered image is delineated to obtain a delineated image containing the delineated region of interest; finally, the delineated image is input into the first trigger analysis model, and the trigger analysis result indicating whether adaptive radiodelivery is triggered is output. That is, in this example, the use of machine learning algorithms to automatically determine whether online adaptive radiodelivery is triggered and obtain adaptive radiodelivery trigger judgment information can save users from the process of repeated confirmation at the treatment site, thereby shortening the trigger time of the adaptive plan, improving trigger efficiency, and improving the consistency of the trigger judgment results.

[0106] In one exemplary embodiment, the first trigger analysis model described above can be a deep learning model that automatically performs feature learning and trigger judgment. In other implementations, the first trigger analysis model can also be a machine learning model other than deep learning. In this case, a separate feature extraction process needs to be performed, and the user needs to set the required features. For example, the first trigger analysis model can include a first feature extraction network and a trigger analysis network. Based on this, such as... Figure 4 As shown, step 306 above may include steps 402 to 404. Wherein:

[0107] Step 402: Input the outlined image and the initial image into the first feature extraction network for feature extraction to obtain target feature information.

[0108] The target feature information may include, but is not limited to, at least one of the following: morphological difference feature information based on the region of interest, image difference feature information, and registration difference feature information; the initial image includes the original delineation result of the region of interest.

[0109] In this example, the delineated image can be the image obtained by delineating the region of interest on the adjusted current fractional image. In this case, the delineated image and the initial image can be input into the first feature extraction network for feature extraction to obtain the target feature information.

[0110] For example, the computer device can input the outlined image into a first feature extraction network to extract the morphological features of the region of interest, thereby obtaining the morphological feature information of the current outlined result; and input the initial image into the first feature extraction network to extract the morphological features of the region of interest, thereby obtaining the morphological feature information of the original outlined result; then, the computer device can compare the morphological feature information of the current outlined result with the morphological feature information of the original outlined result to obtain the morphological difference feature information of the region of interest; optionally, the morphological feature information may include, but is not limited to, information such as the position, shape, and size of the region of interest, and correspondingly, the morphological difference feature information may include feature information such as the positional difference, shape difference, and size difference between the current outlined result and the original outlined result.

[0111] For example, the computer device can also input the outlined image and the initial image into a first feature extraction network to perform image difference analysis and obtain image difference feature information; wherein, the image difference analysis may include, but is not limited to, overall image difference analysis, difference analysis of regions of interest in the image, pixel difference analysis, contrast difference analysis, etc.

[0112] For example, the computer device can also input the outlined image and the initial image into a first feature extraction network for registration difference analysis to obtain registration difference feature information; wherein, the registration difference feature information may include deformation field feature information based on registration, etc.

[0113] For example, when morphological difference feature information, image difference feature information, and registration difference feature information based on the region of interest are obtained, the morphological difference feature information, image difference feature information, and registration difference feature information can be merged to obtain the aforementioned target feature information.

[0114] It should be noted that the feature extraction process and the type of feature information can be defined by the user, and the user can also manually extract features so that the extracted feature information can be input into the trigger analysis network for analysis.

[0115] Step 404: Input the target feature information into the trigger analysis network and output the trigger analysis result.

[0116] Optionally, the trigger analysis network can be a machine learning network based on one or more modeling methods, such as random forest, decision tree, support vector machine, regression, etc. After obtaining the target feature information through the first feature extraction network, the target feature information can be input into the trigger analysis network for processing to obtain the trigger analysis result of whether adaptive radio delivery is triggered, i.e., adaptive radio delivery trigger judgment information.

[0117] In this embodiment, the first trigger analysis model may include a first feature extraction network and a machine learning-based trigger analysis network. The computer device extracts features from the delineated image and the initial image by inputting them into the first feature extraction network to obtain target feature information. Then, the target feature information is input into the trigger analysis network to output the trigger analysis result. The initial image includes the original delineation result of the region of interest. Using the method in this example, automatic trigger determination for adaptive radiometric delivery is achieved through machine learning algorithms, obtaining a direct result of whether adaptive radiometric delivery is triggered, improving the efficiency of trigger determination and the consistency of trigger determination results.

[0118] In one exemplary embodiment, the methods described in the above embodiments can be used to achieve adaptive radiodelivery triggering. This adaptive radiodelivery triggering is a technique that determines whether or when adaptive radiodelivery adjustments are needed during treatment based on patient images, dose, and other information. It can assist the user's judgment and thus improve machine treatment efficiency. However, the aforementioned adaptive radiodelivery triggering technology, in assisting the judgment process, can only provide the user with the direct result of whether adaptive radiodelivery has been triggered, which has limited clinical application value and rationality. Compared to providing information on whether adaptive radiodelivery has been triggered, providing information on how much benefit adaptive radiodelivery adjustments bring to the dose indicators of interest compared to image-guided radiodelivery would be more rational and interpretable in clinical application. Therefore, using only models that predict whether adaptive radiodelivery needs to be triggered is insufficient to meet clinical needs, leaving users lacking specific reference information when determining whether adaptive radiodelivery adjustments are necessary.

[0119] Based on this, this application also proposes another trigger analysis model that can predict the benefits of adaptive radiodelivery. This model can predict different dose indicators of clinical concern, thereby assisting clinical assessment in determining whether adaptive radiodelivery is necessary for the target population. By predicting the benefits at different dose indicators, users can more comprehensively assess whether adaptive radiodelivery adjustments are needed, thus better balancing the relationship between clinical resources and patient benefits.

[0120] For example, the above-mentioned trigger analysis model may further include a second trigger analysis model, which is used to output trigger attention information including benefit information of one or more dose indicators; based on this, such as Figure 5 As shown, step 204 above may further include steps 502 to 506. Wherein:

[0121] Step 502: Perform the first delineation operation on the region of interest in the current segmented image based on the original delineation result in the initial image to obtain the first delineated image.

[0122] The initial image includes the original delineation result of the region of interest. For example, when delineating the region of interest in the current segmented image based on the original delineation result, the original delineation result in the initial image can be rigidly copied to the current segmented image to obtain a first delineated image containing the original delineation result. Optionally, this first delineated image can be the current segmented image plus the original delineation result, that is, adding the original delineated outline or original region of the region of interest to the current segmented image; or it can be a single image obtained by registering and fusing the current segmented image and the original delineation result.

[0123] For example, before delineation, rigid registration can be performed on the initial image and the current segmented image. Then, based on the result of rigid registration, the original delineation result in the initial image is rigidly copied to the current segmented image, resulting in a first delineated image containing the original delineation result. This process copies the original delineation result of the region of interest in the initial image to the corresponding position of the region of interest in the current segmented image, avoiding significant positional offsets between the original delineation result and the region of interest, thereby improving the accuracy of the first delineated image.

[0124] It should be noted that the position and shape of the region of interest (ROI) in the current segmented image may differ from those in the initial image. However, after rigid copying, it is essential to ensure that the original delineation result corresponds to the ROI in the current segmented image, and not to other regions outside the ROI. For example, due to differences in the placement and posture of the target object, the ROI in the initial image may be located at the center of the entire image, while the ROI in the current segmented image may be located at the lower left of the entire image. Therefore, when rigidly copying the original delineation result to the current segmented image, it is desirable that the original delineation result is located at the position of the ROI at the lower left, rather than at the center of the current segmented image.

[0125] Step 504: Perform a second delineation operation on the region of interest in the current segmented image to obtain the second delineated image.

[0126] For example, a computer device can use a preset segmentation algorithm to automatically segment the region of interest (ROI) in the current segmented image, and then perform a second delineation operation on the ROI in the current segmented image based on the segmentation result, to obtain a second delineated image including the current delineation result of the ROI. It should be noted that the preset segmentation algorithm can be any form of segmentation network or model, and this application embodiment does not specifically limit it.

[0127] Step 506: Input the first outlined image and the second outlined image into the second trigger analysis model, and output the trigger analysis result.

[0128] Wherein, the first delineation image is the original delineation result of the current fractional image including the region of interest, and the second delineation image is the current delineation result of the current fractional image including the region of interest; for example, the second trigger analysis model can determine the benefit information of multiple dose indicators by performing difference analysis on the original delineation result and the current delineation result of the region of interest.

[0129] For example, a computer device can input a first outlined image and a second outlined image into a second trigger analysis model for difference analysis to obtain trigger analysis results including trigger attention parameters.

[0130] In one implementation, the second trigger analysis model may include a second feature extraction network and a decision network. The computer device may input the first outlined image into the second feature extraction network for feature extraction to obtain first feature information; and input the second outlined image into the second feature extraction network for feature extraction to obtain second feature information; then, input the first feature information and the second feature information into the decision network to output the trigger analysis result.

[0131] For example, given the first feature information and the second feature information, the first feature information and the second feature information can be concatenated, and the concatenated feature information can be input into the decision network to output the triggered analysis result. It should be noted that the network structure of the second feature extraction network and the decision network in this embodiment is not specifically limited. The first feature information and the second feature information can be a first feature vector and a second feature vector, or a first feature map and a second feature map. The representation of the feature information is also not specifically limited in this embodiment.

[0132] For example, for the benefit information of one or more dose indices in the trigger analysis results, the greater the benefit information, the higher the dosimetric benefit of adaptive planning adjustment for the target object in the radiation delivery fraction; optionally, the benefit information can be the degree of benefit or the benefit value, etc.

[0133] In this embodiment, the trigger analysis model includes a second trigger analysis model, which can be used to output trigger attention parameters, i.e., benefit information of one or more dose indicators. When using the second trigger analysis model for trigger judgment, the computer device can perform a first delineation operation on the region of interest in the current fractional image based on the original delineation results in the initial image, obtaining a first delineated image; and then perform a second delineation operation on the region of interest in the current fractional image, obtaining a second delineated image; then, the first and second delineated images are input into the second trigger analysis model, and the trigger analysis result including the trigger attention parameters is output. That is, in this example, the dose indicator benefit of adaptive radiodelivery adjustment compared to image-guided radiodelivery can be predicted based on changes in the patient's anatomy, serving as an auxiliary decision-making tool to help users determine whether adaptive radiodelivery adjustment is necessary; this method can provide more valuable and referential trigger attention information for clinicians, improving the rationality and interpretability of adaptive radiodelivery trigger judgment.

[0134] In an exemplary embodiment, based on the above embodiments, when trigger concern information including benefit information of one or more dose indicators is obtained, it is further possible to determine whether adaptive radiodelivery is triggered based on the trigger concern information, and obtain a trigger analysis result of whether adaptive radiodelivery is triggered; for example, when the trigger concern information includes benefit information of a dose indicator, the trigger analysis result of adaptive radiodelivery can be obtained if the benefit information of the dose indicator is greater than or equal to the preset benefit information threshold, according to the relationship between the benefit information of the dose indicator and the preset benefit information threshold.

[0135] For example, when the triggering information of concern includes benefit information from multiple dose indicators, a comprehensive judgment can be made on whether to trigger adaptive radiation delivery based on the benefit information from multiple dose indicators; based on this, such as Figure 6 As shown, the above method may further include steps 602 to 604. Wherein:

[0136] Step 602: Perform weighted summation on the benefit information of each dose index in the triggering attention parameters to obtain comprehensive benefit information.

[0137] For example, the weights of each dose indicator can be the same or different; the weight of each dose indicator can be flexibly set according to the importance of the actual dose indicator, so as to obtain the comprehensive benefit information by weighted summation based on the benefit information of each dose indicator and the weight of each dose indicator.

[0138] Step 604: Based on the comprehensive benefit information and the preset benefit information threshold, determine whether to trigger the trigger analysis result.

[0139] For example, if the overall benefit information is greater than or equal to a preset benefit information threshold, it can be determined that adaptive radio delivery is triggered; if the overall benefit information is less than the preset benefit information threshold, it can be determined that adaptive radio delivery is not triggered.

[0140] In this embodiment, after obtaining benefit information for multiple dose indicators through the second trigger analysis model, the benefit information for each dose indicator in the trigger attention parameters can be further weighted and summed to obtain comprehensive benefit information. Then, based on the comprehensive benefit information and a preset benefit information threshold, the trigger analysis result for whether adaptive radiodelivery is triggered is determined. In other words, the method of this embodiment not only obtains the trigger attention parameters for adaptive radiodelivery, providing reference information for triggering adaptive radiodelivery, but also automatically determines whether the trigger analysis result for adaptive radiodelivery is triggered based on these trigger attention parameters, i.e., adaptive radiodelivery trigger judgment information, providing direct trigger analysis results to the user. This method improves the efficiency of trigger judgment, as well as the rationality and interpretability of the trigger judgment.

[0141] In one exemplary embodiment, such as Figure 7 As shown, a method for training a trigger analysis model is also provided, which can be used to determine adaptive radiometric delivery. This method is applied to... Figure 1 The following steps are used as an example of computer equipment, including steps 702 to 704. Wherein:

[0142] Step 702: Obtain training sample data; the training sample data includes the initial sample images used by different sample objects when determining the initial radiographic delivery plan, the sample images of the sample objects at different radiographic delivery fractions, and the corresponding sample trigger analysis results.

[0143] The sample trigger analysis results may include sample triggering concern parameters and / or sample adaptive radiation delivery triggering judgment information. For example, when the sample trigger analysis results include sample triggering concern parameters, these parameters may include benefit information labels for one or more dose parameters.

[0144] For example, the process of obtaining the benefit information label corresponding to each fractional sample image may include: for each fractional sample image, obtaining the index difference of each dose index obtained after performing image-guided radiodelivery planning and adaptive planning for the corresponding radiodelivery fraction of the fractional sample image; normalizing the index difference of each dose index to obtain the benefit information label of each dose index corresponding to the fractional sample image.

[0145] For example, taking a single dose index as an example, when normalizing the index difference of a dose index, the index difference of the dose index in each fractional sample image can be normalized based on the index differences of the dose index in multiple fractional sample images. For instance, the maximum index difference of the dose index can be determined from the index differences of the dose index in multiple fractional sample images, and the quotient of the index difference of the dose index in each fractional sample image and the maximum index difference can be used as the benefit information label of the dose index in each fractional sample image. That is, the index difference of the dose index in each fractional sample image can be normalized to between 0 and 1, thereby obtaining the benefit information label of the dose index. Other dose indices can also use the same normalization method, which will not be repeated here.

[0146] Step 704: Train the initial trigger analysis network based on the training sample data to obtain the trigger analysis model; the trigger analysis model is used to output the trigger analysis results for determining whether to perform adaptive radiodelivery.

[0147] For example, the initial trigger analysis network can be any type of deep learning network or any type of machine learning network. This application embodiment does not specifically limit the network type of the trigger analysis network.

[0148] For example, the initial sample image and the corresponding sub-sample images from the training sample data can be input into the initial trigger analysis network to obtain intermediate trigger analysis results. Then, based on the difference between the intermediate trigger analysis results and the corresponding sample trigger analysis results, the loss function of the initial trigger analysis network is determined. The network parameters of the initial trigger analysis network are then adjusted according to the value of the loss function, and multiple iterations of training are performed until a preset training cutoff condition is met. The resulting trigger analysis network that meets the preset training cutoff condition is then considered a trained trigger analysis model. Optionally, the preset training cutoff condition may include, but is not limited to, a preset number of iterations, a loss function value less than a preset threshold, etc.

[0149] By employing the training method of the aforementioned trigger analysis model, a trigger analysis model applicable to adaptive radiodelivery determination can be pre-trained. Based on this trigger analysis model, a comparative analysis of adaptive radiodelivery and image-guided radiodelivery can be performed before radiodelivery is executed in multiple stages, thereby obtaining a treatment plan with higher benefits. This not only improves the trigger judgment efficiency of adaptive radiodelivery but also enhances the effectiveness and rate of radiodelivery while reducing the number of radiodelivery operations.

[0150] In one exemplary embodiment, a trigger determination method is provided for determining whether adaptive radiodelivery is triggered based on a first trigger analysis model, wherein the first trigger analysis model is used to obtain the trigger analysis result of whether adaptive radiodelivery is triggered.

[0151] Referring to Figure 8(a), a schematic diagram of a first application of the first trigger analysis model is shown. In this example, the first trigger analysis model includes a feature extraction network (i.e., the first feature extraction network mentioned above) and a machine learning model (i.e., the trigger analysis network mentioned above). Before performing the current radiotherapy fraction, the IGRT image of the target object is scanned as the current fraction image. The IGRT image of the target object is rigidly registered with the initial image of the target object to obtain a registered image. The target area and organs at risk are delineated on the registered image to obtain a delineated image containing the current delineation results. Then, the feature extraction network is used to extract features from the delineated image and the initial image containing the original delineation results to obtain target feature information. The target feature information may include at least one of morphological difference feature information based on the region of interest, image difference feature information, and registration difference feature information. Finally, the target feature information is input into the machine learning model, and the output is the trigger analysis result of whether adaptive radiotherapy is triggered.

[0152] For example, the methods for automatically delineating the target area and organs at risk described above may include, but are not limited to, automatic segmentation models and deformation registration, as described in step 304 above, and will not be repeated here. Furthermore, feature extraction can include various forms of features, including but not limited to morphological features based on the Region of Interest (ROI), image features based on the scan, and deformation fields based on registration, as described in step 402 above, and will not be repeated here. The machine learning model can include various modeling methods, such as random forests, decision trees, support vector machines, regression, and deep learning. The output of this machine learning model can include both regression and classification methods. When using regression, a threshold can be designed to convert it into a classification form, ultimately obtaining the trigger analysis result for whether adaptive radiotherapy delivery is triggered.

[0153] Referring to Figure 8(b), a second application diagram of the first trigger analysis model is shown. In this example, the first trigger analysis model can be a deep learning model. Referring to the first application described above, before performing the current radiotherapy fraction, the IGRT image of the target object is scanned as the current fraction image; the IGRT image of the target object is rigidly registered with the initial image of the target object to obtain a registered image, and the target area and organs at risk are delineated on the registered image to obtain a delineated image containing the current delineation results; then, the delineated image and the initial image containing the original delineation results are input into the deep learning model for feature learning and trigger determination, and the trigger analysis result of whether adaptive radiotherapy is triggered is output.

[0154] In the above method, machine learning algorithms are used to automatically determine whether adaptive radiotherapy is triggered, eliminating the need for users to repeatedly confirm on-site based on registered images and clinical experience. This shortens the triggering time for adaptive radiotherapy, improves the triggering efficiency of adaptive radiotherapy, and increases the consistency of the triggering results.

[0155] In an exemplary embodiment, a triggering judgment method is provided for determining whether to trigger an adaptive plan based on a second triggering analysis model. Referring to Figure 9(a), the second triggering analysis model may include a feature extraction network (i.e., the second feature extraction network described above) and a decision network. The feature extraction network maps the input image information of the second triggering analysis model to a new space, and the decision network makes a corresponding judgment on each output dose index by performing a similarity measurement on the obtained feature vectors, thereby obtaining the benefit information of each dose index.

[0156] For example, the feature extraction network may include feature extraction network 1 and feature extraction network 2. Feature extraction network 1 and feature extraction network 2 may have the same network structure and share parameters, so that feature extraction networks 1 and 2 can obtain more consistent feature map pairs. Each feature extraction network may consist of 4 convolutional blocks, and each convolutional block may be formed by a combination of batch normalization layer, convolutional layer, activation function and pooling layer.

[0157] Before performing the current radiotherapy fraction, an IGRT image of the target object is scanned and used as the current fraction image. The original delineation result of the region of interest (ROI) from the original scan image is rigidly copied to the current fraction image to obtain a first delineated image, which is used as input image 1. The ROI in the current fraction image is automatically delineated to obtain a second delineated image, which is used as input image 2. The first delineated image is input to feature extraction network 1 for feature extraction to obtain a first feature map. The second delineated image is input to feature extraction network 2 for feature extraction to obtain a second feature map. Then, the first and second feature maps are concatenated and fed into a decision network for prediction. The output contains multiple nodes, each node representing the benefit information of the dose index of interest. The greater the benefit information of the dose index, the higher the dosimetric benefit of adaptive planning adjustment for the target object in this radiotherapy fraction.

[0158] Optionally, the decision network can consist of two fully connected modules, each of which includes four parts: batch normalization, activation function, random deactivation, and fully connected layers.

[0159] Referring to Figure 9(b), the training process of the second trigger analysis model is illustrated. The second trigger analysis model requires two phases before use: training and testing. During the training phase, labels are generated based on the obtained historical data. Optionally, the radiation delivery plan and adaptive radiation delivery plan can be guided based on the images of each radiation delivery segment to obtain the dose differences of clinically relevant dose indicators between the two plans, thus obtaining the set of differences for each dose indicator corresponding to the sample images of each radiation delivery segment. For each dose indicator, by normalizing the set of differences corresponding to the dose indicator, the benefit information label (e.g., a value between 0 and 1) corresponding to that dose indicator can be obtained for each sample scan data. Then, based on the generated benefit information labels, the learning network can be guided to continuously update weights, extract, and summarize features with the help of a loss function. Depending on the level of model training, the parameters of the learning network can be selectively frozen for inference in the testing phase.

[0160] During the testing phase, by inputting the acquired scan images of a certain radiodelivery segment into a trained second trigger analysis model for processing, the benefits of adaptive plans for various dose indicators of clinical concern compared to image-guided radiodelivery plans can be predicted in real time.

[0161] The above method employs a deep learning-based adaptive radiotherapy delivery trigger analysis model. This model can predict the dose-related benefits of adaptively adjusting the current radiotherapy delivery sequence compared to image-guided radiotherapy based on changes in the patient's anatomy. This serves as an auxiliary decision-making tool to help users determine whether adaptive radiotherapy adjustments are necessary. Furthermore, this method can automatically generate labels to assist clinicians in providing training labels, thus reducing clinical workload.

[0162] 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.

[0163] Based on the same inventive concept, this application also provides an adaptive radiation delivery determination apparatus for implementing the adaptive radiation delivery determination 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 embodiments of the adaptive radiation delivery determination apparatus provided below can be found in the limitations of the adaptive radiation delivery determination method described above, and will not be repeated here.

[0164] In one exemplary embodiment, such as Figure 10 As shown, an adaptive radiation delivery determination device is provided, comprising: an acquisition module 1002, a first determination module 1004, and a second determination module 1006, wherein:

[0165] The acquisition module 1002 is used to acquire the current fraction image of the target object corresponding to the current radiographic delivery fraction, and to acquire the initial image used by the target object when determining the initial radiographic delivery plan.

[0166] The first determining module 1004 is used to determine the trigger analysis result based on the current segmented image, the initial image, and the trigger analysis model.

[0167] The second determining module 1006 is used to determine whether to perform adaptive radio delivery based on the trigger analysis results.

[0168] In one embodiment, the trigger analysis results include trigger concern parameters and / or adaptive radiation delivery trigger determination information, wherein the trigger concern parameters include benefit information of one or more dose indices.

[0169] In one embodiment, the trigger analysis model includes a first trigger analysis model; the first determination module 1004 includes:

[0170] The registration unit is used to register the current segmented image and the initial image to obtain a registered image.

[0171] The first delineation unit is used to delineate the region of interest in the registered image to obtain a delineated image, which includes the delineation result of the region of interest.

[0172] The first determining unit is used to input the outlined image into the first trigger analysis model and output the trigger analysis result.

[0173] In one embodiment, the first trigger analysis model includes a first feature extraction network and a trigger analysis network, and a first determination unit, specifically used to input the outlined image and the initial image into the first feature extraction network for feature extraction to obtain target feature information; input the target feature information into the trigger analysis network and output the trigger analysis result, wherein the initial image includes the original outlined result of the region of interest.

[0174] In one embodiment, the target feature information includes at least one of morphological difference feature information based on the region of interest, image difference feature information, and registration difference feature information.

[0175] In one embodiment, the trigger analysis model includes a second trigger analysis model; the first determining module 1004 includes:

[0176] The second delineation unit is used to perform a first delineation operation on the region of interest in the current segmented image based on the original delineation result in the initial image, so as to obtain the first delineated image.

[0177] The third delineation unit is used to perform a second delineation operation on the region of interest in the current segmented image to obtain a second delineated image.

[0178] The second determining unit is used to input the first outlined image and the second outlined image into the second trigger analysis model and output the trigger analysis result.

[0179] In one embodiment, the second outlining unit is specifically used to rigidly copy the original outlining result in the initial image to the current segmented image to obtain a first outlining image containing the original outlining result.

[0180] In one embodiment, the second trigger analysis model includes a second feature extraction network and a decision network, and a second determination unit, specifically used to input the first outline image into the second feature extraction network for feature extraction to obtain first feature information; input the second outline image into the second feature extraction network for feature extraction to obtain second feature information; input the first feature information and the second feature information into the decision network, and output the trigger analysis result.

[0181] In one embodiment, the trigger analysis results include trigger interest parameters, and the device further includes:

[0182] The processing module is used to perform weighted summation of the benefit information of each dose index in the triggering attention parameters to obtain comprehensive benefit information;

[0183] The third determining module is used to determine the adaptive radioactive delivery trigger judgment information based on the comprehensive benefit information and the preset benefit information threshold.

[0184] The modules in the aforementioned adaptive radiation delivery determination 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.

[0185] Based on the same inventive concept, this application also provides a training apparatus for a trigger analysis model to implement the training method for the trigger analysis model described above. The solution provided by this apparatus is similar to the implementation described in the above method. Therefore, the specific limitations of one or more training apparatus embodiments for trigger analysis models provided below can be found in the limitations of the training method for trigger analysis models described above, and will not be repeated here.

[0186] In one exemplary embodiment, such as Figure 11 As shown, a training device for triggering an analysis model is provided, comprising: a first acquisition module 1102 and a training module 1104, wherein:

[0187] The first acquisition module 1102 is used to acquire training sample data; the training sample data includes the initial sample images used by different sample objects when determining the initial radiographic delivery plan, the sample images of the sample objects at different radiographic delivery stages, and the corresponding sample trigger analysis results.

[0188] The training module 1104 is used to train the initial trigger analysis network based on training sample data to obtain the trigger analysis model; the trigger analysis model is used to output the trigger attention parameters and / or the trigger analysis results based on the adaptive radiometric delivery trigger judgment information determined by the trigger attention parameters.

[0189] In one embodiment, the sample-triggered analysis results include sample-triggered attention parameters, which include benefit information labels for one or more dose indicators; the device further includes:

[0190] The second acquisition module is used to acquire the difference in dose indicators obtained after image-guided radiodelivery planning and adaptive planning are performed on the corresponding radiodelivery segments of each fractional sample image.

[0191] The processing module is used to normalize the differences between each dose index to obtain benefit information labels for one or more dose indices corresponding to the sample images.

[0192] Each module in the training device for the aforementioned trigger analysis model 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.

[0193] In one exemplary embodiment, a computer device is provided, which may be a terminal, a server, or a radiographic delivery device; taking the computer device as a terminal as an example, its internal structure diagram may be as follows. Figure 12As shown, the computer device includes a processor, memory, input / output interfaces, a communication interface, a display unit, and an input device. The processor, memory, and input / output interfaces are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interfaces. The processor provides computational 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 input / output interfaces are used for exchanging information between the processor and external devices. 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 adaptive radiometric delivery determination method and / or triggers the training method of an analysis model. The display unit is used to form a visually visible image and can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the casing of the computer device, or external keyboards, touchpads, or mice, etc.

[0194] Those skilled in the art will understand that Figure 12 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.

[0195] In one exemplary embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of the adaptive radiodelivery determination method and / or the training method for triggering the analysis model in any of the above embodiments.

[0196] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps of the adaptive radiodelivery determination method and / or the training method for triggering the analysis model in any of the above embodiments.

[0197] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps of the adaptive radiometric delivery determination method and / or the method for triggering the training of the analysis model in any of the above embodiments.

[0198] It should be noted that the data involved in this application (including but not limited to data used for analysis, data stored, data displayed, etc.) are all information and data that have been fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0199] Those skilled in the art will understand that all or part of the processes in 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. When executed, the computer program can include the processes of the embodiments described above. 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.

[0200] 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.

[0201] 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 adaptive radiometric delivery determination method, characterized in that, include: Acquire the current fraction image of the target object corresponding to the current radiographic delivery fraction, and acquire the initial image of the target object used when determining the initial radiographic delivery plan; The trigger analysis result is determined based on the current segmented image, the initial image, and the trigger analysis model; Based on the trigger analysis results, determine whether to execute adaptive radio delivery.

2. The adaptive radiation delivery determination method according to claim 1, characterized in that, The trigger analysis results include trigger concern parameters and / or adaptive radiation delivery trigger judgment information, wherein the trigger concern parameters include benefit information for one or more dose indices.

3. The adaptive radiation delivery determination method according to claim 1, characterized in that, The trigger analysis model includes a first trigger analysis model; determining the trigger analysis result based on the current segmented image, the initial image, and the trigger analysis model includes: The current segmented image is registered with the initial image to obtain a registered image; The region of interest in the registered image is delineated to obtain a delineated image, which includes the delineation result of the region of interest; The outlined image is input into the first trigger analysis model, and the trigger analysis result is output.

4. The adaptive radiation delivery determination method according to claim 3, characterized in that, The first trigger analysis model includes a first feature extraction network and a trigger analysis network. The step of inputting the outlined image into the first trigger analysis model and outputting the trigger analysis result includes: The outlined image and the initial image are input into the first feature extraction network for feature extraction to obtain target feature information; The target feature information is input into the trigger analysis network, and the trigger analysis result is output. The initial image includes the original delineation result of the region of interest.

5. The adaptive radiation delivery determination method according to claim 4, characterized in that, The target feature information includes at least one of morphological difference feature information based on the region of interest, image difference feature information, and registration difference feature information.

6. The adaptive radiation delivery determination method according to claim 1, characterized in that, The trigger analysis model includes a second trigger analysis model; determining the trigger analysis result based on the current segmented image, the initial image, and the trigger analysis model includes: Based on the original delineation result in the initial image, a first delineation operation is performed on the region of interest in the current segmented image to obtain a first delineated image; A second delineation operation is performed on the region of interest in the current segmented image to obtain a second delineated image; The first outlined image and the second outlined image are input into the second trigger analysis model, and the trigger analysis result is output.

7. The adaptive radiation delivery determination method according to claim 6, characterized in that, The step of performing a first delineation operation on the region of interest in the current segmented image based on the original delineation result in the initial image to obtain a first delineated image includes: The original drawing result in the initial image is rigidly copied to the current segmented image to obtain a first drawing image containing the original drawing result.

8. The adaptive radiation delivery determination method according to claim 6, characterized in that, The second trigger analysis model includes a second feature extraction network and a decision network. The step of inputting the first outlined image and the second outlined image into the second trigger analysis model and outputting the trigger analysis result includes: The first outlined image is input into the second feature extraction network for feature extraction to obtain the first feature information; The second outlined image is input into the second feature extraction network for feature extraction to obtain the second feature information; The first feature information and the second feature information are input into the decision network, and the trigger analysis result is output.

9. A method for training a trigger analysis model, said trigger analysis model being used for adaptive radiographic delivery determination, characterized in that, The training method includes: Acquire training sample data; the training sample data includes initial sample images used by different sample objects when determining the initial radiographic delivery plan, sample images of the sample objects at different radiographic delivery fractions, and corresponding sample trigger analysis results; The initial trigger analysis network is trained based on the training sample data to obtain the trigger analysis model; the trigger analysis model is used to output the trigger analysis result for determining whether to perform adaptive radiodelivery.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the adaptive radiodelivery determination method according to any one of claims 1 to 8 and / or the training method of the trigger analysis model according to claim 9.

Images