Deformation prediction method and device for tumor area, equipment and storage medium

Abstract

The invention provides a deformation prediction method and device for a tumor area, equipment and a storage medium, and the method comprises the steps: obtaining a first medical image of a target tumor area at a first target historical time, and a second medical image of the target tumor area at a second target historical time after the first target historical time; determining a basic deformation field based on the morphological change of the target tumor area of the second medical image relative to the first medical image, and inputting the basic deformation field into a trained deformation field prediction model to obtain a deformation field prediction result, the result indicates that the form of at least one future time after the second target historical time changes relative to the form of the first target historical time; a deformation field sequence sample used for model training comprises a plurality of deformation field samples sorted according to time, and the deformation field samples are obtained based on the change of the form of the tumor area sample at a second historical time after the first historical time relative to the form of the first historical time. According to the embodiment of the invention, the prediction precision of the model can be improved.

Description

Technical Field

[0001] This application relates to the field of radiotherapy technology, and more specifically, to a method, apparatus, device, and storage medium for predicting deformation of tumor regions. Background Technology

[0002] Radiotherapy is one of the core clinical treatment methods for malignant tumors. In clinical practice, it is usually carried out in the form of "fractionated radiotherapy," which achieves continuous killing of tumors through multiple courses of precise dose delivery. However, during the continuous treatment cycle of fractionated radiotherapy, the tissues and organs in the radiation area will undergo dynamic changes due to various factors, which is one of the core technical bottlenecks for improving the precision of radiotherapy.

[0003] During fractionated radiotherapy, not only does the tumor target area undergo localized non-uniform retreat due to the treatment response, but the surrounding normal organs and tissues covered by radiation (such as nearby sensitive organs like the parotid gland, lung tissue, esophagus, and spinal cord) also experience irregular changes in shape and position due to multiple factors. These include the traction effect caused by tumor retreat, as well as tissue swelling / contraction caused by the patient's physiological activities and early radiotherapy damage. These localized non-uniform spatial deformations of the target area (tumor) and surrounding normal tissues also bring corresponding clinical risks. Summary of the Invention

[0004] In view of this, this application provides a method, apparatus, device and storage medium for predicting deformation of tumor regions, so as to at least solve the problems existing in the related art.

[0005] Specifically, this application is implemented through the following technical solution: This application provides a method for predicting deformation in a tumor region, including: Acquire a medical image sequence; the medical image sequence includes a first medical image and a second medical image, the first medical image refers to a medical image of the target tumor region at a first target historical time, and the second medical image refers to a medical image of the target tumor region at a second target historical time; the second target historical time is after the first target historical time; The basic deformation field is determined based on the morphological changes of the target tumor region in the second medical image relative to the target tumor region in the first medical image. The basic deformation field is input into the trained deformation field prediction model to obtain the deformation field prediction result; the deformation field prediction result is used to indicate the change of the shape of the second target after at least one future time relative to the shape of the first target after the historical time. The trained deformation field prediction model is trained based on a set of deformation field sequence samples, wherein the set of deformation field sequence samples includes multiple deformation field sequence samples, which are sorted by time. The deformation field samples are obtained based on the change in the morphology of the tumor region sample at a second historical time relative to the morphology at a first historical time; the second historical time is any time after the first historical time.

[0006] This application also provides a deformation prediction device for a tumor region, comprising: An image sequence acquisition module is used to acquire medical image sequences; the medical image sequences include a first medical image and a second medical image, wherein the first medical image refers to a medical image of the target tumor region at a first target historical time, and the second medical image refers to a medical image of the target tumor region at a second target historical time; the second target historical time is after the first target historical time; The basic deformation field determination module is used to determine the basic deformation field based on the morphological changes of the target tumor region in the second medical image relative to the target tumor region in the first medical image. The deformation field prediction module is used to input the basic deformation field into the trained deformation field prediction model to obtain the deformation field prediction result; the deformation field prediction result is used to indicate the change of the shape of the second target after at least one future time relative to the shape of the first target after the historical time. The trained deformation field prediction model is trained based on a set of deformation field sequence samples, wherein the set of deformation field sequence samples includes multiple deformation field sequence samples, which are sorted by time. The deformation field samples are obtained based on the change in the morphology of the tumor region sample at a second historical time relative to the morphology at a first historical time; the second historical time is any time after the first historical time.

[0007] This application also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the deformation prediction method for tumor regions described in any of the foregoing embodiments.

[0008] 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 deformation prediction method for tumor regions described in any of the foregoing embodiments.

[0009] This application also provides a computer program product, including a computer program that, when run by a processor, performs the steps of any of the possible tumor region deformation prediction methods described above.

[0010] The technical solutions provided by the embodiments of this application may include the following beneficial effects: In this embodiment of the application, based on the above-mentioned deformation field sequence samples, the trained deformation field prediction model can learn the spatiotemporal correlation between the morphological changes of the tumor region after each cycle of radiotherapy (that is, the model learns the complex mapping relationship of inferring the long-term deformation based on the early deformation). In this way, the deformation field prediction results output by the trained deformation field prediction model can improve the prediction accuracy of the deformation field in the future, and thus can provide prospective planning for subsequent clinical tumor treatment.

[0011] Furthermore, in this embodiment, the tumor region includes not only the tumor target area but also organs and / or tissues affected by radiotherapy. Thus, the deformation field prediction results include not only the future deformation of the tumor target area but also the future deformation of organs and / or tissues affected by radiotherapy. This allows for the prediction of complications of radiotherapy in advance, thereby providing support for subsequent treatment plans.

[0012] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this specification. Attached Figure Description

[0013] Figure 1 This is a flowchart illustrating an exemplary embodiment of the present application of a method for predicting deformation of a tumor region; Figure 2 This is a flowchart illustrating the generation of a deformation field sequence according to an exemplary embodiment of this application; Figure 3 This is a flowchart illustrating the training process of a deformation field prediction model according to an exemplary embodiment of this application; Figure 4 This is a schematic diagram of the structure of a deformation prediction device for a tumor region, as illustrated in an exemplary embodiment of this application. Figure 5 This is a schematic diagram of the structure of another deformation prediction device for a tumor region, as illustrated in an exemplary embodiment of this application; Figure 6 This is a hardware structure diagram of a computer device illustrated in an exemplary embodiment of this application. Detailed Implementation

[0014] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0015] The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The singular forms “a,” “the,” and “the” used in this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0016] It should be understood that although the terms first, second, third, etc., may be used in this application to describe various information, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, without departing from the scope of this application, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to determination."

[0017] Radiotherapy is one of the core clinical treatment methods for malignant tumors. In clinical practice, it is usually carried out in the form of "fractionated radiotherapy," which achieves continuous killing of tumors through multiple courses of precise dose delivery. However, during the continuous treatment cycle of fractionated radiotherapy, the tissues and organs in the radiation area will undergo dynamic changes due to various factors, which is one of the core technical bottlenecks for improving the precision of radiotherapy.

[0018] During fractionated radiotherapy, not only does the tumor target area undergo localized non-uniform retreat due to the treatment response, but the surrounding irradiated normal organs and tissues (such as nearby sensitive organs like the parotid gland, lung tissue, esophagus, and spinal cord) also experience irregular changes in shape and position due to multiple factors. These include the traction effect caused by tumor retreat, as well as tissue swelling / contraction caused by the patient's physiological activities and early radiotherapy damage. These localized non-uniform spatial deformations of the target area (tumor) and surrounding normal tissues also bring corresponding clinical risks: on the one hand, they reduce the accuracy of radiotherapy dose delivery to the tumor target area, leading to insufficient dose in some tumor areas; on the other hand, they significantly increase the probability of accidental irradiation of surrounding normal tissues, and may even cause serious complications such as radiation pneumonitis and esophagitis.

[0019] Clinically, online adaptive radiotherapy (ART) is used to adjust the tumor target area and delineate organs at risk, and generate new plans online to improve the accuracy of radiotherapy. However, currently, online ART lacks intelligent judgment on the timing of initiation. Clinically, for patients who need online ART, the workflow of initiating ART in full fractions is often adopted, resulting in low work efficiency and waste of medical resources. At the same time, the time of a single online ART session is long, and patients have poor tolerance. Therefore, predicting the local non-uniform spatial deformation of the tumor target area and surrounding normal tissues, and realizing the early prediction of the timing of adaptive radiotherapy initiation, has important clinical significance and research value.

[0020] While related technologies have been studied for predicting the timing of adaptive radiotherapy initiation, they typically have the following drawbacks: (1) Insufficient adaptability to the scenario: The relevant technologies usually focus on the long-term growth of tumors or the overall spread of the target area, and are not designed for "local non-uniform deformation during the radiotherapy fractionation process", which cannot meet the actual needs of clinical radiotherapy fractionation adjustment.

[0021] (2) Insufficient spatiotemporal feature fusion: None of the related technologies have achieved deep spatiotemporal feature fusion, and have failed to take into account the relationship between time series and spatial morphology. From the input data, most of them are static medical image data of discrete time. Such data can only reflect the spatial anatomical structure information of a single time. The prediction model needs to rely on its own feature extraction ability to mine the spatial morphological transformation between two time periods, and then try to simply splice or shallowly fuse the extracted spatial transformation information with the evolution information of the time dimension, which makes it difficult for the model to accurately capture the spatiotemporal coupling characteristics.

[0022] (3) Insufficient input and output accuracy: Some image sequence-based prediction models focus on the time series changes at the image pixel level, rather than on the geometric morphology of the tumor.

[0023] (4) Lack of personalized early warning during treatment: The relevant technology does not provide patients with personalized early warning and auxiliary evaluation basis for subsequent fractionation deformation during radiotherapy. Doctors need to make accurate decisions by taking into account individual differences, which may not be able to support the need for dynamic adjustment of the treatment plan in a timely and accurate manner.

[0024] Based on the above research, this application provides a method for predicting the deformation of a tumor region. The method first acquires a medical image sequence, which includes a first medical image and a second medical image. The first medical image refers to a medical image of the target tumor region at a first target historical time, and the second medical image refers to a medical image of the target tumor region at a second target historical time. The second target historical time is after the first target historical time. Then, based on the morphological changes of the target tumor region in the second medical image relative to the target tumor region in the first medical image, a basic deformation field is determined. Finally, the basic deformation field is input into a trained deformation field prediction model to obtain a deformation field prediction result. The deformation field prediction result is used to indicate the morphological changes of the tumor region at least one future time after the second target historical time relative to the morphological changes of the tumor region at the second historical time relative to the first target historical time. The trained deformation field prediction model is trained based on a set of deformation field sequence samples, which includes multiple deformation field sequence samples, sorted by time. The deformation field samples are obtained based on the morphological changes of the tumor region sample at the second historical time relative to the first historical time. The second historical time is any time after the first historical time.

[0025] In this embodiment, based on the deformation field sequence samples including multiple deformation field samples sorted by time, the trained deformation field prediction model can learn the spatiotemporal correlation between morphological changes in tumor regions over time (that is, the model learns the complex mapping relationship of inferring long-term deformation based on early deformation). Thus, the deformation field prediction results output by the trained deformation field prediction model can improve the prediction accuracy of deformation fields in the future, and can then provide prospective planning for subsequent clinical tumor treatment.

[0026] Furthermore, in this embodiment, the tumor region includes not only the tumor target area but also organs and / or tissues affected by radiotherapy. Thus, the deformation field prediction results include not only the future deformation of the tumor target area but also the future deformation of organs and / or tissues affected by radiotherapy. This allows for the prediction of complications of radiotherapy in advance, thereby providing support for subsequent treatment plans.

[0027] To facilitate understanding of this embodiment, a method for predicting tumor region deformation disclosed in this application will first be described in detail. The execution entity of this method is generally a computer device, which can be a server. This server can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing initial cloud computing services such as cloud services, cloud databases, cloud computing, cloud storage, big data, and artificial intelligence platforms. In other embodiments, the computer device can also be a terminal device, which can be a mobile device, terminal, handheld device, computing device, vehicle-mounted device, etc.

[0028] In other embodiments, the method can also be applied to an implementation environment consisting of computer equipment and servers, or an implementation environment consisting of terminal equipment and servers. Furthermore, the method for predicting the deformation of the tumor region can also be implemented by a processor calling computer-readable instructions stored in memory.

[0029] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings.

[0030] Please see the appendix Figure 1 This is a flowchart illustrating a method for predicting deformation of a tumor region, as shown in an exemplary embodiment of this application. Figure 1 As shown, the deformation prediction method for the tumor region in this application embodiment may include the following steps S101~S103: S101: Acquire a medical image sequence; the medical image sequence includes a first medical image and a second medical image, the first medical image refers to a medical image of the target tumor region at a first target historical time, and the second medical image refers to a medical image of the target tumor region at a second target historical time; the second target historical time is after the first target historical time.

[0031] The medical imaging sequence may include, but is not limited to, at least one of the following: computed tomography (CT), color Doppler ultrasound, and magnetic resonance imaging (MRI).

[0032] Specifically, the first medical image can be a medical image of the target tumor area before radiotherapy, or a medical image of the target tumor area after radiotherapy, and the number of radiotherapy treatments can be any number, without limitation. The second medical image refers to a medical image of the target tumor area at a second target historical time, and the second target historical time is after the first target historical time.

[0033] For example, if the first medical image is a medical image of the target tumor region at the first target historical time before radiotherapy, then the second medical image can be a medical image after the first radiotherapy, or a medical image after the first target historical time before radiotherapy, or a medical image after multiple radiotherapy sessions; as another example, if the first medical image is a medical image of the target tumor region at the first target historical time after two radiotherapy sessions, then the second medical image can be a medical image at any time after the first target historical time, such as a medical image after the third radiotherapy, or a medical image at any time after the second radiotherapy and before the third radiotherapy.

[0034] In this embodiment, the target tumor region is the area formed by including the tumor target volume and normal organs and / or normal tissues (such as the parotid gland, spinal cord, lung, liver, small intestine, bladder, etc.) affected by radiotherapy to the tumor target volume. It is understood that the tumor target volume includes, but is not limited to, the gross tumor volume (GTV), clinical target volume (CTV), and planning target volume (PTV). The target tumor region can refer to any part of the body, including but not limited to the head and neck, abdomen, pelvis, and thoracic cavity.

[0035] S102: Determine the basic deformation field based on the morphological changes of the target tumor region in the second medical image relative to the target tumor region in the first medical image.

[0036] In this step, the morphological changes of the target tumor region in the second medical image relative to the target tumor region in the first medical image are determined by using the morphology of the target tumor region in the first medical image as a reference, and the basic deformation field is finally generated.

[0037] For example, taking radiotherapy cycles as an example, if the first medical image is an image of the first target at the first historical time before radiotherapy, and the second medical image is an image of the target tumor area after the first week of radiotherapy, then the basic deformation field can refer to the deformation field between week 0 and week 1; as another example, if the target medical image is an image of the target tumor area after the fourth week of radiotherapy, then the basic deformation field can refer to the deformation field between week 0 and week 4.

[0038] Here, the deformation field can refer to the deformation information of each voxel in the target tumor region. In this embodiment, each voxel has three-dimensional position information, and the deformation field can refer to the change in the three-dimensional position of each voxel in the target tumor region at the second target historical time relative to the three-dimensional position at the first target historical time. In other embodiments, the deformation field can also refer to other deformation information, such as change trajectory, change intensity, and change direction, which have the same effect as position change. Of course, in some other embodiments, the deformation information can also be derived information of the above-mentioned deformation information, such as the gradient tensor of each deformation information.

[0039] S103: Input the basic deformation field into the trained deformation field prediction model to obtain the deformation field prediction result; the deformation field prediction result is used to indicate the change in the shape of the second target after at least one future time relative to the first target after the historical time; the trained deformation field prediction model is trained based on a set of deformation field sequence samples, wherein the set of deformation field sequence samples includes multiple deformation field sequence samples, the deformation field sequence samples include multiple deformation field samples sorted by time, and the deformation field samples are obtained based on the change in the shape of the tumor region sample at the second historical time relative to the first historical time; the second historical time is any time after the first historical time.

[0040] In step S103, for the trained deformation field prediction model, the model input is the basic deformation field, and the model output is the deformation field prediction result.

[0041] The deformation field prediction results are used to indicate the changes in the shape of the second target at least one future time after the historical time, relative to the shape of the first target at the historical time.

[0042] Following the example above which uses treatment cycles as the time frame, the deformation field prediction result can indicate the changes in morphology after at least one week of radiotherapy following the Mth week of radiotherapy, relative to the morphology without radiotherapy.

[0043] In some implementations, the deformation field prediction results include a predicted deformation field sequence; the predicted deformation field sequence includes multiple predicted deformation fields ordered by time, the multiple predicted deformation fields being used to characterize the changes in the morphology of the target tumor region at multiple consecutive future times after the second target historical time relative to the morphology of the first target historical time.

[0044] For example, if the first target historical time T1 corresponds to the first form and the second target historical time T2 corresponds to the second form, then multiple consecutive future times can be T3, T4 and T5. Thus, multiple predicted deformation fields include: the change of the form of future time T3 relative to the first form, the change of the form of future time T4 relative to the first form, and the change of the form of future time T5 relative to the first form.

[0045] It should be noted that the time interval between each future time and the previous future time is the same as the time interval between the first target historical time and the second target historical time. For example, if the time interval between the first target historical time and the second target historical time is ΔT, then the time interval between future time T3 and the second target historical time T2 is also ΔT, and the time interval between future time T4 and future time T3 is also ΔT.

[0046] Following the example of the treatment cycles above, the predicted deformation field sequence includes multiple predicted deformation fields ordered according to the predicted treatment cycles. These multiple predicted deformation fields are used to characterize the changes in the morphology of the target tumor region after multiple consecutive weeks of radiotherapy following the Mth week of radiotherapy, relative to the morphology without radiotherapy. Specifically, if M is 1, the predicted deformation field sequence includes the predicted deformation fields corresponding to the second, third, and fourth weeks, respectively; if M is 2, the predicted deformation field sequence includes the predicted deformation fields corresponding to the fourth, sixth, and eighth weeks, respectively.

[0047] Optionally, the upper limit of the number of predicted deformation fields in the predicted deformation field sequence can be set according to actual needs. Specifically, the upper limit can be obtained in response to user input, or it can be a preset number, which is not limited here.

[0048] In other embodiments, the deformation field prediction result may also include a predicted deformation field, wherein the predicted deformation field may be a week adjacent to the Mth week or a week not adjacent to the Mth week.

[0049] For example, if M is 1, then at least one week after the first week of radiotherapy can be the third week, the fourth week, or the seventh week.

[0050] In this embodiment of the application, the deformation field sequence sample set includes multiple deformation field sequence samples, wherein each deformation field sequence sample corresponds to a patient.

[0051] The deformation field sequence samples include multiple deformation field samples ordered by treatment weeks (e.g., week 1, week 2, ... week 5). The deformation field samples are obtained based on the changes in the morphology of the tumor region sample after the Nth week of radiotherapy relative to its initial morphology. Thus, by training the deformation field prediction model to be trained based on multiple deformation field sequence samples, the trained deformation field prediction model can learn the spatiotemporal correlation between the morphological changes of the tumor region after each week of radiotherapy (week sequence and morphological change correlation). Therefore, by inputting the basic deformation field into the trained deformation field prediction model, the deformation field prediction result can be obtained.

[0052] The training process for the trained deformation field prediction model will be described in detail later.

[0053] In this embodiment of the application, based on the above-mentioned deformation field sequence samples, the trained deformation field prediction model can learn the spatiotemporal correlation between the morphological changes of the tumor region after each cycle of radiotherapy (that is, the model learns the complex mapping relationship of inferring the long-term deformation based on the early deformation). In this way, the deformation field prediction results output by the trained deformation field prediction model can improve the prediction accuracy of the deformation field in the future, and thus can provide prospective planning for subsequent clinical tumor treatment.

[0054] Furthermore, in this embodiment, the tumor region includes not only the tumor target area but also organs and / or tissues affected by radiotherapy. Thus, the deformation field prediction results include not only the future deformation of the tumor target area but also the future deformation of organs and / or tissues affected by radiotherapy. This allows for the prediction of complications of radiotherapy in advance, thereby providing support for subsequent treatment plans.

[0055] In some implementations, after obtaining the deformation field prediction result in step S103, the target tumor region in the first medical image can be morphologically adjusted based on the deformation field prediction result to obtain a predicted medical image.

[0056] In this embodiment, the predicted deformation field (such as a set of displacement vectors) from the deformation field prediction result is applied (or "mapped") to a first medical image serving as a spatial reference. Specifically, based on the displacement vector of each voxel in the predicted deformation field, the position of the corresponding voxel in the first medical image can be moved or deformed to simulate the morphology of the target tumor region at a future time. In this way, a new, synthetic medical image (or the most critical outline of the target tumor region) can be obtained. On this image, the target tumor region has changed accordingly based on the prediction, visually demonstrating "what the target tumor region will look like in a future week if treatment continues as originally planned".

[0057] Furthermore, after obtaining the predicted medical images, clinical decision information can be generated based on the predicted medical images. For example, the volume change and overlap (Dice coefficient) between the predicted morphology and the original planned target area can be automatically calculated, or the predicted morphology can be superimposed with the original dose distribution to recalculate key dosimetric parameters and determine whether they exceed the clinical safety threshold.

[0058] Specifically, predictive medical images can be applied to at least one of the following scenarios: (1) Deformation warning: When the predicted deformation (such as target area shrinkage / displacement) reaches the preset warning threshold, a warning message is generated to remind the doctor to pay attention.

[0059] (2) Radiotherapy initiation assessment: Specifically refers to adaptive radiotherapy (ART) initiation assessment. By analyzing the dosimetric results under predicted morphology, an assessment conclusion is given on whether to recommend re-formulating the radiotherapy plan, providing an objective basis for initiating the time-consuming and laborious adaptive radiotherapy process.

[0060] (3) Adjustment of radiotherapy plan: Based on the prediction information, doctors can make advance and proactive adjustments to the treatment parameters of subsequent fractions (such as the direction of the irradiation field, weight, etc.) or to develop a completely new adaptive radiotherapy plan.

[0061] (4) Deformation compensation: Based on the predicted medical images, the target tumor area is deformed and compensated.

[0062] In this embodiment, the practicality can be improved by using a closed-loop solution that integrates data processing, intelligent prediction, and clinical decision-making.

[0063] Optionally, based on the deformation field prediction results, visualization information (such as deformation heat maps, volume change curves, DVH comparison charts, etc.) can be generated to provide corresponding medical support for clinical treatment.

[0064] The following is combined with Figure 2 The generation process of the deformation field sequence is described in detail. Figure 2 This is a flowchart illustrating the generation of a deformation field sequence, provided as an exemplary embodiment of this application. Figure 2 As shown, steps S201 to S203 are included: S201: Obtain a medical image sample sequence; the medical image sample sequence includes a first medical image sample and a plurality of second medical image samples; the first medical image sample is a medical image of the tumor region at the first historical time, and the plurality of second medical image samples are medical images of the tumor region at a plurality of consecutive second historical times after the first historical time.

[0065] The image types of each medical image sample in the medical image sample sequence may include, but are not limited to, at least one of computed tomography (CT), color Doppler ultrasound, and magnetic resonance imaging (MRI). Preferably, the image types of each image sample may be the same, for example, all of them may be CT images or all of them may be MRI images.

[0066] In this embodiment, the tumor region is the area formed by including the tumor target volume and normal organs and / or normal tissues (such as the parotid gland, spinal cord, lung, liver, small intestine, bladder, etc.) affected by radiotherapy to the tumor target volume. It is understood that the tumor target volume includes, but is not limited to, the gross tumor volume (GTV), clinical target volume (CTV), and planning target volume (PTV). The tumor region can refer to any part of the body, including but not limited to the head and neck, abdomen, pelvis, and thoracic cavity.

[0067] The first medical image sample is a medical image of the tumor region at the first historical time. Multiple second medical image samples are medical images of the tumor region at multiple consecutive second historical times after the first historical time. This sequence of medical image samples completely represents the deformation and evolution of the tumor region of a specific patient throughout the entire treatment process.

[0068] For example, a medical image sample sequence may include multiple medical image images of a patient arranged according to the shooting time, including A1, A2, A3, A4, A5, and A6. Then, the first medical image sample may be A1, and the multiple second medical image samples may be A2, A3, A4, A5, and A6, respectively. Alternatively, the first medical image sample may be A1, and the multiple second medical image samples may be A3 and A5, respectively. Or, the first medical image sample may be A2, and the multiple second medical image samples may be A3, A4, A5, and A6, respectively. Or, the first medical image sample may be A2, and the multiple second medical image samples may be A4 and A6, respectively.

[0069] In addition, it should be noted that the first medical image sample can be a medical image without radiotherapy or a medical image after radiotherapy, without any limitation. The second historical time of the second medical image sample can be after the first historical time.

[0070] Regarding step S201, when acquiring the medical image sample sequence, the following steps (a) to (c) may be included: (a) Obtaining an initial medical image sequence; the initial medical image sequence includes multiple initial medical images ordered by time; each initial medical image has identification information for the tumor region.

[0071] Here, the identification information for the tumor region can refer to identification information (such as an identification box) that distinguishes the tumor region from other unrelated regions.

[0072] (b) For each initial medical image, based on the identification information of the tumor region, the initial medical image is cropped to obtain a local image including the tumor region.

[0073] In this way, for each initial medical image, the region can be cropped based on the identification information to obtain a local image including the tumor region.

[0074] Optionally, before cropping the initial medical images, standardized preprocessing can be performed. Specifically, standardized preprocessing can include windowing, grayscale normalization, and uniform pixel spacing. Windowing is used to adjust the display contrast and brightness of the image. The images are stored in Henle units, whose numerical range (typically -1000 to +3000 HU) far exceeds the grayscale range of ordinary display devices.

[0075] Windowing involves setting a window width and a window level. The window width determines the image contrast; the narrower the window, the higher the contrast. The window level determines the image brightness. In this way, the center point of the HU value of interest can be linearly mapped to the entire grayscale level of the display, thereby clearly displaying the boundaries of the tumor target area (GTV) and the surrounding soft tissue organs affected by radiotherapy, providing the best visual basis for subsequent delineation, extraction, and deformation calculation.

[0076] Gray-scale normalization refers to scaling the pixel values ​​of an image to a fixed numerical range, usually [0, 1], thereby unifying the intensity values ​​of all images to the same scale, avoiding deviations caused by scale differences, and making the model optimization process more stable and faster.

[0077] Uniform pixel spacing refers to adjusting the physical size of voxels in all images to be consistent. Since deformation field calculations, distance measurements, and volume calculations all depend on the consistency of the true physical size represented by voxels, inconsistent pixel spacing will severely distort the calculated displacements and volumes. Furthermore, deep learning models (especially convolutional networks) typically require input data to have a regular grid structure in the spatial dimension. Uniform pixel spacing ensures the consistency of spatial sampling for all samples.

[0078] In some implementations, when a region can be cropped based on the identification information for each initial medical image to obtain a local image including the tumor region, it is also possible to crop a region based on the identification information for each initial medical image to obtain an initial local image including the tumor region, and then perform spatial cropping on the initial local image to obtain a local image.

[0079] Specifically, for each initial medical image, the three-dimensional geometric center of the tumor target area (GTV) extracted from the initial medical image is used as the spatial reference point. Around this spatial reference point, a fixed voxel range is extracted along the X, Y, and Z axes in the image coordinate system to obtain an initial local image. Thus, each initial local image is cropped into an image block with the same three-dimensional size. This operation ensures that the image data of each segment are spatially aligned in the local coordinate system centered on the tumor.

[0080] (c) Generate the medical image sample sequence based on each local image and the corresponding time.

[0081] After obtaining local images including the tumor region based on the above steps, the medical image sample sequence is generated by combining each local effect with the corresponding time.

[0082] In this embodiment of the application, the preprocessing described above can improve the standardization of each image sample, thereby improving the accuracy of the model.

[0083] S202: For each second medical image sample, the deformation field of the tumor region is determined based on the change in the morphology of the tumor region in the second medical image sample relative to the morphology of the tumor region in the first medical image sample.

[0084] For each second medical image sample, its deformation field relative to the reference image (first medical image sample) is calculated using techniques such as image registration. Each deformation field accurately records the cumulative spatial deformation of the tumor region at that specific cycle.

[0085] In this embodiment, differential geometry tools are used to model the tumor region in relation to the first treatment fraction (e.g., week 0) and any subsequent treatment fraction (e.g., week 1, week 2, etc.). By calculating the displacement and other features of the geometric shape of the tumor region between adjacent fractions, deformation fields between fractions are generated to characterize the local non-uniform deformation pattern of the tumor region.

[0086] S203: Based on the time information of each second medical image sample and the deformation field corresponding to the tumor region in each second medical image sample, generate the deformation field sequence sample.

[0087] Here, temporal information refers to the time point label of each second medical image sample and its corresponding deformation location. This information is key metadata for constructing spatiotemporal sequences and enabling the model to understand the temporal attributes of the deformation process.

[0088] All the deformation fields obtained in step S202 are arranged in their corresponding time order, and time tags are associated and bound to each deformation field to form a complete deformation field sequence sample.

[0089] For example, if we take the treatment week as the time label, for a patient with follow-up data for weeks 1, 2, and 3, the final generated sample is a deformation field sequence [week 0 → week 1, week 0 → week 2, week 0 → week 3], and each is attached with a week label (or time label) [1, 2, 3].

[0090] In this embodiment of the application, the first medical image sample is used as a unified spatial reference. The anatomical morphological changes at different future time points are quantified into a set of deformation fields with time labels, thereby constructing standardized sequence data for training the spatiotemporal prediction model. This provides a high-quality data foundation for the model to learn the law of "how deformation evolves over time".

[0091] In some implementations, patient health information corresponding to the medical image sample sequence can also be obtained, and deformation field sequence samples can be generated based on the patient health information, the time information of each second medical image sample, and the deformation field corresponding to the tumor region in each second medical image sample.

[0092] The patient's health information may include age, medication dosage, treatment method, and physical condition (such as blood pressure).

[0093] Furthermore, after generating the deformation field sequence samples, it can be obtained through methods such as... Figure 3 The model training flowchart shown illustrates the steps for training the model to obtain the trained deformation field prediction model. The architecture of the deformation field prediction model in this application includes an encoder, a spatiotemporal modeling module, and a decoder, specifically including steps S301 to S305: S301: Input each deformation field sequence sample in the deformation field training sample set into the encoder to obtain deformation field sequence features.

[0094] Here, multiple deformation field sequences can be input into the encoder in sequence, and the encoder can extract features from each deformation field in the deformation field sequence to obtain the deformation field sequence features.

[0095] S302: For each deformation field sequence feature, construct a graph structure corresponding to the deformation field sequence based on the spatial neighborhood relationship between each voxel; the graph structure is used to characterize the spatial correlation of the deformation of the tumor region with the treatment cycle.

[0096] The tumor region comprises multiple voxels, and the deformation field includes deformation information of each voxel.

[0097] In this step, based on the spatial neighborhood relationships (such as vertical, horizontal, front-back, and vertical) between voxels in the deformation field, the feature map (deformation field sequence features) at each time point is transformed into a graph structure. Each voxel corresponds to a node in the graph structure, and the node feature is the value of that voxel in the encoded features. Connecting edges are established between spatially adjacent voxel nodes. By connecting nodes in the same spatial location, the graphs at different times are combined into a spatiotemporal graph structure. Based on this graph structure, the complex spatial relationships of tumor region deformation (such as how the contraction of a certain voxel affects neighboring voxels) and temporal connections can be represented more flexibly and effectively.

[0098] S303: The spatiotemporal modeling module is used to fuse the temporal information of the deformation field sequence and the graph structure to generate spatiotemporal fusion features.

[0099] Among them, the spatiotemporal modeling module (such as graph attention network) transmits and aggregates information between nodes along the spatial and temporal edges of the graph structure. Specifically, it includes spatial fusion: nodes collect information from their spatial neighbors and learn the interaction of local deformations; temporal fusion: nodes obtain information from their own state in previous time and learn the evolution trajectory of deformations, and finally obtain spatiotemporal fusion features. That is, after fusion, the features of each node not only include its own information, but also incorporate its spatiotemporal context.

[0100] In some implementations, the temporal information of the deformation field sequence can be encoded using one-hot encoding to obtain temporal features, and then fused with the graph structure (such as splicing, weighted fusion, dynamic weight allocation driven by graph attention mechanism, etc.) to obtain spatiotemporal fusion features.

[0101] S304: Input the spatiotemporal fusion features into the decoder to obtain the first predicted deformation field sequence.

[0102] Here, the decoder (usually composed of deconvolutional layers or graph upsampling operations) upsamples and transforms the spatiotemporal fusion features, mapping them back to the same dimension and spatial resolution as the deformation field sequence, to obtain the first predicted deformation field sequence.

[0103] For example, still taking the treatment week as the time label, if the deformation field sequence includes the first week deformation field, the second week deformation field, then the first predicted deformation field sequence may include the third week deformation field, the fourth week deformation field, and so on. Of course, the number of deformation fields included in the first predicted deformation field sequence can be set according to actual needs. For example, the number can be a fixed value, such as 2 or 3, or the number can be the same as the number of deformation fields in the actual deformation field sequence.

[0104] It should be noted that during the training process described above, multiple deformation fields in the deformation field sequence can be divided into input data and corresponding labels (i.e., the real deformation field sequence) according to time.

[0105] Optionally, the deformation fields within the first time range of the deformation field sequence can be used as input data, and multiple consecutive deformation fields after the first time range can be used as labels. For example, taking the treatment weeks as the time label, the deformation fields of the first 3 weeks can be used as input data, and the deformation fields of the 4th and 5th weeks can be used as labels.

[0106] Optionally, the deformation field at any time in the deformation field can be used as input data, and multiple consecutive deformation fields after that time can be used as labels; alternatively, the deformation field within a first time range in the deformation field sequence can be used as input data, and the deformation field at any time in multiple consecutive deformation fields after the first time range can be used as labels.

[0107] For example, still using treatment weeks as the time label, the deformation fields of the first m weeks in the deformation field sequence can be used as input data, and the deformation fields of the next n consecutive weeks after the m-th week can be used as labels, where m is a positive integer greater than or equal to 1, and n is a positive integer greater than or equal to 1. For example, the deformation fields of the first 3 weeks can be used as input data, and the deformation fields of the 4th and 5th weeks can be used as labels. Alternatively, the deformation field of the m-th week in the deformation field sequence can be used as input data, and the deformation fields of the next n consecutive weeks after the m-th week can be used as labels. Still alternatively, the deformation fields of the first m weeks in the deformation field sequence can be used as input data, and the deformation field of any week in the next n consecutive weeks can be used as labels.

[0108] S305: Determine the model loss based on the first predicted deformation field sequence and the actual deformation field sequence, and adjust the parameters of the deformation field prediction model based on the model loss.

[0109] In this embodiment, the above process is supervised training.

[0110] Specifically, when determining the model loss based on the first predicted deformation field sequence and the actual deformation field sequence, the difference between the first predicted deformation field sequence and the actual deformation field sequence can be determined first to determine the error loss. Then, based on the three-dimensional displacement gradient of each voxel in each first predicted deformation field in the first predicted deformation field sequence, the smoothness loss can be determined. Finally, based on the mean error loss and the smoothness loss, the model loss can be determined.

[0111] Among them, the error loss includes, but is not limited to, any one of mean square error loss, mean absolute error loss, and cosine similarity loss, and the smoothness loss is used to characterize the smoothness of the first predicted deformation field in physical space.

[0112] In other implementations, physical constraint loss can also be determined, and model loss can be determined based on physical constraint loss, error loss, and smoothness loss. Physical constraint loss may include deformation continuity loss, volume conservation loss, etc., which are not limited here.

[0113] In this embodiment, the optimizer used during training is the Adam optimizer, the initial learning rate is set to 1e-4, and a learning rate decay strategy is configured to improve training stability. The number of training iterations is set to 10,000 rounds. The dataset is randomly divided into training set and validation set in a ratio of 8:2. An early stopping strategy is adopted during training. Training is terminated when the loss of the validation set does not decrease for several consecutive rounds, which effectively avoids model overfitting and ensures the generalization ability of the model.

[0114] It should be noted that, in the embodiments of this application, in addition to the above-mentioned technical solution for constructing deformation fields based on ensemble differential tools, traditional image registration methods or deep learning registration methods can also be used to complete the construction of inter-level deformation fields; the architecture of the above-mentioned deformation field prediction model can also be other combinations, such as using fully connected neural networks or temporal prediction networks (e.g., Transformer networks, LSTM prediction networks, etc.) to replace graph neural networks to complete deformation field prediction; each solution maintains the core logic of data preprocessing, spatial uniform cropping, and clinical application unchanged, among which the combination of traditional registration + simple network is adapted to clinical scenarios with low modeling complexity requirements, and the combination of deep learning registration + high-precision temporal network is adapted to high-precision prediction scenarios.

[0115] Corresponding to the embodiments of the aforementioned tumor region deformation prediction method, this application also provides embodiments of a tumor region deformation prediction device.

[0116] Please see Figure 4 This is a schematic diagram illustrating the structure of a deformation prediction device for a tumor region, as shown in an exemplary embodiment of this application. Figure 4 As shown, the deformation prediction device 400 for the tumor region includes: The image sequence acquisition module 410 is used to acquire a medical image sequence; the medical image sequence includes a first medical image and a second medical image, wherein the first medical image refers to a medical image of the target tumor region at a first target historical time, and the second medical image refers to a medical image of the target tumor region at a second target historical time; the second target historical time is after the first target historical time. The basic deformation field determination module 420 is used to determine the basic deformation field based on the morphological changes of the target tumor region in the second medical image relative to the target tumor region in the first medical image. The deformation field prediction module 430 is used to input the basic deformation field into the trained deformation field prediction model to obtain the deformation field prediction result; the deformation field prediction result is used to indicate the change of the shape of the second target after at least one future time relative to the shape of the first target after the historical time. The trained deformation field prediction model is trained based on a set of deformation field sequence samples, wherein the set of deformation field sequence samples includes multiple deformation field sequence samples, which are sorted by time. The deformation field samples are obtained based on the change in the morphology of the tumor region sample at a second historical time relative to the morphology at a first historical time; the second historical time is any time after the first historical time.

[0117] In some implementations, the deformation field prediction results include a predicted deformation field sequence; the predicted deformation field sequence includes multiple predicted deformation fields ordered by time, the multiple predicted deformation fields being used to characterize the changes in the morphology of the target tumor region at multiple consecutive future times after the second target historical time relative to the morphology at the first target historical time.

[0118] Please see Figure 5 This is a schematic diagram illustrating the structure of a deformation prediction device for a tumor region, as shown in an exemplary embodiment of this application. Figure 5 As shown, the deformation prediction device 400 for the tumor region further includes a sequence sample generation module 440, which is used for: Obtain a medical image sample sequence; the medical image sample sequence includes a first medical image sample and a plurality of second medical image samples; the first medical image sample is a medical image of the tumor region at a first historical time, and the plurality of second medical image samples are medical images of the tumor region at a plurality of consecutive second historical times after the first historical time; For each second medical image sample, the deformation field of the tumor region is determined based on the change in the morphology of the tumor region in the second medical image sample relative to the morphology of the tumor region in the first medical image sample. Based on the week information of each second medical image sample and the deformation field corresponding to the tumor region in each second medical image sample, the deformation field sequence sample is generated.

[0119] In some embodiments, the deformation prediction device 400 for the tumor region further includes a model training module 450, wherein the tumor region comprises multiple voxels, and the deformation field includes deformation information of each voxel; the model training module 450 is used for: Multiple deformation field sequences from the deformation field training sample set are input into the encoder to obtain deformation field sequence features; For each deformation field sequence feature, a graph structure corresponding to the deformation field sequence is constructed based on the spatial neighborhood relationship between each voxel; the graph structure is used to characterize the spatial correlation of deformation changes in the tumor region over time. The spatiotemporal modeling module is used to fuse the temporal information of the deformation field sequence and the graph structure to generate spatiotemporal fusion features; The spatiotemporal fusion features are input into the decoder to obtain the first predicted deformation field sequence; Based on the first predicted deformation field sequence and the actual deformation field sequence, the model loss is determined, and the parameters of the deformation field prediction model are adjusted based on the model loss.

[0120] In some implementations, the first predicted deformation field sequence includes multiple first predicted deformation fields; the model training module 450 is specifically used for: Based on the three-dimensional displacement gradient of each voxel in each of the first predicted deformation fields in the first predicted deformation field sequence, a smoothness loss is determined; the smoothness loss is used to characterize the smoothness of the first predicted deformation field in physical space. Determine the error loss between the first predicted deformation field and the actual deformation field sequence; The model loss is determined based on the smoothness loss and the error loss.

[0121] In some embodiments, the sequence sample generation module 440 is specifically used for: Acquire an initial medical image sequence; the initial medical image sequence includes multiple initial medical images ordered chronologically; each initial medical image has identification information for the tumor region; For each initial medical image, based on the identification information of the tumor region, the initial medical image is cropped to obtain a local image including the tumor region; The medical image sample sequence is generated based on each local image and its corresponding time.

[0122] In some embodiments, the deformation prediction device 400 for the tumor region further includes an image generation module 460, which is used for: Based on the predicted medical images, clinical decision information is generated; the clinical decision information is used for at least one of the following: deformation warning of the target tumor region, radiotherapy initiation assessment, or radiotherapy plan adjustment.

[0123] In some embodiments, the deformation prediction device 400 for the tumor region further includes a decision information generation module 470, which is used for: Based on the predicted medical images, clinical decision information is generated; the clinical decision information is used for at least one of the following: deformation warning of the target tumor region, radiotherapy initiation assessment, or radiotherapy plan adjustment.

[0124] The specific implementation process of the functions and roles of each unit in the above device can be found in the implementation process of the corresponding steps in the above method, and will not be repeated here.

[0125] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this application according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0126] Corresponding to the above-mentioned method for predicting deformation of tumor regions, this application also provides a computer device, such as... Figure 6 The diagram shown is a structural schematic of a computer device provided in an embodiment of this application. Figure 6As shown, the computer device 600 includes a processor 610, an internal bus 620, memory 630, a network interface 640, and non-volatile memory 650, and may also include other hardware required for its functions. One or more embodiments of this specification can be implemented in software, for example, the processor 610 reads the corresponding computer program from the non-volatile memory 650 into the memory 630 and then runs it. Of course, besides software implementation, one or more embodiments of this specification do not exclude other implementation methods, such as logic devices or a combination of hardware and software, etc. That is to say, the execution entity of the following processing flow is not limited to individual logic units, but can also be hardware or logic devices.

[0127] The memory 630, also known as internal memory, is used to temporarily store the computational data in the processor 610, as well as the data exchanged with non-volatile memory 650 such as hard disk. The processor 610 exchanges data with non-volatile memory 650 through the memory 630.

[0128] In this embodiment, memory 630 is specifically used to store application code that executes the solution of this application, and its execution is controlled by processor 610. That is, when the computer device is running, processor 610 communicates with network interface 640, memory 630 and non-volatile memory 650 through internal bus 620, so that processor 610 executes the application code stored in memory 630 and non-volatile memory 650, thereby executing the tumor region deformation prediction method described in the above method embodiment.

[0129] Processor 610 may be an integrated circuit chip with signal processing capabilities. The aforementioned processor can be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it can also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware microservices. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor can be a microprocessor or any conventional processor.

[0130] It is understood that the structures illustrated in the embodiments of this application do not constitute a specific limitation on the computer device 600. In other embodiments of this application, the computer device 600 may include more or fewer components than illustrated, or combine some components, or split some components, or have different component arrangements. The illustrated components may be implemented in hardware, software, or a combination of software and hardware.

[0131] This application also provides a computer-readable storage medium storing a computer program. When executed by a processor, the computer program performs the steps of the tumor region deformation prediction method described in the above-described method embodiments. The storage medium can be either volatile or non-volatile computer-readable storage.

[0132] This application also provides a computer program product carrying program code. The program code includes instructions that can be used to execute the steps of the tumor region deformation prediction method in the above method embodiments. For details, please refer to the above method embodiments, which will not be repeated here.

[0133] The aforementioned computer program product can be implemented through hardware, software, or a combination thereof. In one optional embodiment, the computer program product is specifically embodied in a computer storage medium; in another optional embodiment, the computer program product is specifically embodied in a software product, such as a software development kit (SDK), etc.

[0134] The embodiments of the subject matter and functional operation described in this specification can be implemented in the following ways: digital electronic circuits, tangibly embodied computer software or firmware, computer hardware including the structures disclosed in this specification and their structural equivalents, or combinations thereof. Embodiments of the subject matter described in this specification can be implemented as one or more computer programs, i.e., one or more modules of computer program instructions encoded on a tangible, non-transitory program carrier for execution by a data processing apparatus or for controlling the operation of a data processing apparatus. Alternatively or additionally, the program instructions may be encoded on artificially generated propagation signals, such as machine-generated electrical, optical, or electromagnetic signals, which are generated to encode information and transmit it to a suitable receiving device for execution by the data processing apparatus. The computer storage medium may be a machine-readable storage device, a machine-readable storage substrate, a random or serial access memory device, or combinations thereof.

[0135] The processing and logic flow described in this specification can be executed by one or more programmable computers that execute one or more computer programs to perform corresponding functions by operating on input data and generating output. The processing and logic flow can also be executed by dedicated logic circuitry—such as FPGAs (Field-Programmable Gate Arrays) or ASICs (Application-Specific Integrated Circuits), and the device can also be implemented as dedicated logic circuitry.

[0136] Computers suitable for executing computer programs include, for example, general-purpose and / or special-purpose microprocessors, or any other type of central processing unit. Typically, the central processing unit receives instructions and data from read-only memory and / or random access memory. Basic computer microservices include a central processing unit for implementing or executing instructions and one or more memory devices for storing instructions and data. Typically, a computer will also include one or more mass storage devices for storing data, such as disks, magneto-optical disks, or optical disks, or the computer will be operatively coupled to such mass storage devices to receive data from or transfer data to them, or both. However, a computer is not required to have such devices. Furthermore, a computer can be embedded in another device, such as a mobile phone, a personal digital assistant (PDA), a mobile audio or video player, a game console, a global positioning system (GPS) receiver, or a portable storage device such as a universal serial bus (USB) flash drive, to name a few.

[0137] Computer-readable media suitable for storing computer program instructions and data include all forms of non-volatile memory, media, and memory devices, such as semiconductor memory devices (e.g., EPROM, EEPROM, and flash memory devices), magnetic disks (e.g., internal hard disks or removable disks), magneto-optical disks, and CD-ROM and DVD-ROM disks. Processors and memory may be supplemented by or incorporated into dedicated logic circuitry.

[0138] While this specification contains numerous specific implementation details, these should not be construed as limiting the scope of any invention or the scope of the claims, but rather are primarily intended to describe features of specific embodiments of a particular invention. Certain features described in the various embodiments herein may also be implemented in combination in a single embodiment. Conversely, various features described in a single embodiment may also be implemented separately in various embodiments or in any suitable sub-combination. Furthermore, while features may function in certain combinations as described above and even initially claimed in this way, one or more features from a claimed combination may be removed from that combination in some cases, and a claimed combination may refer to a sub-combination or a variation thereof.

[0139] Similarly, although the operations are depicted in a specific order in the accompanying drawings, this should not be construed as requiring these operations to be performed in the specific order shown or sequentially, or requiring all illustrated operations to be performed to achieve the desired result. In some cases, multitasking and parallel processing may be advantageous. Furthermore, the separation of various system modules and microservices in the above embodiments should not be construed as requiring such separation in all embodiments, and it should be understood that the described program microservices and systems can generally be integrated together in a single software product or packaged into multiple software products.

[0140] Thus, specific embodiments of the subject matter have been described. Other embodiments are within the scope of the appended claims. In some cases, the actions recited in the claims may be performed in a different order and still achieve the desired result. Furthermore, the processes depicted in the drawings are not necessarily shown in a specific order or sequence to achieve the desired result. In some implementations, multitasking and parallel processing may be advantageous.

[0141] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for predicting deformation in a tumor region, characterized in that, include: Acquire a medical image sequence; the medical image sequence includes a first medical image and a second medical image, the first medical image refers to a medical image of the target tumor region at a first target historical time, and the second medical image refers to a medical image of the target tumor region at a second target historical time; the second target historical time is after the first target historical time; The basic deformation field is determined based on the morphological changes of the target tumor region in the second medical image relative to the target tumor region in the first medical image. The basic deformation field is input into the trained deformation field prediction model to obtain the deformation field prediction result; the deformation field prediction result is used to indicate the change of the shape of the second target after at least one future time relative to the shape of the first target after the historical time. The trained deformation field prediction model is trained based on a set of deformation field sequence samples, wherein the set of deformation field sequence samples includes multiple deformation field sequence samples, which are sorted by time. The deformation field samples are obtained based on the change in the morphology of the tumor region sample at a second historical time relative to the morphology at a first historical time; the second historical time is any time after the first historical time.

2. The method according to claim 1, characterized in that, The deformation field prediction result includes a predicted deformation field sequence; the predicted deformation field sequence includes multiple predicted deformation fields ordered by time, and the multiple predicted deformation fields are used to characterize the changes in the morphology of the target tumor region at multiple consecutive future times after the second target historical time relative to the morphology at the first target historical time.

3. The method according to claim 1, characterized in that, The deformation field sequence sample is generated through the following steps: Obtain a medical image sample sequence; the medical image sample sequence includes a first medical image sample and a plurality of second medical image samples; the first medical image sample is a medical image of the tumor region at a first historical time, and the plurality of second medical image samples are medical images of the tumor region at a plurality of consecutive second historical times after the first historical time; For each second medical image sample, the deformation field of the tumor region is determined based on the change in the morphology of the tumor region in the second medical image sample relative to the morphology of the tumor region in the first medical image sample. Based on the week information of each second medical image sample and the deformation field corresponding to the tumor region in each second medical image sample, the deformation field sequence sample is generated.

4. The method according to claim 3, characterized in that, The tumor region comprises multiple voxels, and the deformation field includes deformation information of each voxel; the deformation field prediction model includes an encoder, a spatiotemporal modeling module, and a decoder. The trained deformation field prediction model is obtained through the following training steps: Multiple deformation field sequences from the deformation field training sample set are input into the encoder to obtain deformation field sequence features; For each deformation field sequence feature, a graph structure corresponding to the deformation field sequence is constructed based on the spatial neighborhood relationship between each voxel; the graph structure is used to characterize the spatial correlation of deformation changes in the tumor region over time. The spatiotemporal modeling module is used to fuse the temporal information of the deformation field sequence and the graph structure to generate spatiotemporal fusion features; The spatiotemporal fusion features are input into the decoder to obtain the first predicted deformation field sequence; Based on the first predicted deformation field sequence and the actual deformation field sequence, the model loss is determined, and the parameters of the deformation field prediction model are adjusted based on the model loss.

5. The method according to claim 4, characterized in that, The first predicted deformation field sequence includes multiple first predicted deformation fields; the step of determining the model loss based on the difference between the first predicted deformation field sequence and the actual deformation field sequence includes: Based on the three-dimensional displacement gradient of each voxel in each of the first predicted deformation fields in the first predicted deformation field sequence, a smoothness loss is determined; the smoothness loss is used to characterize the smoothness of the first predicted deformation field in physical space. Determine the error loss between the first predicted deformation field and the actual deformation field sequence; The model loss is determined based on the smoothness loss and the error loss.

6. The method according to claim 3, characterized in that, The acquisition of medical image sample sequences includes: Acquire an initial medical image sequence; the initial medical image sequence includes multiple initial medical images ordered chronologically; each initial medical image has identification information for the tumor region; For each initial medical image, based on the identification information of the tumor region, the initial medical image is cropped to obtain a local image including the tumor region; The medical image sample sequence is generated based on each local image and its corresponding time.

7. The method according to claim 1, characterized in that, After obtaining the deformation field prediction result, the method further includes: Based on the deformation field prediction results, the target tumor region in the first medical image is morphologically adjusted to obtain a predicted medical image.

8. The method according to claim 7, characterized in that, The method further includes: Based on the predicted medical images, clinical decision information is generated; the clinical decision information is used for at least one of the following: deformation warning of the target tumor region, radiotherapy initiation assessment, or radiotherapy plan adjustment.

9. A device for predicting deformation of a tumor region, characterized in that, The device includes: An image sequence acquisition module is used to acquire medical image sequences; the medical image sequences include a first medical image and a second medical image, wherein the first medical image refers to a medical image of the target tumor region at a first target historical time, and the second medical image refers to a medical image of the target tumor region at a second target historical time; the second target historical time is after the first target historical time; The basic deformation field determination module is used to determine the basic deformation field based on the morphological changes of the target tumor region in the second medical image relative to the target tumor region in the first medical image. The deformation field prediction module is used to input the basic deformation field into the trained deformation field prediction model to obtain the deformation field prediction result; the deformation field prediction result is used to indicate the change of the shape of the second target after at least one future time relative to the shape of the first target after the historical time. The trained deformation field prediction model is trained based on a set of deformation field sequence samples, wherein the set of deformation field sequence samples includes multiple deformation field sequence samples, which are sorted by time. The deformation field samples are obtained based on the change in the morphology of the tumor region sample at a second historical time relative to the morphology at a first historical time; the second historical time is any time after the first historical time.

10. A computer device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the deformation prediction method for the tumor region according to any one of claims 1-8.

11. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements the steps of the deformation prediction method for tumor regions according to any one of claims 1-8.

Images