A system for dynamic prediction of radiation pneumonitis

Abstract

The application provides a radioactive pneumonia dynamic prediction system, through non-rigid registration of CBCT images at multiple time points, a deformation representation graph reflecting lung tissue volume change is constructed, and a deformation reliability graph is further generated by introducing a deformation quality evaluation mechanism, so that a reliable lung area is screened or weighted in the lung range, deformation related features are extracted and constructed only in the reliable area, and the adverse effects of noise, artifacts and registration errors on subsequent feature modeling are effectively reduced, and on this basis, the application does not only use deformation information at a single time point, but organizes deformation representation results at multiple time points into a time sequence to construct trajectory features, so that the system modeling of the time evolution process of anatomical changes is realized, and the model can capture dynamic change signals related to the occurrence of radioactive pneumonia earlier and more comprehensively.

Description

Technical Field

[0001] This invention belongs to the field of dynamic prediction technology of radiation pneumonia, and specifically relates to a dynamic prediction system for radiation pneumonia. Background Technology

[0002] Radiation pneumonitis is a common radiation injury with potentially serious consequences during radiotherapy for thoracic tumors. Existing risk assessment methods for radiation pneumonitis are mainly based on static predictions using dose-volume parameters (such as MLD and V20) during the radiotherapy planning phase and some clinical factors, which fail to fully reflect the dynamic changes in the patient's lung anatomy over time during radiotherapy.

[0003] With the widespread application of cone-beam computed tomography (CBCT) in radiotherapy, lung images at multiple time points can be acquired during radiotherapy, providing a data foundation for characterizing the morphological changes of lung tissue during treatment. However, the existing technology still has the following shortcomings: 1) CBCT images have high noise and many artifacts. The deformation field obtained based on image registration is not reliable enough in some areas. Existing methods usually do not evaluate the reliability of deformation information, which makes the extracted features susceptible to error interference. 2) Existing methods mostly use only a single time point or cumulative deformation information, without systematically modeling the temporal evolution of deformation at multiple time points, making it difficult to reflect the dynamic trajectory characteristics of lung tissue changes; 3) Existing risk models typically treat the entire lung as a homogeneous structure, ignoring the differences in anatomical location, radiation dose, and biological sensitivity among different lung regions, resulting in insufficient spatial specificity and accuracy of the prediction results. Summary of the Invention

[0004] Based on this, the present invention provides a dynamic prediction system for radiation pneumonia, which aims to utilize the anatomical deformation information of cone-beam computed tomography (CBCT) images at multiple time points during radiotherapy, combined with a reliable lung area screening and regional sensitivity weighting mechanism, to achieve dynamic prediction of radiation pneumonia risk.

[0005] A first aspect of this invention provides a dynamic prediction system for radiation-induced pneumonia, the system comprising: The module is used to build patient-level time-series image data structures, including patient pCT, multi-time-node CBCT, dose, and to acquire multi-time-point image-dose datasets. The registration module is used to automatically segment the lung images after preprocessing all images to obtain the lung mask, select the reference image, and perform non-rigid registration on the CBCT at each time point to obtain the corresponding deformation field. The first calculation module is used to calculate the Jacobian determinant for each deformation field, and to calculate one or more of the indicators such as registration residual, inverse consistency error and deformation smoothness for each time point, and output the deformation characterization map and the corresponding confidence map. The processing module is used to process the confidence map using a threshold method and take the intersection of multiple time nodes as a stable and reliable lung region. The second calculation module is used to construct a time series for each voxel in the stable and reliable lung region, and calculate trajectory features based on the time series; The segmentation module is used to divide the stable and reliable lung region into several sub-regions, and to statistically summarize the trajectory features of each sub-region to obtain the region-level features, which are then spliced ​​together to form a region-level feature set. The fusion module is used to generate corresponding regional sensitivity weights based on the features of each region, and to perform weighted fusion of dose information of each sub-region to obtain a patient-level effective dose representation vector. The input module is used to input the patient-level effective dose representation vector into the trained risk prediction model and output the risk probability.

[0006] Furthermore, in the step of calculating the Jacobian determinant for each deformation field, and calculating one or more of the indicators such as registration residual, inverse consistency error, and deformation smoothness for each time point, and outputting the deformation characterization map and the corresponding confidence map, the formula for calculating the Jacobian determinant is: ; The formula for calculating the registration residual is: ; The formula for calculating inverse consistency error is: ; The formula for calculating deformation smoothness is: ; in, Let det(·) be the Jacobian determinant value at voxel x at time point k, where k = 1, 2, ..., n, and n is the total number of time points. Let be the gradient of the deformation field at time point k. The grayscale / intensity value of the reference image at voxel x. To transform the image at time point k through a deformation field After transformation to the reference image space, the grayscale / intensity value at voxel x is... The registration residual at voxel x at time point k is... The inverse consistency error at voxel x at time point k is... Let k be the deformation field at time point k. Let x be the deformation smoothness at time point k, voxel x. Let be the gradient of the displacement field at time point k; After normalizing one or more of the indicators, such as registration residual, inverse consistency error, and deformation smoothness, at each time point, a reliability map is obtained.

[0007] Furthermore, in the step of processing the confidence map using a threshold method and taking the intersection of multiple time points as the stable and reliable lung region, the expression for processing the confidence map using the threshold method is as follows: ; The expression for taking the intersection of multiple time points as the stable and reliable lung region is: ; in, For the lung mask, For the aforementioned credibility graph, For threshold parameters, To stabilize the reliable lung region, This is the credibility graph at the k-th time point.

[0008] Furthermore, in the step of constructing a time series for each voxel within the stable and reliable lung region, and calculating trajectory features based on the time series, the trajectory features include one or more of the following: cumulative deformation intensity, average rate of change, and acceleration of change. The formula for calculating the cumulative deformation intensity is: ; The formula for calculating the average rate of change is: ; The formula for calculating changing acceleration is: .

[0009] Furthermore, in the step of dividing the stable and reliable lung region into several sub-regions, statistically summarizing the trajectory features of each sub-region to obtain region-level features, and then concatenating them to form a region-level feature set, the stable and reliable lung region is divided into m sub-regions. The effective voxel set for each sub-region is For each sub-region The trajectory features are statistically summarized to form a statistical subset, including one or more of the following: mean, standard deviation, maximum value, quantile, and percentage exceeding the threshold. Based on the statistical calculation subset, determine the statistical vector of the corresponding trajectory feature; The corresponding sub-region The statistical vectors of trajectory features are concatenated to obtain the partition deformation feature vectors, forming a regional feature set.

[0010] Furthermore, the step of generating corresponding regional sensitivity weights based on the features of each region, and weighting and fusing the dose information of each sub-region to obtain the patient-level effective dose representation vector includes: Based on the dose distribution, calculate the partitioned dose feature vector for each sub-region. The partitioned dose feature vector includes one or more of the partitioned average dose, partitioned maximum dose, and partitioned volumetric dose index. Based on the linear mapping, the partition deformation feature vector is mapped to a region sensitivity score, and the region sensitivity score is normalized by Softmax to obtain the region weight. The patient-level effective dose characterization vector is obtained by weighting and summing the dose feature vectors of each region using the aforementioned regional weights.

[0011] A second aspect of the present invention provides a computer-readable storage medium storing computer instructions that enable a computer to implement the radiation pneumonia dynamic prediction system provided in the first aspect.

[0012] A third aspect of the present invention provides an electronic device, comprising: At least one processor, at least one memory, a communication interface, and a bus; wherein, The processor, memory, and communication interface communicate with each other through the bus; The memory stores program instructions that can be executed by the processor, which calls the program instructions to implement the radiation pneumonia dynamic prediction system provided in the first aspect.

[0013] The radiation pneumonia dynamic prediction system provided in this invention constructs a deformation characterization map reflecting changes in lung tissue volume by performing non-rigid registration on CBCT images at multiple time points. Furthermore, a deformation quality assessment mechanism is introduced to generate a deformation confidence map, thereby selecting or weighting a reliable lung region within the lung area. Deformation-related features are extracted and constructed only within this reliable region, effectively reducing the adverse effects of noise, artifacts, and registration errors on subsequent feature modeling. Based on this, the invention does not only utilize deformation information from a single time point but organizes the deformation characterization results from multiple time points into a time series to construct trajectory features, thereby achieving systematic modeling of the temporal evolution of anatomical changes. This allows the model to capture dynamic change signals related to the occurrence of radiation pneumonia earlier and more comprehensively. Attached Figure Description

[0014] Figure 1This is a structural block diagram of a dynamic prediction system for radiation-induced pneumonia provided in Embodiment 1 of the present invention; Figure 2 This is a structural block diagram of an electronic device provided in Embodiment 3 of the present invention. Detailed Implementation

[0015] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. Several embodiments of the invention are illustrated in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.

[0016] It should be noted that when a component is said to be "fixed to" another component, it can be directly on the other component or there may be an intervening component. When a component is said to be "connected to" another component, it can be directly connected to the other component or there may be an intervening component. The terms "vertical," "horizontal," "left," "right," and similar expressions used in this document are for illustrative purposes only.

[0017] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0018] Example 1 Please see Figure 1 , Figure 1 The diagram shows a structural block diagram of a dynamic prediction system for radiation-induced pneumonia provided in Embodiment 1 of the present invention. The dynamic prediction system 200 for radiation-induced pneumonia includes: an establishment module 21, a registration module 22, a first calculation module 23, a processing module 24, a second calculation module 25, a partitioning module 26, a fusion module 27, and an input module 28, wherein: Module 21 is used to establish a patient-level time-series image data structure, including the patient's pCT, CBCT at multiple time points, dose, and to acquire a multi-time point image-dose dataset.

[0019] In this embodiment of the invention, the patient-level time-series image data structure can be represented as follows: , For pCT, For the CBCT image data at the k-th time point, This represents the cumulative dose or planned dose mapping result.

[0020] The registration module 22 is used to automatically segment the lung images after preprocessing all images to obtain the lung mask, and select the reference image to perform non-rigid registration on the CBCT at each time point to obtain the corresponding deformation field.

[0021] Specifically, all images are subjected to voxel size standardization (e.g., 2mm×2mm×2mm) and grayscale normalization (e.g., HU truncated to [-1000, 1000] and then linearly mapped). The lung images are automatically segmented to obtain the lung mask L(x), and a reference image is selected. (Preferred pCT), finally, for Performing non-rigid registration yields the deformation field. , represented as: .

[0022] The first calculation module 23 is used to calculate the Jacobian determinant for each deformation field, and to calculate one or more of the following indices for each time point: registration residual, inverse consistency error, and deformation smoothness, and output the deformation characterization map and the corresponding confidence map.

[0023] It should be noted that the formula for calculating the Jacobian determinant is: ; The formula for calculating the registration residual is: ; The formula for calculating inverse consistency error is: ; The formula for calculating deformation smoothness is: ; in, Let det(·) be the Jacobian determinant value at voxel x at time point k, where k = 1, 2, ..., n, and n is the total number of time points. Let be the gradient of the deformation field at time point k. The grayscale / intensity value of the reference image at voxel x. To transform the image at time point k through a deformation field After transformation to the reference image space, the grayscale / intensity value at voxel x is... The registration residual at voxel x at time point k is... The inverse consistency error at voxel x at time point k is... Let k be the deformation field at time point k. Let x be the deformation smoothness at time point k, voxel x. Let be the gradient of the displacement field at time point k; After normalizing one or more of the indicators—registration residual, inverse consistency error, and deformation smoothness—at each time point, a reliability map is obtained. In this embodiment of the invention, the registration residual, inverse consistency error, and deformation smoothness are calculated at each time point, and the above indicators are normalized to obtain: , This is a credibility graph.

[0024] The processing module 24 is used to process the confidence map using a threshold method and take the intersection of multiple time nodes as a stable and reliable lung region.

[0025] Specifically, the expression for processing the confidence graph using the threshold method is as follows: ; The expression for taking the intersection of multiple time points as the stable and reliable lung region is: ; in, For the lung mask, For the aforementioned credibility graph, For threshold parameters, To stabilize the reliable lung region, This is the credibility graph at the k-th time point.

[0026] In other embodiments of the present invention, the percentile screening method or the weighted method can be used to replace the threshold method. Percentile screening method: selecting... Voxels located in the first p%; weighting method: It is directly used as a voxel weight in subsequent feature calculations.

[0027] The second calculation module 25 is used to construct a time series for each voxel in the stable and reliable lung region, and calculate trajectory features based on the time series.

[0028] Specifically, trajectory features include one or more of the following: cumulative deformation intensity, average rate of change, and acceleration. In this embodiment of the invention, a time series is constructed for each voxel x ∈ TLR(x): ; The cumulative strain strength, average rate of change, and acceleration are calculated. The formula for calculating the cumulative strain strength is: ; The formula for calculating the average rate of change is: ; The formula for calculating changing acceleration is: .

[0029] The segmentation module 26 is used to divide the stable and reliable lung region into several sub-regions, and to statistically summarize the trajectory features of each sub-region to obtain the region-level features, which are then spliced ​​together to form a region-level feature set.

[0030] In the embodiments of the present invention, the method of dividing the stable and reliable lung region is at least one of the following: Divide into dose ranges: for example, Dose ≥ D1, D2 ≤ Dose <D1, Dose<D2; Divisions based on anatomical structure: upper lobe / lower lobe, central / peripheral; Zoned by distance from the target area: d <d1, d1 ≤ d<d2, d ≥ d2; The final set of regions is obtained as R1, R2, ... .

[0031] It should be noted that the stable and reliable lung region is divided into m sub-regions. The effective voxel set for each sub-region is For each sub-region Trajectory features , and Statistical summaries were performed separately to form a statistical subset. Including the mean Standard deviation Maximum value quantiles (like ), Over-threshold ratio One or more of them; Based on the statistical calculation subset, a statistical vector corresponding to the trajectory feature is determined. For example, Then the subregion The three types of regional statistical vectors are defined as follows: ; ; ; S(·) is an operator, for example, To make feature f sum Substituting operator S into the region Ω i Calculate this statistic within the corresponding sub-region. The statistical vectors of trajectory features are concatenated to obtain the partitioned deformation feature vectors, forming a region-level feature set. For example, the concatenation result is: ; in, (like If there are 7 statistics, then ), the vector set of all sub-regions As input for later use.

[0032] The fusion module 27 is used to generate corresponding regional sensitivity weights based on the regional features, and to perform weighted fusion of dose information of each sub-region to obtain a patient-level effective dose representation vector.

[0033] Specifically, based on dose distribution Calculate the partitioned dose feature vector for each sub-region. The partitioned dose feature vector includes the partitioned average dose. Maximum dose in each region , zonal volume dose index (such as , , One or more of the following; Based on the linear mapping, the partition deformation feature vector Mapped to region sensitivity score The region sensitivity scores are then Softmax normalized to obtain region weights. For example, a linear mapping can be used. ; in and For learnable parameters, a lightweight network containing 1-2 fully connected layers can also be used. accomplish ; Furthermore, regarding Softmax normalization is performed to obtain the region weights. It satisfies the nonnegativity and normalization constraints: .

[0034] Using the region weights Dose feature vectors for each region Weighted summation yields the patient-level effective dose representation vector. , represented as: .

[0035] Input module 28 is used to input the patient-level effective dose representation vector into the trained risk prediction model and output the risk probability.

[0036] In this embodiment of the invention, a clinical feature vector is incorporated. (e.g., age, baseline lung function, comorbidities, chemotherapy history, etc.), and patient-level effective dose representation vector. Composition of model input vector , .

[0037] Furthermore, the input vector Input risk prediction model Output RP risk probability : ; in, For the Sigmoid function, This represents the probability of an RP of level ≥2 occurring. It should be noted that... Classification models such as logistic regression, support vector machine, random forest, gradient boosting tree (such as XGBoost / LightGBM) or multilayer perceptron (MLP) can be used.

[0038] Furthermore, supervised learning can be used to train the model, and the loss function can be either binary cross-entropy or focal loss to adapt to class imbalance; regularization constraints can also be introduced to improve generalization ability.

[0039] Set threshold RP risk probability Mapped to low / medium / high risk levels; when In such cases, an early warning message is output, which can be combined with a set of regional sensitivity weights. The distribution of sensitive lung areas can provide a basis for dose adjustment or follow-up strategies.

[0040] In summary, the radiation pneumonia dynamic prediction system in the above embodiments of the present invention constructs a deformation characterization map reflecting changes in lung tissue volume by performing non-rigid registration on CBCT images at multiple time points. Furthermore, it introduces a deformation quality assessment mechanism to generate a deformation confidence map, thereby selecting or weighting a reliable lung region within the lung area. Deformation-related features are extracted and constructed only within this reliable region, effectively reducing the adverse effects of noise, artifacts, and registration errors on subsequent feature modeling. Furthermore, the present invention does not merely utilize deformation information from a single time point, but organizes the deformation characterization results from multiple time points into a time series to construct trajectory features, thereby achieving systematic modeling of the temporal evolution of anatomical changes. This allows the model to capture dynamic change signals related to the occurrence of radiation pneumonia earlier and more comprehensively.

[0041] Example 2 The dynamic prediction system for radiation-induced pneumonia provided in Embodiment 2 of this invention differs from the dynamic prediction system for radiation-induced pneumonia provided in Embodiment 1 of this invention in that it does not calculate the Jacobian determinant, but instead uses the deformation displacement modulus. ; Let be the displacement field at time point k. Understandably, the trajectory feature adaptation is changed to be based on... The cumulative deformation intensity, average rate of change, and acceleration of change.

[0042] Example 3 The system in this embodiment of the invention is implemented using electronic devices; therefore, it is necessary to introduce the relevant electronic devices. For this purpose, the embodiments of the present invention provide an electronic device, such as… Figure 2 As shown, the electronic device includes at least one processor 201, a communication interface 204, at least one memory 202, and a communication bus 203. The at least one processor 201, the communication interface 204, and the at least one memory 202 communicate with each other via the communication bus 203. The at least one processor 201 can invoke logical instructions stored in the at least one memory 202 to implement various systems provided in the system embodiments.

[0043] Furthermore, the logical instructions in at least one of the aforementioned memory 202 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the system described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0044] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. 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 embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0045] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to implement the methods or systems described in the various embodiments or some parts of the embodiments.

[0046] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. Based on this understanding, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those shown in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, or sometimes in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0047] In this patent, the terms "comprising," "including," or any other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes said element.

[0048] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A dynamic prediction system for radiation-induced pneumonia, characterized in that, The system includes: The module is used to build patient-level time-series image data structures, including patient pCT, multi-time-node CBCT, dose, and to acquire multi-time-point image-dose datasets. The registration module is used to automatically segment the lung images after preprocessing all images to obtain the lung mask, select the reference image, and perform non-rigid registration on the CBCT at each time point to obtain the corresponding deformation field. The first calculation module is used to calculate the Jacobian determinant for each deformation field, and to calculate one or more of the indicators such as registration residual, inverse consistency error and deformation smoothness for each time point, and output the deformation characterization map and the corresponding confidence map. The processing module is used to process the confidence map using a threshold method and take the intersection of multiple time nodes as a stable and reliable lung region. The second calculation module is used to construct a time series for each voxel in the stable and reliable lung region, and calculate trajectory features based on the time series; The segmentation module is used to divide the stable and reliable lung region into several sub-regions, and to statistically summarize the trajectory features of each sub-region to obtain the region-level features, which are then spliced ​​together to form a region-level feature set. The fusion module is used to generate corresponding regional sensitivity weights based on the features of each region, and to perform weighted fusion of dose information of each sub-region to obtain a patient-level effective dose representation vector. The input module is used to input the patient-level effective dose representation vector into the trained risk prediction model and output the risk probability.

2. The dynamic prediction system for radiation-induced pneumonia according to claim 1, characterized in that, In the step of calculating the Jacobian determinant for each deformation field, and calculating one or more of the indicators such as registration residual, inverse consistency error, and deformation smoothness for each time point, and outputting the deformation characterization map and the corresponding confidence map, the formula for calculating the Jacobian determinant is as follows: ； The formula for calculating the registration residual is: ； The formula for calculating inverse consistency error is: ； The formula for calculating deformation smoothness is: ； in, Let det(·) be the Jacobian determinant value at voxel x at time point k, where k = 1, 2, ..., n, and n is the total number of time points. Let be the gradient of the deformation field at time point k. The grayscale / intensity value of the reference image at voxel x. To transform the image at time point k through a deformation field After transformation to the reference image space, the grayscale / intensity value at voxel x is... The registration residual at voxel x at time point k is... The inverse consistency error at voxel x at time point k is... for The corresponding inverse transform function, Let k be the deformation field at time point k. Let x be the deformation smoothness at time point k, voxel x. Let be the gradient of the displacement field at time point k; After normalizing one or more of the indicators, such as registration residual, inverse consistency error, and deformation smoothness, at each time point, a reliability map is obtained.

3. The dynamic prediction system for radiation-induced pneumonia according to claim 2, characterized in that, In the step of processing the confidence map using a threshold method and taking the intersection of multiple time nodes as the stable and reliable lung region, the expression for processing the confidence map using the threshold method is as follows: ； The expression for taking the intersection of multiple time points as the stable and reliable lung region is: ； in, For the lung mask, For the aforementioned credibility graph, For threshold parameters, To stabilize the reliable lung region, This is the credibility graph at the k-th time point.

4. The dynamic prediction system for radiation-induced pneumonia according to claim 3, characterized in that, In the step of constructing a time series for each voxel within the stable and reliable lung region, and calculating trajectory features based on the time series, the trajectory features include one or more of the following: cumulative deformation intensity, average rate of change, and acceleration of change. The formula for calculating the cumulative deformation intensity is as follows: ； The formula for calculating the average rate of change is: ； The formula for calculating changing acceleration is: 。 5. The dynamic prediction system for radiation-induced pneumonia according to claim 4, characterized in that, In the step of dividing the stable and reliable lung region into several sub-regions, statistically summarizing the trajectory features of each sub-region to obtain region-level features, and then concatenating them to form a region-level feature set, the stable and reliable lung region is divided into m sub-regions. The effective voxel set for each sub-region is For each sub-region The trajectory features are statistically summarized to form a statistical subset, including one or more of the following: mean, standard deviation, maximum value, quantile, and percentage exceeding the threshold. Based on the statistical calculation subset, determine the statistical vector of the corresponding trajectory feature; The corresponding sub-region The statistical vectors of trajectory features are concatenated to obtain the partition deformation feature vectors, forming a regional feature set.

6. The dynamic prediction system for radiation-induced pneumonia according to claim 5, characterized in that, The steps of generating corresponding regional sensitivity weights based on the features of each region, and weighting and fusing the dose information of each sub-region to obtain the patient-level effective dose representation vector include: Based on the dose distribution, calculate the partitioned dose feature vector for each sub-region. The partitioned dose feature vector includes one or more of the partitioned average dose, partitioned maximum dose, and partitioned volumetric dose index. Based on the linear mapping, the partition deformation feature vector is mapped to a region sensitivity score, and the region sensitivity score is normalized by Softmax to obtain the region weight. The patient-level effective dose characterization vector is obtained by weighting and summing the dose feature vectors of each region using the aforementioned regional weights.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that cause the computer to implement the dynamic prediction system for radiation pneumonia as described in any one of claims 1 to 6.

8. An electronic device, characterized in that, include: At least one processor, at least one memory, a communication interface, and a bus; wherein, The processor, memory, and communication interface communicate with each other through the bus; The memory stores program instructions that can be executed by the processor, which invokes the program instructions to implement the dynamic prediction system for radiation pneumonia as described in any one of claims 1 to 6.

Images