Multi-center, multi-tumor-species and multi-prescription dosage prediction method for pelvic tumor and related equipment

Abstract

The invention provides a pelvic tumor multi-center, multi-tumor-species and multi-prescription dose prediction method and related equipment, and is applied to the technical field of data processing. The method comprises the following steps: preprocessing a training sample set to generate a training sample set with target feature data; processing a preset dose distribution processing model based on a SwinTransform network architecture based on the training sample set with the target feature data, and generating a target dose distribution processing model; performing feature extraction processing on the CT image information of the target user to generate a target feature vector; processing the physiological state information of the target user to generate a dose limiting factor of the target user; processing the physiological parameter information of the target user to generate a dynamic dose influence factor of the target user; and processing the target feature vector, the dose limiting factor of the target user and the dynamic dose influence factor of the target user based on a target dose distribution processing model to generate target dose distribution information.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a dose prediction method and related equipment for multi-center, multi-tumor and multi-prescription pelvic tumors. Background Art

[0002] Radiotherapy is one of the important means of treating pelvic tumors. The accuracy of radiotherapy faces challenges due to the daily changes of pelvic organs, such as changes in the filling of the bladder and rectum, and changes in the tumor target area during treatment. In addition, considering that pelvic tumors include cervical cancer, prostate cancer, bladder cancer and other diseases, a single deep learning dose prediction model is difficult to be compatible with all diseases and ensure that the dose prediction for all diseases is accurate enough. Therefore, it is necessary to carry out personalized model design research.

[0003] Existing technologies are often models obtained by training a single center based on a single tumor type and a single prescription. The dose distribution of other types of tumors with the same prescription can only be obtained through methods such as model fine-tuning, but this often sacrifices the accuracy of dose prediction. For different prescriptions, even the same type of tumor cannot be accurately predicted. Moreover, compared with the model directly trained using the corresponding type and prescription tumor radiotherapy plan, the predicted dose distribution error obtained by these schemes such as model fine-tuning is significantly increased, and there is a problem of unstable output quality. In addition, the combination of treatment planning systems, linear accelerators, and measurement equipment from different manufacturers may affect the classification performance of the prediction model, which limits the universality and generalization ability of the model.

[0004] It should be noted that the information disclosed in the above background technology section is only used to enhance the understanding of the background of the present disclosure, and therefore may include information that does not constitute the prior art known to ordinary technicians in the field. Summary of the invention

[0005] The purpose of this application is to provide a multi-center, multi-tumor, and multi-prescription dose prediction method and related equipment for pelvic tumors, at least to a certain extent, to overcome the problems existing in the prior art, by carefully pre-processing the sample set, mining features into the original library according to the multi-element grouping of cancer, and screening key features through a deep neural network after scientifically dividing the set, training an innovative architecture model, and iteratively optimizing. When processing the target user's CT image, the target vector is obtained by enhancing normalization, dynamic segmentation, and deep fusion of texture and other features. Analyze the dose interval and limiting factors of the physiological state, and mine physiological parameters to generate dynamic influencing factors. Finally, multiple factors are integrated according to the target model, and coding, warnings, and dynamic values ​​are generated according to rules to construct a personalized dose report, improve the quality and efficiency of radiotherapy, and drive the precision process.

[0006] Other features and advantages of the present application will become apparent from the following detailed description, or may be learned in part by the practice of the present invention.

[0007] According to one aspect of the present application, a dose prediction method for multi-center, multi-tumor and multi-prescription pelvic tumors is provided, comprising: obtaining CT image information of a target user, physiological state information of the target user, physiological parameter information of the target user, a training sample set and a preset dose distribution processing model based on a SwinTransformer network architecture, wherein the CT image information of the target user includes CT image information of different cancer information of the target user, the physiological state information of the target user includes the age, gender, physical condition and specific characteristics of the tumor of the target user, and the training sample set includes CT image information of different users with different cancer information; preprocessing the training sample set to generate a training sample set with target feature data; based on the training sample set with The training sample set of target feature data is processed on the preset dose distribution processing model based on the SwinTransformer network architecture to generate a target dose distribution processing model; the CT image information of the target user is subjected to feature extraction processing to generate a target feature vector; the physiological state information of the target user is processed to generate a dose limitation factor of the target user; the physiological parameter information of the target user is processed to generate a dynamic dose influence factor of the target user; the target feature vector, the dose limitation factor of the target user and the dynamic dose influence factor of the target user are processed based on the target dose distribution processing model to generate target dose distribution information, wherein the target dose distribution information is used to characterize the dose distribution report corresponding to different cancer information.

[0008] Another aspect of the present application is a multi-center, multi-tumor, and multi-prescription dose prediction device for pelvic tumors, characterized in that it includes: an acquisition module for acquiring CT image information of a target user, physiological state information of the target user, physiological parameter information of the target user, a training sample set, and a preset dose distribution processing model based on a SwinTransformer network architecture, wherein the CT image information of the target user includes CT image information of different cancer information of the target user, the physiological state information of the target user includes the age, gender, physical condition, and tumor-specific characteristics of the target user, and the training sample set includes CT image information of different users with different cancer information; a processing module for preprocessing the training sample set to generate a training sample set with target feature data; Sample set; based on the training sample set with target feature data, the preset dose distribution processing model based on the SwinTransformer network architecture is processed to generate a target dose distribution processing model; the CT image information of the target user is subjected to feature extraction processing to generate a target feature vector; the physiological state information of the target user is processed to generate a dose limitation factor of the target user; the physiological parameter information of the target user is processed to generate a dynamic dose influence factor of the target user; based on the target dose distribution processing model, the target feature vector, the dose limitation factor of the target user and the dynamic dose influence factor of the target user are processed to generate target dose distribution information, wherein the target dose distribution information is used to characterize the dose distribution report corresponding to different cancer information.

[0009] According to another aspect of the present application, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a second processor, the above-mentioned dose prediction method for multi-center, multi-tumor type and multi-prescription pelvic tumors is implemented.

[0010] The present application provides a multi-center, multi-tumor, and multi-prescription dose prediction method and related equipment for pelvic tumors. The sample set is carefully preprocessed, and features are mined into the original library according to the multi-element grouping of cancer. After scientifically dividing the set, the key features are screened through a deep neural network, and the innovative architecture model is trained and iteratively optimized. When processing the CT image of the target user, the target vector is obtained by enhancing normalization, dynamic segmentation, and deep fusion of texture and other features. The dose interval and limiting factors of the physiological state are analyzed, and the physiological parameters are mined to generate dynamic influencing factors. Finally, multiple factors are integrated according to the target model, and coding, warnings, and dynamic values ​​are generated according to the rules to construct a personalized dose report, improve the quality and efficiency of radiotherapy, and drive the precision process.

[0011] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1A flow chart showing a method for predicting dose of pelvic tumors in multiple centers, multiple tumor types and multiple prescriptions provided in one embodiment of the present application; Figure 2 A schematic diagram of the structure of a multi-center, multi-tumor, and multi-prescription dose prediction device for pelvic tumors provided in one embodiment of the present application is shown; Figure 3 A diagram showing a multi-head and multi-tail training mode for a preset dose distribution processing model based on a SwinTransformer network architecture provided by an embodiment of the present application is shown. DETAILED DESCRIPTION

[0013] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0014] Combine the following Figure 1 To describe the dose prediction method for multi-center, multi-tumor and multi-prescription pelvic tumors according to the exemplary embodiment of the present application. It should be noted that the following application scenarios are only shown to facilitate understanding of the spirit and principles of the present application, and the embodiments of the present application are not limited in this regard. On the contrary, the embodiments of the present application can be applied to any applicable scenario.

[0015] In one embodiment, the present application also proposes a dose prediction method and related equipment for multi-center, multi-tumor, and multi-prescription pelvic tumors. Figure 1 The flowchart of a method for predicting dose of pelvic tumors in multiple centers, multiple tumor types and multiple prescriptions according to an embodiment of the present application is schematically shown. Figure 1 As shown, the method is applied to a server, comprising: S101, obtaining CT image information of a target user, physiological state information of the target user, physiological parameter information of the target user, a training sample set, and a preset dose distribution processing model based on a SwinTransformer network architecture.

[0016] In one embodiment, in the pelvic tumor radiotherapy dose prediction scenario, accurate determination of dose distribution is crucial to improving intervention effects and reducing side effects. Take a 60-year-old male user as an example. He suffers from prostate cancer and mild cardiovascular disease (comorbidities), and his nutritional status is good (normal body mass index and normal serum protein levels). A high-resolution CT image data set of the user's pelvic area is obtained from the hospital imaging system, including axial, coronal and sagittal images. The image shows the morphology, size, location and adjacent relationship of the prostate tumor with surrounding tissues (bladder, rectum, seminal vesicle, etc.). The tumor is circular, with a diameter of about 3 cm, located in the peripheral zone of the right lobe of the prostate, with clear boundaries, uneven internal density, and low-density necrotic areas, which are typical features of CT images of different cancer information. In the axial image, the distance between the tumor and the posterior wall of the bladder and the anterior wall of the rectum can be clearly measured, providing a key basis for the design of organ protection strategies for radiotherapy planning, such as determining the safe boundary of the irradiation field to avoid dangerous organs, affecting the dose distribution adjustment.

[0017] Age factors: 60-year-old users have declining physical functions and weaker tissue repair capabilities than young people. Radiotherapy dose planning needs to be more cautious to prevent normal tissue damage from causing serious complications, such as radiation cystitis and proctitis. Gender affects the growth metabolism of prostate cancer. Androgens are closely related to tumor progression. Endocrine interventions such as androgen receptor antagonists can change the sensitivity of tumors to radiotherapy. Radiotherapy doses are adjusted dynamically according to the intervention stage and response. Cardiovascular disease is a complication of users. Although it is mild, it affects cardiac function reserve and circulatory stability. The heart is affected by scattered rays during pelvic radiotherapy. Cardiac tolerance is assessed based on cardiac ejection fraction and myocardial enzyme indicators. The radiotherapy plan optimizes the field direction and modulates the dose rate to reduce the heart dose load. The dose is tracked and adjusted in real time according to changes in the condition. In terms of tumor-specific characteristics, the pathological type is adenocarcinoma, with glandular duct-like structure and specific cell morphology. Its invasiveness, metastatic potential, and radiotherapy sensitivity are different from other types of prostate cancer. The basic dose range is determined according to the pathological subtype; the degree of differentiation is moderate, and the cell atypia is moderately different from normal tissue. The dose intensity is adjusted according to the degree of differentiation to ensure a balance between tumor control and function retention; genetic testing shows specific gene mutations (such as BRCA2 mutations), which affect the DNA repair pathway and make the tumor sensitive to radiotherapy. The dose is accurately adjusted according to genetic characteristics to achieve personalized and precise radiotherapy, improve efficacy and reduce side effects.

[0018] Regarding physiological motion parameters, wearable sensors are used to continuously monitor respiratory waveforms. The respiratory cycle is about 4-6 seconds, the inhalation amplitude is about 800-1000ml, and the exhalation amplitude is about 600-800ml. Respiratory motion causes prostate tumor displacement of about 0.5-1cm. The respiratory parameters provide data for respiratory gating technology or dose compensation algorithm to ensure that the tumor receives uniform and accurate doses in different phases of respiratory motion. ECG signal monitoring shows heart rate of 70-80 beats / min, sinus rhythm, no significant changes in ST segment, stable cardiac rhythm and function, but micro-displacement of the heart and dose fluctuations are still considered during radiotherapy, and dose distribution is dynamically optimized based on ECG signals. Blood oxygen saturation is 95%-98%, reflecting normal tissue oxygenation, which has little effect on radiotherapy sensitivity, but continuous monitoring and early warning of oxygenation changes are used to adjust the dose. In terms of physiological structure information, MRI or CT reconstruction technology is used to construct a three-dimensional model of pelvic organs, showing that the prostate capsule is intact, the bladder wall thickness is uniform, and there is no obvious lesion in the rectum and the course of peripheral blood vessels. During radiotherapy, the dose distribution is adjusted according to the organ's morphological structure and relative position changes (such as changes in bladder and rectal filling causing organ displacement) to ensure tumor coverage while protecting normal tissue.

[0019] The SwinTransformer network architecture plays a core role in the pelvic tumor dose prediction model, and its unique design meets the needs of complex tumor radiotherapy scenarios. Overall architecture and multi-head multi-tail mode: The overall network has a hierarchical structure, which includes multiple stages and gradually extracts features from low-level to high-level. The multi-head multi-tail mode is a key innovation. For different tumor types such as cervical cancer, prostate cancer, bladder cancer, rectal cancer and endometrial cancer, each has an independent input path (head). Taking the cervical cancer input path as an example, the image data is initially processed by a specific convolutional layer and normalization operation, and the basic features such as tumor location and morphology are extracted before entering the shared network layer. The shared network layer is composed of multiple SwinTransformer modules stacked together. Each module contains a multi-head attention mechanism and a multi-layer perceptron, which captures the association between local and global features through window multi-head attention. For example, in the feature extraction of prostate cancer, attention is paid to the relationship between the tumor and the surrounding neurovascular bundles and the internal microstructural heterogeneity. The training data of different tumor types are shared here for weight parameter optimization. The output layer customizes independent branches (tails) for each tumor and outputs the corresponding dose distribution prediction results. This not only ensures tumor type-specific processing, but also improves learning efficiency and model generalization through shared parameters, reducing the risk of overfitting.

[0020] Feature extraction and fusion mechanism: After the input CT image is enhanced and normalized to improve the quality and comparability, the convolution layer initially extracts the contour and texture features. In the SwinTransformer module, the multi-head attention mechanism focuses on different feature dimensions according to different heads. One head focuses on the change of tumor cell density to reflect tumor activity, and the other head captures the organ boundary to accurately locate the target area and endangered organs. The multi-layer perceptron fuses the feature conversion and constructs a descriptor of the complex feature relationship between tumors and organs. For example, in the treatment of bladder cancer, the tumor cell proliferation characteristics and the bladder wall muscle layer structural characteristics are integrated to provide a detailed basis for dose calculation. The model learning process integrates tumor imaging features, physiological state and parameter association features, extracts fusion features from multi-source information of tumor characteristics (pathology, stage, gene), user physiology (age, gender, comorbidities) and physiological motion structure (respiration, electrocardiogram, organ displacement structure), optimizes the network weights through back propagation, accurately maps the relationship between features and dose distribution, and realizes accurate dose prediction under complex clinical factors.

[0021] The training sample set contains a large number of pelvic tumor user CT images and corresponding radiotherapy plan data from multiple centers, covering cases of cervical cancer, prostate cancer, bladder cancer, rectal cancer, and endometrial cancer of all ages, stages, and pathological types. Taking prostate cancer as an example, there are early localized cases that are cured by radical radiotherapy, and there are also cases of palliative radiotherapy for symptom relief of advanced bone metastasis. Based on this sample set, the SwinTransformer network architecture model is trained, and its multi-head and multi-tail mode sets independent input branches and shared network weight parameters for different tumor types. The input CT image is subjected to multi-layer convolution and attention mechanism to extract tumor and organ features, and the dose limit and dynamic dose factor are integrated to output high-precision dose distribution prediction results. The model learns the relationship between tumor imaging features and dose, the user's physiological characteristics and dose response laws, and the effect of physiological movement and structural changes on dose. After a large amount of data training and verification, the performance is optimized to provide reliable dose distribution prediction for new users (such as the above case), and realize multi-center, multi-tumor, and multi-prescription accurate radiotherapy dose planning for pelvic tumors. The sources of multi-center data are rich and diverse, covering the medical levels, equipment differences and patient population characteristics in different regions, providing a broad data basis for model learning, helping it to grasp the radiotherapy dosage rules of multiple tumor types under different conditions, and avoid deviations caused by the limitations of single-center data. For example, genetic differences among people in different regions affect tumor occurrence and development and radiotherapy response. Multi-center data fusion can fully capture such difference characteristics.

[0022] Integrate multi-center data training models to overcome the limitations of single-center data, enable the model to learn data characteristics and laws in multiple scenarios, and enhance the adaptability of radiotherapy dose calculation for different regions, equipment, and populations. For example, the international multi-center data fusion model is applicable to radiotherapy institutions around the world, whether in developed or developing regions, to accurately handle the dose requirements of multiple tumor types, promote global standardization and precision application of radiotherapy technology, and reduce dose calculation errors and fluctuations in treatment effects caused by regional differences. Process multi-center data and complete secondary optimization of multi-center data to obtain homogeneous dose distribution, ultimately solving the problem of uneven quality of patient data from multiple machines and multiple centers.

[0023] S102, preprocessing the training sample set to generate a training sample set with target feature data.

[0024] In one embodiment, the training data set is grouped to generate a grouped training data set, wherein the grouped training data set includes physiological characteristic information of users of different age groups and different cancer information. Massive pelvic tumor user data is collected from multiple centers and divided into children (0-14 years old), youth (15-44 years old), middle-aged (45-64 years old), and elderly (65 years old and above) groups according to age. The children's group has multiple embryonic tumors, such as sacrococcygeal teratomas, which grow fast and are sensitive to radiotherapy; the elderly group has more prostate cancer and cervical cancer, which are related to tissue degeneration and hormonal changes. The types of cancer include cervical cancer, squamous cell carcinoma and adenocarcinoma subtypes according to cancer cell differentiation and HPV infection, and the tumor morphology, growth rate and radiotherapy response are different; prostate cancer is divided into multiple types according to Gleason score and genetic characteristics; bladder cancer has great differences in pathological stage and tumor grade; rectal cancer affects surgical and radiotherapy strategies according to location and pathological type; endometrial cancer affects prognosis and intervention according to estrogen receptor status. In terms of quantity, the proportion of tumor samples of each type is allocated according to incidence and research focus to ensure balanced model learning. In terms of duration, early-stage tumors are localized and symptoms are mild, while late-stage tumors are metastatic and symptoms are severe. Grouping helps the model grasp the characteristics of tumor progression and lays the foundation for accurate dose prediction.

[0025] The grouped training data set is subjected to feature extraction to generate the original feature library, and the original feature library is divided into various feature data sets to generate training sets and validation sets. After grouping, image features are extracted from the data, such as the texture of the tumor area in the CT image (grayscale co-occurrence matrix calculates roughness, and contrast reflects tumor heterogeneity), shape (area, perimeter, and circularity evaluate growth mode and invasiveness), density (CT value difference identifies tumor tissue composition); MRI image functional parameters (ADC value indicates tumor cell density and diffusion) and physiological parameters (blood routine reflects immune inflammation, liver and kidney function indicators are associated with metabolic detoxification, and cardiopulmonary function test data help evaluate radiotherapy tolerance), which are collected into the original feature library. Then, the training set and validation set are divided according to the ratio (such as 7:3) or stratified sampling, and the training set model and validation set parameters are optimized to ensure model performance and generalization.

[0026] Based on the classifier, the validation sets divided by the original feature library are predicted to generate prediction results. The preset algorithm is used to divide the training sets in the original feature library for training to generate validation set prediction results. The prediction results and validation set prediction results are processed to generate target feature data. The target feature data is used to characterize the image features and physiological parameters that affect the user's physiological abnormalities. The validation set is predicted by classifiers such as decision trees and support vector machines to evaluate the initial performance of the model and the effectiveness of the features. The training set is trained using algorithms such as random gradient descent and Adam, and the network weights are optimized according to the loss function (such as mean square error and cross entropy) to obtain the validation set prediction results. The prediction and verification results are integrated to select key image textures, physiological parameters, etc. as target feature data. For example, specific texture feature combinations can accurately locate tumor boundaries and key physiological parameters can reflect the user's radiotherapy tolerance, providing a core basis for personalized and precise radiotherapy dose decisions, improving intervention effects and reducing side effects.

[0027] S103, processing a preset dose distribution processing model based on the SwinTransformer network architecture based on a training sample set with target feature data to generate a target dose distribution processing model.

[0028] In one embodiment, if Figure 3 As shown in the figure, in the data collection and integration stage, a large amount of pelvic tumor case data was collected from multiple medical institutions, including CT images, medical records, genetic testing reports, etc. of users of different age groups (children to the elderly), cancer types (cervical cancer, prostate cancer, bladder cancer, rectal cancer, endometrial cancer), stages and physiological states, to construct the original training sample set. According to the grouping of tumor characteristics and user physiological parameters, the feature data such as tumor image texture, shape, metabolic parameters and user age-related physiological decline indicators and gene mutations affecting radiotherapy sensitivity were extracted and integrated into the sample set to form a training sample set with target feature data.

[0029] The preset model of the SwinTransformer network architecture includes an input layer, a multi-layer feature extraction module, a fusion layer, and an output layer. The input layer accepts the training sample set image and feature data, and is initially processed by convolution and normalization. In the feature extraction module, the window multi-head attention mechanism focuses on the local microstructure and overall morphology of the tumor, and different heads capture features such as tumor cell density, nuclear-cytoplasmic ratio, peripheral vascular distribution, and relative position relationship of organs; the multi-layer perceptron integrates features and mines nonlinear relationships, such as analyzing the interactive effects of gene-image features on dose distribution. The fusion layer integrates multi-module features, coordinates the weights of tumors and physiological factors, and adjusts the network weights after repeated training and optimization to make the model output dose distribution close to clinical reality. For example, in the case of cervical cancer, the model learns the growth characteristics of tumors in different stages, the relationship between user age and constitution and radiotherapy tolerance, accurately predicts the dose distribution of tumor targets and surrounding organs, and generates an individualized target dose distribution processing model, providing key technical support for accurate irradiation of pelvic tumors and improving efficacy and safety.

[0030] In another implementation, the SwinTransformer model was evaluated on a dataset we prepared, which contained 500 pelvic tumor radiotherapy plans, including 100 cases each of cervical cancer, prostate cancer, bladder cancer, rectal cancer, and endometrial cancer, for a total of 500 cases. The patient dataset was randomly divided into a training set of 400 cases, a validation set of 25 cases, and a test set of 75 cases. These plans specify the maximum prescription doses dmax, dmedium_1, dmedium_2, ... dmedium_n, and dmin for different target areas, according to the prescription from large to small. Multi-cancer information provides the data cornerstone for multi-prescription processing, and the model architecture is the core computing engine, which can mine the dose distribution rules from complex data, such as learning the dose association rules under different tumor types and individual differences of patients from different cancer images and physiological characteristics, laying a solid foundation for multi-prescription dose calculation. Different cancer pathological types, stages, locations, and individual differences of patients require diversified radiotherapy dose schemes. Multi-prescription processing accurately calculates the dose for each tumor type. For example, for tumors in different parts of the brain and brain tumors of different pathological types, personalized doses are determined based on the malignancy of the tumor, the distribution of surrounding nerves and blood vessels, etc. For example, different doses are prescribed for gliomas and meningiomas, which improves the tumor control rate, reduces the risk of recurrence, and strikes tumor cells with precise doses while protecting normal tissues.

[0031] The input of the model includes multiple channels, the first of which is the planning target volume (PTV) image, in which the voxels within the PTV are assigned corresponding prescription dose values, which are normalized according to dmax. For example, if the prescription dose is 50Gy, 45Gy or 40Gy, then in the PTV image, the corresponding voxel value will be set to 1, 0.9 or 0.8. For different values ​​of prescription dose, the PTV image is defined as follows. in, They are The other 2-4 channels are the normalized CT image, the dose mask, and all the combinations of organs at risk. The output of the model is a normalized dose image that predicts the dose distribution in the entire 3D space, including the target area at different prescription doses.

[0032] When multiple tumor types occur concurrently, radiotherapy for each tumor type affects each other, and multi-prescription treatment coordinates the planning of doses to avoid conflicts and achieve synergy. For example, for patients with lung cancer and liver metastasis, the physiological characteristics, tolerance and tumor development of different organs of the lung and liver are taken into consideration, and the prescription doses of lung and liver tumors are coordinated to prevent lung tumor treatment from damaging the liver or liver doses from affecting lung respiratory function, optimize the overall radiotherapy effect, improve the patient's quality of life, prolong survival, and ensure that radiotherapy for multiple tumor types is orderly and efficient. During radiotherapy, the patient's physiological state and tumor change dynamically, and multi-prescription treatment adjusts the dose in real time according to physiological parameters and tumor progression. If the patient is weak during radiotherapy or the tumor changes its response to dose, the dose plan is optimized in a timely manner to ensure that the dose is accurate and effective throughout the radiotherapy, and to continuously maintain a balance between treatment effect and safety, so that radiotherapy can be flexibly adjusted according to the condition and patient status, and the overall treatment quality and reliability can be improved.

[0033] S104, performing feature extraction processing on the CT image information of the target user to generate a target feature vector.

[0034] In one implementation, the CT image of the target user is enhanced and normalized to generate a target CT image. After obtaining the pelvic CT image of the target user, the image enhancement technology performs histogram equalization, filtering denoising, and contrast stretching operations based on the image characteristics and tumor characteristics. For example, histogram equalization improves the grayscale distribution of the image, making the fine structure and edges inside the tumor clearer in low-contrast areas; filtering denoising removes image artifacts and noise interference to ensure that the image details are true and reliable; contrast stretching highlights the difference between tumors and normal tissues and improves recognition. Normalization processing unifies the image pixel values ​​to a specific range (such as 0-1), eliminates the impact of differences in equipment and scanning parameters, and lays a solid foundation for subsequent quantitative analysis. After this processing, a high-quality target CT image is generated to prepare high-quality data for accurate tumor feature extraction.

[0035] The target CT image is dynamically segmented to generate some initial tumor region information. Dynamic segmentation accurately outlines the tumor: When dynamically segmenting the target CT image, the adaptive region growing algorithm combines image gradient, grayscale information and prior knowledge to flexibly define the tumor range. The algorithm starts from the seed point (determined by the doctor's mark or automatic detection of suspected tumor points), grows toward the periphery based on similarity criteria (such as grayscale difference, texture consistency), continuously adjusts the growth threshold to adapt to changes in the tumor boundary, accurately outlines the tumor contour, and generates initial tumor region information. This process takes into account the irregular shape of the tumor, blurred edges, and adhesions of surrounding tissues. For example, when prostate cancer invades the capsule and surrounding tissues, the algorithm accurately distinguishes between tumors and normal tissues, defines the accurate range for subsequent feature analysis, avoids misjudgment and miscutting, and improves the accuracy of radiotherapy.

[0036] The texture feature analysis and processing of several initial tumor region information were performed to generate tumor texture information. For the initial tumor region, the grayscale co-occurrence matrix method was used to calculate the texture parameters. High contrast shows that the tumor cells are unevenly distributed and the internal structure is complex; correlation reflects the regularity of the texture, and low correlation means that the tumor grows disorderly; entropy measures the randomness of the texture, and high entropy indicates that the tumor is highly heterogeneous. For example, the texture analysis of cervical cancer tumors helps doctors understand the differences between the active proliferation area and the necrotic area of ​​tumor cells, providing a microscopic basis for radiotherapy dose distribution planning, ensuring that the dose in the high-activity area is sufficient and the tumor core is accurately hit. The shape feature quantification processing of several initial tumor region information was performed to generate tumor morphological information. When quantifying the shape features, parameters such as tumor area, perimeter, circularity, and compactness were calculated. Area is positively correlated with tumor load; the ratio of perimeter to area evaluates the complexity of the tumor edge, and a high ratio indicates that the tumor is highly invasive and has irregular boundaries; a circularity close to 1 is a regular circle, and the more it deviates from 1, the more irregular the shape and the more invasive it is. Taking bladder cancer as an example, the quantification of shape features helps doctors determine the growth pattern of the tumor (expandability or invasiveness), providing a key reference for radiotherapy field design and dose adjustment, and achieving a balance between full tumor coverage and protection of normal tissue.

[0037] Several initial tumor region information are grayed out to generate tumor gray feature information. After graying, the gray mean, standard deviation, skewness, kurtosis and other statistics of the tumor region are analyzed. The gray mean reflects the mean of tissue density. The density of tumor tissue is often higher than that of normal tissue. The standard deviation measures the degree of gray value dispersion. A large standard deviation indicates uneven density inside the tumor. Skewness and kurtosis describe the symmetry and sharpness of gray distribution, which helps to identify differences in tumor tissue components. For example, the gray features of tumors with different Gleason grades of prostate cancer are different, which provides a strong basis for accurate staging and personalized dose formulation, and improves the accuracy and efficacy of radiotherapy. The tumor texture information, tumor morphology information and tumor gray feature information are fused to generate target feature information. The texture, shape and gray features are fused into target feature information, using methods such as principal component analysis (PCA) or neural network. PCA reduces the dimension to extract key feature components, retains the core information and reduces data redundancy; the neural network learns the fusion weights adaptively based on feature association. The fusion features comprehensively characterize the tumor morphology, structure, density and heterogeneity, and provide key input for the radiotherapy planning system to accurately plan the dose distribution. For example, in rectal cancer radiotherapy, the irradiation field shape, size, angle and dose intensity distribution are accurately designed based on the fusion features, thereby improving the tumor control rate, reducing complications, and optimizing the user intervention experience and prognosis.

[0038] S105, processing the physiological status information of the target user to generate a dose limitation factor of the target user.

[0039] In one embodiment, the physiological state information of the target user is processed to generate the target user's cancer type characteristic information and the target user's physiological indicator information, and the target user's cancer type characteristic information is processed to generate the dose preset interval value corresponding to each target area information. Based on the target user's physiological indicator information, the dose preset interval value corresponding to each target area information is processed to generate the target user's dose limiting factor. Take a 60-year-old female user as an example. She suffers from cervical cancer and endometrial cancer at the same time, which brings great challenges to radiotherapy dose planning. Characteristics of cervical cancer: Pathology is squamous cell carcinoma, the tumor grows exophytically, is about 4cm×3cm in size, is located at the external os of the cervix, invades the upper 1 / 3 of the vagina, and the FIGO stage is ⅡB. Genetic testing shows that the PIK3CA gene mutation affects cell proliferation and radiotherapy sensitivity, indicating that the dose strategy may need to be adjusted to enhance the efficacy. The morphological characteristics of squamous cell carcinoma and genetic changes jointly outline the tumor invasiveness and radiotherapy response characteristics, which are the core basis for dose planning. Characteristics of endometrial cancer: Pathology is endometrioid adenocarcinoma, moderately differentiated, the tumor diffusely infiltrates more than 1 / 2 of the uterine myometrium, the uterus is enlarged as if it were 8 weeks pregnant, and there is no obvious involvement of the bilateral adnexa, which is stage III. The microsatellite instability test is highly unstable (MSI-H), which suggests the potential for immune intervention, but in terms of radiotherapy dose considerations, tumor stage, differentiation and genetic characteristics jointly determine its control difficulty and the need for normal tissue protection, especially the depth of invasion of the uterine myometrium affects the dose distribution strategy of the uterus and surrounding organs.

[0040] Physiological index assessment: The user suffers from hypertension (blood pressure controlled at around 130 / 80 mmHg), mild diabetes (HbA1c 6.5%), and osteoporosis. Routine blood tests showed mild anemia (Hb 100 g / L), normal liver and kidney function indicators, and mild cardiopulmonary function impairment (maximum oxygen uptake 22 ml / kg / min). These physiological indicators affect radiotherapy tolerance and dose safety range from different dimensions. For example, hypertension and diabetes affect microvascular lesions and tissue perfusion, anemia interferes with tissue oxygen supply and radiotherapy sensitivity, and cardiopulmonary function impairment restricts the intensity and implementation of radiotherapy plans, which are key considerations for generating dose limiting factors.

[0041] Target area dose range for cervical cancer: According to NCCN guidelines and clinical experience, for radical radiotherapy of stage IIB cervical cancer, the conventional PTV dose for cervical tumors is 80-85Gy. Due to tumor size, invasion range and genetic characteristics, the planned PTV dose is 82-85Gy to ensure tumor control. Dose limits for organs at risk: Rectal dose limit V40Gy <30%, V50Gy <20% to prevent radiation proctitis from causing severe diarrhea and bleeding; bladder dose limit V50Gy <40%, V60Gy <25% to protect bladder function, and fine-tune the dose according to the relationship between organ function reserve and tumor location. For example, when the tumor is close to the posterior wall of the bladder, the anterior wall dose can be moderately increased by 3%-5% to ensure coverage, and the posterior wall dose can be accurately controlled according to bladder tolerance, and the dose distribution is dynamically optimized according to organ response.

[0042] Endometrial cancer target area dose range: The PTV dose standard for stage III endometrial cancer is 60-65Gy. In view of the tumor stage, differentiation and MSI-H characteristics, the dose intensity is moderately increased and set to 63-66Gy, taking into account the potential opportunity selection of tumor control and immune combined intervention. Adjacent organ dose control: Small intestine dose limit V40Gy <25%, V50Gy <15%, to maintain intestinal function and reduce the risk of intestinal obstruction; bladder dose limit V50Gy <40%, V60Gy <25% (partially coordinated with cervical cancer bladder limit optimization), adjust the dose according to the changes in uterine morphology and tumor shrinkage, such as adjusting the dose distribution according to the bladder filling degree when the uterus shrinks, and optimizing the dose distribution under the premise of maintaining bladder function to ensure sufficient tumor dose and normal tissue safety.

[0043] Hypertension and diabetes increase the risk of cardiovascular disease and microvascular disease. Blood pressure, blood sugar fluctuations and cardiac function indicators (such as changes in the ST-T segment of the electrocardiogram and EF value of cardiac ultrasound) should be closely monitored during radiotherapy. If blood pressure exceeds 140 / 90 mmHg for 3 consecutive days or blood sugar fluctuates greatly (blood sugar >11.1mmol / L or <3.9mmol / L more than 2 times 2 hours after a meal), the PTV dose should be reduced by 5%-8% or the fractionation mode should be adjusted (single dose reduced by 10%, and the number of fractions increased by 10%-15%) according to cardiac function and overall tolerance to reduce the risk of cardiovascular complications. Anemia limits tissue oxygen supply. According to the oxygen enhancement ratio theory and tumor hypoxia detection results, if the proportion of tumor hypoxia area exceeds 20%, oxygen therapy should be used to moderately increase the tumor hypoxia area dose (5%-8%) and the overall dose should be fine-tuned according to tolerance to improve radiotherapy sensitivity and ensure intervention effect. Decreased cardiopulmonary function limits radiotherapy dose intensity and respiratory management strategies. Based on lung function indicators (FEV1, DLCO) and cardiac reserve function, optimize the respiratory gating parameters of the radiotherapy plan and reduce the proportion of lung volume in the high-dose area (such as V20Gy < 25%). If the user's dyspnea worsens during radiotherapy (NYHA heart function grade increases by one level), suspend radiotherapy to evaluate and adjust the dose (reduce the dose by 5%-10% or optimize the fractionation) to ensure user tolerance and reduce respiratory and cardiac complications. Continuously optimize the dose limiting factor based on the dynamic changes of physiological indicators to achieve accurate, safe and efficient radiotherapy for multiple cancer types.

[0044] In another implementation, accurate determination of the radiotherapy dose is crucial to improving the effectiveness of pelvic tumor intervention and reducing side effects, and requires comprehensive consideration of multiple factors.

[0045] Physiological state analysis and feature extraction for cervical cancer: A 48-year-old female user suffered from stage ⅡB cervical cancer, squamous cell carcinoma. In terms of physiological state, she had mild anemia (Hb105g / L), which affected tissue oxygen supply and radiotherapy sensitivity; her body mass index was 24, and her body fat was evenly distributed; she had no other chronic diseases and her physical function was acceptable. Pathological type: squamous cell carcinoma, with unique cell morphology and invasion pattern, mostly exogenous growth, easy to invade cervical stroma in the early stage, and may involve paracervical tissue, vagina and pelvic wall in the late stage. Genetic testing showed PIK3CA gene mutation, which affects cell signaling pathways, tumor proliferation, and radiotherapy response. The preset dose interval was initially determined: According to the staging and guidelines, the radical dose of the cervical tumor target volume (PTV) is 80-85Gy. Because stage ⅡB tumors involve paracervical tissues, the paracervical lymphatic drainage area needs to be covered, and the PTV dose is preset in this interval. Considering the size of the tumor (about 4 cm in diameter) and location (involving the anterior lip of the cervix and some paracervical tissues), the dose of organs at risk is preset: the rectal dose limit is V40Gy <30%, V50Gy <20%; the bladder dose limit is V50Gy <40%, V60Gy <25%; the small intestine dose limit is V40Gy <20% to prevent radiotherapy damage from causing organ dysfunction. Dose limiting factor generation: user age and anemia are key factors. At the age of 48, the organ function is normal but anemia affects oxygen supply. According to the oxygen enhancement ratio theory, the dose is appropriately reduced or the fractionation scheme is adjusted (such as moderately increasing the number of fractions and reducing the single dose) to improve the tumor oxygenation and enhance the radiotherapy response. The dose limiting factor is adjusted accordingly, such as reducing the overall dose by 3%-5%, and dynamically optimizing the dose according to tumor regression and user tolerance. In terms of genetic characteristics, PIK3CA mutant tumor cells are sensitive to PI3K-Akt pathway inhibitors combined with radiotherapy. The radiotherapy dose can be moderately reduced and the targeted drug dose can be increased. The dose limiting factor is synergistically optimized according to clinical research and individual response to achieve the best intervention ratio.

[0046] Physiological state analysis and feature extraction for prostate cancer: 62-year-old male, prostate cancer stage III, adenocarcinoma. Combined with hyperlipidemia, long-term diet control. Weak body, reduced muscle mass, BMI23 but poor body composition. Tumor pathology Gleason score 8 points (4+4), highly invasive; gene detection found AR gene amplification and TP53 mutation, affecting androgen receptor function and cell cycle regulation, radiotherapy resistance. The dose preset interval is initially determined: stage III prostate cancer, the tumor exceeds the prostate capsule, the seminal vesicle is involved, and the PTV dose is preset to 78-82Gy. According to the size of the tumor (about 5cm in diameter, bilateral lobes are involved), location (close to the anterior wall of the rectum and bladder neck), the dose of organs at risk is preset: rectal dose limit V40Gy <40%, V50Gy <25%; bladder dose limit V50Gy <50%, V60Gy <30%; urethral dose limit Dmax <75Gy, to maintain the urination function of the organ, taking into account both tumor control and functional protection. Generation of dose-limiting factors: Age and comorbidities are important. At the age of 62, functional decline and hyperlipidemia increase cardiovascular risks. The cardiac dose of radiotherapy is strictly monitored, and the PTV dose is fine-tuned according to cardiac function indicators (ejection fraction 55%). Weak constitution affects recovery ability, and the fractionation mode is adjusted (such as moderately reducing the single dose and increasing the fractionation interval during large-fraction radiotherapy) and reducing the dose intensity. The dose-limiting factor is adjusted accordingly, such as reducing the overall dose by 5%-8%, and dynamically optimizing the dose and reducing side effects according to the intervention response and physical status. In terms of genetic characteristics, AR gene amplification is treated with anti-androgen intervention combined with radiotherapy, and the dose is adjusted according to the efficacy and toxicity of the drug; TP53 mutation increases the radiotherapy dose while exploring the possibility of combining new targeted drugs or immune interventions. The dose-limiting factor is optimized through multidisciplinary collaboration to improve efficacy and quality of life.

[0047] Physiological state analysis and feature extraction for bladder cancer: 70-year-old male, bladder cancer T3, transitional cell carcinoma. Suffering from chronic bronchitis, mild lung function impairment (FEV170% predicted value); general nutritional status, BMI21; 40-year smoking history, affecting tumor microenvironment and prognosis. Tumor pathology is multifocal, some are papillary and some are invasive; genetic testing shows FGFR3 gene mutation and TP53 mutation, affecting cell proliferation, differentiation and radiotherapy sensitivity. The dose preset range is initially determined: T3 bladder cancer has deep muscle layer invasion, and the PTV dose is preset to 60-66Gy. Tumors vary in size (maximum diameter is about 6 cm) and are located on the side wall of the bladder near the ureteral orifice. The dose of organs at risk is preset as follows: rectal dose limit V40Gy <35%, V50Gy <20%; small intestinal dose limit V40Gy <25%; bilateral ureteral dose limit Dmax <50Gy, to prevent radiotherapy-induced ureteral stenosis, obstruction and intestinal damage, and to preserve organ function. Dose-limiting factor generation: age and cardiopulmonary function are key. At the age of 70, the organ reserve is reduced and the lung function is poor. Radiotherapy respiratory management is challenging. The radiotherapy plan is optimized according to lung function indicators (such as adjusting respiratory gating parameters and avoiding high-dose lung areas). The PTV dose is fine-tuned according to the relationship between respiratory reserve and tumor location. The inflammatory response of chronic bronchitis interferes with the evaluation of radiotherapy efficacy, and the inflammatory indicators and tumor changes need to be strictly monitored to adjust the dose. Nutritional status affects repair, and nutritional support should be strengthened. The dose should be appropriately reduced or the fractionation should be optimized according to nutritional indicators to ensure user tolerance. In terms of genetic characteristics, for FGFR3 mutations, explore the combination of targeted drugs and radiotherapy to reduce the dose; for TP53 mutations, increase the radiotherapy dose and explore the synergy of immune intervention, dynamically optimize the dose limiting factor based on clinical response and genetic changes, and improve the benefits and tolerance of the intervention.

[0048] For rectal cancer, physiological state analysis and feature extraction: 58-year-old female, rectal cancer T4 stage, adenocarcinoma. Combined with type 2 diabetes (stable blood sugar control), mild peripheral neuropathy; BMI26, abdominal obesity; moderate daily activity, good physical endurance. Tumor pathology is poorly differentiated adenocarcinoma, highly invasive and easy to metastasize; genetic testing shows KRAS mutation and BRAF mutation, which affect the efficacy of EGFR targeted intervention and radiotherapy sensitivity. The dose preset interval is initially determined: T4 rectal cancer penetrates the intestinal wall to the surrounding tissue, and the PTV dose is preset to 54-60Gy. According to the size of the tumor (about 5.5cm in diameter) and location (5cm from the anus to the posterior wall of the rectum), the dose of organs at risk is preset: bladder dose limit V50Gy <45%, V60Gy <30%; small intestine dose limit V40Gy <30%; sacral plexus dose limit Dmax <50Gy, protect nerve function, intestinal and bladder function and user quality of life. Dose limiting factor generation: diabetes and obesity are prominent factors. Diabetes affects microvascular lesions, tissue perfusion and repair. Before radiotherapy, optimize blood sugar control, monitor microvascular indicators, and adjust the dose according to blood sugar fluctuations and neuropathy progression. Obesity changes the pelvic anatomical structure and affects the distribution of radiation dose. The dose is accurately assessed based on CT image reconstruction, and the dose is appropriately increased to compensate for the absorption and scattering effects of adipose tissue. The dose limiting factor is dynamically optimized based on changes in blood sugar, body weight, and neurological function. For example, the overall dose is adjusted by ±5% based on fluctuations in the disease. A multidisciplinary team is jointly developed to develop an individualized radiotherapy plan that takes into account tumor control and complication prevention to improve the user's long-term prognosis. In terms of genetic characteristics, KRAS and BRAF mutant tumors are resistant to conventional targeted drugs. New target drugs are explored in combination with radiotherapy. The dose strategy is dynamically optimized based on drug development and clinical evidence to improve the intervention effect and quality of life.

[0049] Physiological state analysis and feature extraction for endometrial cancer: 65-year-old female, endometrial cancer stage III, adenocarcinoma. Suffering from hypertension (stable with drug control) and osteoporosis; BMI22, balanced nutrition; 15 years of menopause, low estrogen level. Tumor pathology is highly differentiated adenocarcinoma, infiltrating more than 1 / 2 of the muscle layer; genetic testing did not show specific driver gene mutations, with microsatellite instability (MSI-H), affecting immune intervention response and radiotherapy sensitivity. The dose preset interval was initially determined: for stage III endometrial cancer, the PTV dose was preset to 60-65Gy. According to the size of the tumor (maximum diameter of about 4.5cm) and location (posterior wall of the uterine body involving parametrial tissue), the dose of organs at risk was preset: bladder dose limit V50Gy<40%, V60Gy<25%; rectal dose limit V40Gy<30%, V50Gy<20%; small intestine dose limit V40Gy<20%, to protect pelvic organ function and optimize the quality of radiotherapy plan. Generation of dose-limiting factors: Age and comorbidities are key. People aged 65 years and above have fragile bones. Before radiotherapy, bone density should be assessed to prevent bone-related complications. Radiotherapy positions and plans should be adjusted according to bone health. Bone protective agents should be used when necessary. Hypertension affects cardiovascular risk. Blood pressure and heart function should be monitored, and PTV dose should be fine-tuned according to cardiac load. Immunointervention of MSI-H phenotype tumors has a good prospect. We should explore the immunotherapy combined with radiotherapy model, adjust the radiotherapy dose and sequential timing according to the immune intervention response (such as changes in tumor markers and immune cell activity), and dynamically optimize the dose-limiting factor according to clinical efficacy and physiological status (blood pressure, bone density, etc.) to enhance the effectiveness and safety of intervention and improve the user's quality of life and survival prognosis.

[0050] S106, processing the physiological parameter information of the target user to generate a dynamic dose influence factor of the target user.

[0051] In one embodiment, the physiological parameter information of the target user is processed to generate the physiological motion parameter information and physiological structure information of the target user. A high-precision chest strap, ECG electrode patch, and fingertip oximeter are used for continuous monitoring. The respiratory rate is 13-16 times / minute, the inspiratory phase is 2-2.5 seconds, the expiratory phase is 2.5-3 seconds, and the tidal volume fluctuates from 400-550ml to generate a respiratory waveform. ECG monitoring shows sinus rhythm, heart rate is 70-80 times / minute, ST segment is occasionally slightly depressed (<0.1mV), T wave morphology is flat, and dynamic ECG signals are collected. The blood oxygen saturation is maintained at 94%-97%, and the dynamic value is recorded for subsequent analysis. A series of pelvic MRI and CT scans are performed before radiotherapy to obtain a morphological and structural information database of organs such as the uterus, cervix, vagina, bladder, and rectum. The uterus is enlarged and irregular in shape, the cervical tumor changes the local anatomy, the bladder wall is slightly thickened due to the adjacent tumor, and the rectum is pushed and displaced. Each organ dynamically changes position and shape depending on the fullness of the bladder and rectum and changes in body posture, forming dynamic physiological structure information, which lays the foundation for accurate planning of radiotherapy doses.

[0052] The target user's physiological motion parameter information is processed to generate dynamic respiratory waveform features, dynamic ECG signal features, and dynamic blood oxygen saturation features. In-depth analysis of respiratory waveforms: Spectral analysis of respiratory waveforms yields a main frequency of 0.2-0.25Hz and weak harmonics, reflecting the stability of respiratory rhythm and potential minor interference. Wavelet transform accurately captures the turning points and slope changes at the end of inspiration and expiration to locate the respiratory displacement of the tumor. For example, at the end of inspiration, the cervical tumor displaces 0.3-0.4cm toward the head, and the endometrial cancer displaces 0.2-0.3cm toward the ventral side, which are key parameters for dose compensation in respiratory gated radiotherapy. A dynamic dose compensation scheme is designed based on the displacement amount and direction. At the end of inspiration, the dose of the anterior wall of the cervix and the anterior wall of the uterus is appropriately reduced (5%-8%), and the dose of the posterior wall is increased (8%-10%) to ensure tumor coverage and reduce normal tissue damage.

[0053] ECG signal feature extraction: Analysis of the RR interval standard deviation of 20-25ms indicates a mild imbalance in autonomic nervous function. Although the overall heart rate is stable, occasional premature beats (1-2 times / minute) and ST-T segment changes indicate potential myocardial ischemia vulnerability. According to the frequency period of arrhythmia, a cardiac dose risk model is constructed. During periods of dense premature beats or ST segment depression, the dose around the heart is reduced by 3%-5%, and the radiation angle is optimized to avoid high-dose areas of the myocardium, protect heart function, and ensure the safety and continuity of radiotherapy, so as to integrate dynamic dose factors into cardiac safety considerations.

[0054] Dynamic analysis of blood oxygen saturation: Analyze the time series of blood oxygen saturation, with a standard deviation of 0.5%-1% and a coefficient of variation of 0.6%-1.1%, reflecting the matching degree between the dynamic reserve of tissue oxygenation and metabolic demand. According to the characteristics of the oxygen dissociation curve and tumor hypoxia imaging, when the tumor hypoxia area accounts for 15%-20%, the dose of the tumor hypoxia area is increased by 5%-8% while the radiotherapy plan is oxygen therapy, and the dynamic optimization is based on the blood oxygen response to improve the sensitivity of radiotherapy, increase the probability of tumor control, and accurately convert blood oxygen information into a key factor for dose adjustment.

[0055] The physiological structure information of the target user is processed to generate dynamic physiological structure information. When the bladder is filled from 100ml to 400ml, the uterus and the tumor move backward by 1-2cm and move 0.5-1cm to the head. The cervical tumor compresses the bladder wall, causing local depression and deformation. The change in rectal filling causes the anterior-posterior diameter of the pelvic organs to change by 1-2cm, and the relative position of the organs to twist by 5°-10°. The uneven infiltration of the uterine muscle layer by cancer causes abnormal contraction and relaxation function, affecting the blood supply and position stability of the tumor, providing an anatomical basis for the dynamic adjustment of radiotherapy dose distribution, such as setting multiple dose plans according to the degree of bladder filling, and selecting the best one according to real-time image guidance to ensure the accuracy of tumor dose and organ function.

[0056] The dynamic respiratory waveform characteristics, dynamic ECG signal characteristics and dynamic blood oxygen saturation characteristics are processed to generate characteristic parameter information of different respiratory cycles. Respiratory monitoring shows that the daily fluctuation of respiratory rate is 12-16 times / minute, and the standard deviation of the inspiratory phase duration is 0.2 seconds, indicating weak rhythm stability. At the end of inspiration, the average displacement of cervical tumor to the head side is 0.35cm, and the displacement of endometrial cancer to the ventral side is 0.28cm, with a standard deviation of 0.05cm, which are key parameters for respiratory gating. The ECG signal shows that the daily variability of heart rate is 5-10 times / minute, the ST segment depression of 0.05-0.1mV accounts for 10%-15% of the recording period, and the T wave flat period accounts for 20%-30%, which is a warning of cardiac dose risk. The coefficient of variation of blood oxygen saturation was low in the morning and high in the afternoon (0.8%-1.2%), which was negatively correlated with the proportion of tumor hypoxic areas (18% in the morning and 12% in the afternoon). It served as the basis for correcting the dose-oxygen effect, and constructed a multi-parameter respiratory cycle characteristic matrix. The dose distribution was adjusted according to the phase of the cycle. At the end of inspiration, the anterior wall dose was reduced by 6%-8%, and the posterior wall dose was increased by 8%-10%, thereby protecting normal tissues and optimizing tumor doses.

[0057] The characteristic parameter information and dynamic physiological structure information of different respiratory cycles are processed to generate dynamic dose influencing factors for the target user. The dynamic physiological structure and respiratory cycle characteristics are integrated to construct a dynamic dose influencing factor model. Taking the third week of radiotherapy as an example, when the bladder is half-full, the respiratory rate is 14 times / minute, and the ST segment is depressed by 0.08mV at the end of inspiration, the model reduces the dose near the heart by 4% and increases the dose to the posterior wall of the bladder by 7% (the tumor is close to the posterior wall) in real time according to the algorithm, and dynamically optimizes the incident angle of the ray according to the displacement and rotation of the organ (such as adjusting the incident angle by 3°-5° according to 8° uterine torsion), ensuring accurate conformal dose of the tumor and safe limit of normal tissue, and continuously optimizing the radiotherapy dose according to the real-time evolution of the user's physiological and pathological state, improving the quality of radiotherapy for multiple cancers and the user's prognosis, and realizing accurate, dynamic, safe and efficient radiotherapy dose planning.

[0058] S107, processing the target feature vector, the dose limitation factor of the target user, and the dynamic dose influence factor of the target user based on the target dose distribution processing model to generate target dose distribution information.

[0059] In one embodiment, The target feature vector is processed based on the target dose distribution processing model to generate tumor feature vector encoding information. Features are extracted from the user's multimodal images and clinical data, and processed into high-dimensional tumor feature vector encoding by the target dose distribution processing model. For example, for cervical cancer, the model focuses on tumor morphology (irregular shape parameter encoding), location (quantitative encoding of millimeter distance from the bladder and rectum), pathological type (squamous cell carcinoma unique texture feature encoding), gene characteristics (PIK3CA gene mutation site and expression level encoding) and surrounding tissue relationship (coding of adhesion degree with cervical stroma and blood vessels); endometrial cancer also integrates tumor size (volume change encoding), invasion depth (ratio of uterine myometrial invasion layer encoding), pathological grade (moderately differentiated cell atypia encoding), microsatellite instability state (MSI-H related molecular marker encoding) and uterine morphological changes (uterine cavity volume, uneven myometrial thickness encoding), forming encoding information that accurately reflects tumor characteristics, providing a core basis for subsequent dose calculations, and helping to distinguish differences in tumor characteristics to determine the basic direction of dose strategies.

[0060] Based on the target dose distribution processing model, the dose limiting factor and tumor feature vector encoding information of the target user are processed to generate a tumor dose warning value. The dose limiting factor covers multiple factors of the user's physiological state. For example, hypertension causes increased cardiovascular risk. Based on the heart function reserve and blood pressure fluctuations, a safe upper limit coefficient of the peripheral cardiac dose is set for pelvic tumors; diabetes causes microvascular lesions and nerve damage. Based on nerve conduction velocity and blood sugar control level, the pelvic organ dose adjustment coefficient is determined to prevent radiotherapy damage from aggravating the condition. Combined with the tumor feature vector, if the cervical cancer tumor is close to the bladder and proliferates actively (high metabolic texture features), based on the bladder function reserve and tumor invasion risk, the model accurately calculates the high dose warning value to ensure tumor control while preventing severe bladder damage (such as bladder dose exceeding 60Gy warning); if the endometrial cancer tumor has deep myometrial infiltration and MSI-H may affect the immune microenvironment radiotherapy response, the dose warning is calculated based on the function of the uterus and adjacent organs (such as myometrial dose exceeding 65Gy warning), and a key safety threshold is set for radiotherapy dose planning to prevent excessive intervention from causing serious complications, protect the user's quality of life and the possibility of subsequent intervention.

[0061] Based on the target dose distribution processing model, the dynamic dose influencing factors of the target user and the tumor feature vector encoding information are processed to generate dynamic dose values. Among the physiological motion parameters, the respiratory cycle has a significant impact. The displacement of the cervical tumor at the end of inspiration causes the dose distribution to change. The model adjusts the dose distribution in real time according to the respiratory frequency, depth, tumor displacement and direction. The dose compensation for the tumor displacement of 0.3-0.4cm is 5%-8%; during the period of abnormal ECG signal (premature beats or ST segment depression), the dose of tumors around the heart is dynamically reduced by 3%-5% to protect heart function; blood oxygen saturation fluctuations reflect changes in tissue oxygenation and tumor hypoxia. The dose in the hypoxic area (based on functional imaging positioning) is dynamically increased by 5%-8% to enhance the radiotherapy effect. The dose is continuously optimized according to the real-time fluctuations of physiological parameters to achieve dynamic and precise radiotherapy, improve tumor control, reduce normal tissue damage, and provide quantitative support for the flexible adjustment of radiotherapy dose according to the user's immediate physiological and pathological state.

[0062] Based on the target dose distribution processing model, the tumor dose warning value and dynamic dose value are processed to generate target dose distribution information, which is used to characterize the dose distribution report corresponding to different cancer information. The report is a three-dimensional distribution map and numerical table of the dose of the target area and organs at risk for cervical cancer and endometrial cancer. The dose of each target area is accurately planned according to the tumor characteristics, physiological state, and physiological movement. For example, the dose of the target area of ​​cervical cancer tumor is accurately distributed according to the tumor stage, genetic characteristics and physiological restrictions of 82-85Gy, and the dose of the surrounding organs at risk is dynamically adjusted according to the distance and function; the dose of endometrial cancer is reasonably distributed according to the stage, differentiation and physiological condition, and the dose of the myometrium tumor is 63-66Gy. The dose of the adjacent bladder and intestine is strictly controlled according to the dynamic position relationship of the organs, providing accurate navigation for the implementation of radiotherapy plans (such as intensity modulated radiotherapy field design and dose rate modulation), ensuring that each radiotherapy dose is accurately irradiated to the tumor and protecting the normal tissue function, improving the radiotherapy efficacy, safety and quality of life of users with multiple cancer types, and realizing personalized, dynamic and precise dose planning and intervention implementation. Full process optimization.

[0063] In another embodiment, the method further includes a calculation formula for obtaining a tumor dose warning value, the calculation formula being: in, Represents the basic dose reference value, represents the weight coefficient, represents the first tumor characteristic function, Represents the tumor pathological type, Represents the degree of differentiation, Represents gene expression characteristics, represents the first user's physiological function, Represents the user's age, Represents the user's gender, Represents the user's body mass index, represents the first organ association function, Represents the distance between the tumor and the target organ, Represents the reserve elements of tumor and organ function, represents the second tumor characteristic function, Represents the clinical stage information of the tumor, Represents the size of the tumor, Represents the anatomical location, represents the second user's physiological function, Represents user comorbidity information, Represents the user's medical history factors, represents the second organ association function, Represents the volume of adjacent organs, Represents radiosensitivity, stands for Regulation Index.

[0064] Take a 60-year-old female user with cervical cancer and endometrial cancer as an example: Assuming a basic dose constant Adjustment Index .

[0065] First tumor characteristic function Tumor pathological type :Cervical cancer is squamous cell carcinoma, the degree of differentiation :Medium differentiation, gene expression signature :There is a PIK3CA gene mutation. Based on the above characteristics, it is assumed that .

[0066] Second tumor characteristic function Clinical stage of tumor : Stage, tumor size :Assuming the tumor size is 4cm×3cm, the anatomical location : Invasion of the upper 1 / 3 of the vagina. Based on the above characteristics, it is assumed that .

[0067] First user physiological function Middle, age : 60 years old, gender :Female, body mass index : 25, the user has high blood pressure and mild diabetes. Based on these characteristics, it is assumed that .

[0068] Second user physiological function Comorbidity information : Hypertension and mild diabetes, past medical history factors : No other major medical history. Based on the above characteristics, it is assumed that .

[0069] First organ association function The distance between the tumor and the target organ :Assuming that the tumor is close to the bladder, the reserve factors of tumor and organ function :The bladder function is normal but affected by the tumor. Based on the above characteristics, it is assumed that .

[0070] First organ association function The volume of adjacent organs : Normal bladder volume, radiosensitivity :The bladder is moderately sensitive to radiotherapy. Based on the above characteristics, it is assumed that .

[0071] Calculate tumor dose warning value , substitute the above assumed values ​​into the formula: .

[0072] The method also includes a calculation formula for obtaining a dynamic dose value, the calculation formula is:

[0073] Among them, n represents the number of factor categories, represents the weight coefficient, represents the group of motion characteristic functions, Represent the characteristic parameters of the respiratory motion cycle and the amplitude change rate, Represent the cardiac rhythm characteristics and the rate of change of cardiac chamber volume, Represents the organ characteristic function group, represents the organ mobility parameter, represents the organ mobility parameter, Represents the heterogeneity of organ tissues. represents the physiological regulation coefficient, Represents the user's physiological state parameter vector, Represents the exponential adjustment factor.

[0074] For the 60-year-old female user with cervical cancer and endometrial cancer: Factor category number n: Assume n=3, consider respiratory, cardiac and tissue oxygenation factors respectively. Weight coefficient :Set the weight of breathing factor Weighting of cardiac factors Set the weight of tissue oxygenation factor Kinematic characteristic function group , respiratory motion characteristics (i=1), respiratory frequency Taking the average . Tidal volume fluctuation amplitude change rate .but , assuming that based on these values .

[0075] Heart motion characteristics (i=2), heart rate (heart motion rhythm characteristics ) is 70-80 times / minute, taking the average value . Heart chamber volume change rate (Assuming that the volume of the heart chamber changes slightly), .but . Assume that based on these values .

[0076] Organ characteristic function group : Bladder organ characteristics (taking the bladder as an example, i=1): Bladder function is normal but is affected by the tumor , assuming .

[0077] Low bladder mobility , assuming .

[0078] Moderate tissue heterogeneity , assuming .

[0079] but , assuming that based on these values .

[0080] Physiological regulation coefficient and index adjustment factor : Respiratory factor (i=1): blood oxygen saturation 94%-97%, the average value is 95.5%, set Related to blood oxygen saturation, assuming .

[0081] Cardiac factors (i=2): Considering cardiac function, assuming (If, related to myocardial ischemia), then .

[0082] Calculate dynamic dose values :

[0083] Substituting the above values: For the breathing factor (i=1): .

[0084] For cardiac factors (i=2): .

[0085] Assume that the tissue oxygenation factor (i=3) is similarly calculated to be 0.3 (the specific calculation process is omitted).

[0086] but .

[0087] This dynamic dose value can be used to adjust the radiotherapy dose according to the user's real-time physiological state to ensure that the tumor can be effectively intervened while minimizing damage to normal tissue during the intervention process.

[0088] This solution uses a meticulous pre-processing training sample set to group samples by cancer type, stage, age, etc., to explore common features within the group and differences between groups. Advanced algorithms are used to extract image texture, shape, density, and physiological parameter-related features from the grouped data into the original feature library, and then the training and validation sets are scientifically divided. The validation set is predicted by a deep neural network classifier, and key features are selected into target feature data based on the predicted training results. This is used to train a dose distribution processing model based on an innovative architecture (such as SwinTransformer optimized with an attention mechanism). After repeated iterative parameter adjustments, an accurate target dose distribution processing model is obtained, which can efficiently handle complex pelvic tumor dose prediction tasks and ensure stable and accurate output of dose distribution in different scenarios.

[0089] When processing the target user's CT image, the image is enhanced by adaptive filtering, histogram equalization, etc., and normalized to the standard range; dynamic segmentation accurately outlines the tumor contour according to the tumor image characteristics and prior knowledge to obtain the initial regional information; in-depth analysis of regional texture (such as grayscale co-occurrence matrix multi-parameter analysis), shape (multi-dimensional quantitative indicators), grayscale (multi-statistical features) features and fusion, to generate accurate and comprehensive target feature vectors, providing core imaging basis for dose prediction. Analyze the physiological state to generate cancer characteristics and physiological indicator information, follow clinical guidelines, big data analysis and physiological and pathological associations to determine the target area dose interval, and correct the dose limiting factor based on comprehensive physiological factors to ensure safe and effective doses.

[0090] Mining physiological parameters to obtain physiological movement (respiratory rhythm frequency, ECG waveform rhythm variability, blood oxygen fluctuation trend) and structure (organ morphology dynamic changes, tissue density distribution) information, generating respiratory cycle multi-parameters (such as different phases of tumor displacement speed direction, cardiopulmonary function change rate) and dynamic physiological structure characteristics through time-frequency analysis and feature modeling, and merging the two to construct dynamic dose influencing factors to quantify the real-time dose impact of physiological factors. Based on the target model, the target vector, limiting factors, and influencing factors are integrated, and coding information, warning values, and dynamic values ​​are generated based on deep learning and clinical rules. The target dose distribution report is accurately constructed to customize personalized, dynamic, and precise dose plans for pelvic tumor radiotherapy, improve intervention benefits and user quality of life, and promote the process of precision radiotherapy.

[0091] In one embodiment, if Figure 2As shown, the present application also provides a multi-center, multi-tumor, and multi-prescription dose prediction device for pelvic tumors, including: An acquisition module 201 is used to acquire CT image information of a target user, physiological state information of the target user, physiological parameter information of the target user, a training sample set, and a preset dose distribution processing model based on a SwinTransformer network architecture, wherein the CT image information of the target user includes CT image information of different cancer information of the target user, the physiological state information of the target user includes age, gender, physical condition, and specific characteristics of the tumor of the target user, and the training sample set includes CT image information of different users with different cancer information; The processing module 202 is used to pre-process the training sample set to generate a training sample set with target feature data; based on the training sample set with target feature data, process the preset dose distribution processing model based on the SwinTransformer network architecture to generate a target dose distribution processing model; perform feature extraction processing on the CT image information of the target user to generate a target feature vector; process the physiological state information of the target user to generate a dose limitation factor of the target user; process the physiological parameter information of the target user to generate a dynamic dose influence factor of the target user; based on the target dose distribution processing model, process the target feature vector, the dose limitation factor of the target user and the dynamic dose influence factor of the target user to generate target dose distribution information, wherein the target dose distribution information is used to characterize the dose distribution report corresponding to different cancer information.

[0092] Each embodiment in the present application is described in a related manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the dose prediction method, electronic device, electronic device, and readable storage medium embodiment for evaluating multi-center, multi-tumor, and multi-prescription doses of pelvic tumors, since they are basically similar to the above-mentioned dose prediction method embodiment for multi-center, multi-tumor, and multi-prescription doses of pelvic tumors, the relevant parts can be referred to the partial description of the above-mentioned dose prediction method embodiment for multi-center, multi-tumor, and multi-prescription doses of pelvic tumors.

Claims

1. A multi-center, multi-tumor, and multi-prescription dose prediction method for pelvic tumors, characterized in that: include: Acquire CT image information of a target user, physiological state information of the target user, physiological parameter information of the target user, a training sample set, and a preset dose distribution processing model based on a SwinTransformer network architecture, wherein the CT image information of the target user includes CT image information of different cancer information of the target user, the physiological state information of the target user includes age, gender, physical condition, and specific characteristics of the tumor of the target user, and the training sample set includes CT image information of different users with different cancer information; Preprocess the training sample set to generate a training sample set with target feature data; Based on the training sample set with target feature data, the preset dose distribution processing model based on the SwinTransformer network architecture is processed to generate a target dose distribution processing model; Perform feature extraction processing on the target user's CT image information to generate a target feature vector; Processing the physiological status information of the target user to generate a dose limiting factor for the target user; Processing the physiological parameter information of the target user to generate the dynamic dose influence factor of the target user; The target characteristic vector, the dose limitation factor of the target user and the dynamic dose influencing factor of the target user are processed based on the target dose distribution processing model to generate target dose distribution information, wherein the target dose distribution information is used to characterize the dose distribution report corresponding to different cancer information.

2. The method according to claim 1, characterized in that Preprocess the training sample set to generate a training sample set with target feature data, including: The training data set is grouped to generate a grouped training data set, wherein the grouped training data set includes physiological characteristic information of users of different age groups and different cancer information; Extract features from the grouped training data set to generate an original feature library; Divide the original feature library into various feature data sets to generate training sets and validation sets; Based on the classifier, each validation set divided by the original feature library is predicted to generate a prediction result; Use the preset algorithm to divide each training set in the original feature library for training and generate the prediction results of the validation set class; The prediction results and the validation set prediction results are processed to generate target feature data, which are used to characterize the image features and physiological parameters that affect the user's physiological abnormalities.

3. The method according to claim 1, characterized in that Perform feature extraction on the target user's CT image information to generate a target feature vector, including: Perform image enhancement and normalization processing on the target user's CT image to generate a target CT image; Perform dynamic segmentation processing on the target CT image to generate some initial tumor region information; Performing texture feature analysis on a number of initial tumor region information respectively to generate tumor texture information; Performing shape feature quantification processing on several initial tumor region information respectively to generate tumor morphology information; Grayscale processing is performed on several initial tumor region information respectively to generate tumor grayscale feature information; The tumor texture information, tumor morphology information and tumor grayscale feature information are fused to generate target feature information.

4. The method according to claim 1, characterized in that The target user's physiological status information is processed to generate the target user's dose limiting factor, including: Processing the physiological status information of the target user to generate cancer type characteristic information and physiological indicator information of the target user; Process the target user's cancer type characteristic information to generate a preset dose interval value corresponding to each target area information; The dose preset interval value corresponding to each target area information is processed based on the target user's physiological index information to generate a dose limitation factor for the target user.

5. The method according to claim 1, characterized in that The physiological parameter information of the target user is processed to generate the dynamic dose influencing factors of the target user, including: Processing the physiological parameter information of the target user to generate physiological motion parameter information of the target user and physiological structure information of the target user; Process the target user's physiological motion parameter information to generate dynamic respiratory waveform features, dynamic ECG signal features, and dynamic blood oxygen saturation features; Processing the physiological structure information of the target user to generate dynamic physiological structure information; Processing dynamic respiratory waveform features, dynamic ECG signal features and dynamic blood oxygen saturation features to generate characteristic parameter information of different respiratory cycles; The characteristic parameter information and dynamic physiological structure information of different respiratory cycles are processed to generate the dynamic dose impact factor of the target user.

6. The method according to claim 5, characterized in that Based on the target dose distribution processing model, the target feature vector, the dose limitation factor of the target user and the dynamic dose influence factor of the target user are processed to generate target dose distribution information, including: Processing the target feature vector based on the target dose distribution processing model to generate tumor feature vector encoding information; Based on the target dose distribution processing model, the dose limitation factor and tumor feature vector encoding information of the target user are processed to generate a tumor dose warning value; Based on the target dose distribution processing model, the dynamic dose influencing factors of the target user and the tumor feature vector encoding information are processed to generate a dynamic dose value; The tumor dose warning value and dynamic dose value are processed based on the target dose distribution processing model to generate target dose distribution information.

7. The method according to claim 6, characterized in that The target characteristic vector, the dose limitation factor of the target user and the dynamic dose influence factor of the target user are processed based on the target dose distribution processing model to generate target dose distribution information, which also includes: The method also includes a calculation formula for obtaining a tumor dose warning value, the calculation formula is: ； in, Represents the basic dose reference value, represents the weight coefficient, represents the first tumor characteristic function, Represents the tumor pathological type, Represents the degree of differentiation, Represents gene expression characteristics, represents the first user's physiological function, Represents the user's age, Represents the user's gender, Represents the user's body mass index, represents the first organ association function, Represents the distance between the tumor and the target organ, Represents the reserve elements of tumor and organ function, represents the second tumor characteristic function, Represents the clinical stage information of the tumor, Represents the size of the tumor, Represents the anatomical location, represents the second user's physiological function, Represents user comorbidity information, Represents the user's medical history factors, represents the second organ association function, Represents the volume of adjacent organs, Represents radiosensitivity, stands for regulation index; The method also includes a calculation formula for obtaining a dynamic dose value, the calculation formula is: ； Among them, n represents the number of factor categories, represents the weight coefficient, represents the group of motion characteristic functions, Represent the characteristic parameters of the respiratory motion cycle and the amplitude change rate, Represent the cardiac rhythm characteristics and the rate of change of cardiac chamber volume, Represents the organ characteristic function group, represents the organ mobility parameter, represents the organ mobility parameter, Represents the heterogeneity of organ tissues. represents the physiological regulation coefficient, Represents the user's physiological state parameter vector, Represents the exponential adjustment factor.

8. A multi-center, multi-tumor, and multi-prescription dose prediction device for pelvic tumors, characterized in that: The device includes: an acquisition module, used to acquire CT image information of a target user, physiological state information of the target user, physiological parameter information of the target user, a training sample set, and a preset dose distribution processing model based on a SwinTransformer network architecture, wherein the CT image information of the target user includes CT image information of different cancer information of the target user, the physiological state information of the target user includes age, gender, physical condition, and specific characteristics of the tumor of the target user, and the training sample set includes CT image information of different users with different cancer information; A processing module is used to preprocess a training sample set to generate a training sample set with target feature data; based on the training sample set with target feature data, a preset dose distribution processing model based on the SwinTransformer network architecture is processed to generate a target dose distribution processing model; feature extraction processing is performed on the CT image information of the target user to generate a target feature vector; the physiological state information of the target user is processed to generate a dose limitation factor of the target user; the physiological parameter information of the target user is processed to generate a dynamic dose influence factor of the target user; based on the target dose distribution processing model, the target feature vector, the dose limitation factor of the target user and the dynamic dose influence factor of the target user are processed to generate target dose distribution information, wherein the target dose distribution information is used to characterize the dose distribution report corresponding to different cancer information.

9. An electronic device, characterized in that: include: a first processor; and a memory for storing executable instructions for the first processor; The first processor is configured to execute the multi-center, multi-tumor, and multi-prescription dose prediction method for pelvic tumors according to any one of claims 1 to 7 by executing executable instructions.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the second processor, the dose prediction method for multi-center, multi-tumor and multi-prescription pelvic tumors according to any one of claims 1 to 7 is implemented.

Images