Automatic segmentation and intelligent plan generation method and device suitable for adaptive radiotherapy of prostate cancer

Abstract

The application relates to the computer technical field and discloses a kind of automatic segmentation and intelligent plan generation method and device suitable for prostate cancer adaptive radiotherapy, comprising: obtaining medical image data of the subject and including the target region and the clinical dose data of the lesion area of the subject and the clinical treatment data of the preset priority level of organ at risk;Automatic segmentation is carried out based on medical image data to obtain target segmentation data;According to target segmentation data, initial dose distribution data is obtained by dose distribution prediction, and it is iteratively optimized using clinical treatment data, to obtain the target dose distribution data of the subject;Based on target dose distribution data, the iterative optimization of radiotherapy plan is carried out, and the target radiotherapy plan of the subject is generated. Thus, on the basis of realizing the accurate and efficient automatic segmentation of prostate cancer target region and organ at risk, multiple rounds of iterative optimization of dose distribution and radiotherapy plan are carried out, and finally a high-quality target radiotherapy plan is obtained.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to an automatic segmentation and intelligent planning method and apparatus suitable for adaptive radiotherapy of prostate cancer. Background Technology

[0002] Globally, prostate cancer is one of the most common malignant tumors in men, and its incidence is rising year by year. Radiotherapy is the main treatment for prostate cancer and is used throughout the entire treatment process. However, existing methods for automatically generating radiotherapy plans have limitations, necessitating an intelligent plan generation method that can improve the quality of prostate cancer radiotherapy plans. Summary of the Invention

[0003] This application aims to at least partially solve one of the technical problems in related technologies. To this end, this application proposes an automated segmentation and intelligent planning method and apparatus suitable for adaptive radiotherapy of prostate cancer. The main technical solutions adopted in this application include:

[0004] In a first aspect, embodiments of this application provide an automatic segmentation and intelligent plan generation method suitable for adaptive radiotherapy of prostate cancer. The method includes: acquiring medical imaging data and clinical treatment data of the patient; wherein the medical imaging data is obtained from imaging examinations of the pelvic region of the patient; the clinical treatment data includes clinical dose data and preset priority levels for the target area and organs at risk within the lesion region of the patient; performing automatic segmentation based on the medical imaging data to obtain target segmentation data; wherein the target segmentation data reflects the distribution of the target area and organs at risk within the lesion region of the patient; predicting the dose distribution based on the target segmentation data to obtain initial dose distribution data; iteratively optimizing the initial dose distribution data using the clinical dose data and preset priority levels to obtain target dose distribution data for the patient; and iteratively optimizing the radiotherapy plan based on the target dose distribution data to generate a target radiotherapy plan for the patient; wherein the target radiotherapy plan describes the target dosimetric data for the target area and organs at risk within the lesion region.

[0005] Secondly, embodiments of this application provide an automatic segmentation and intelligent plan generation device suitable for adaptive radiotherapy of prostate cancer. The device includes: a data acquisition module for acquiring medical imaging data and clinical treatment data of the treated subject; wherein the medical imaging data is obtained from imaging examinations of the pelvic region of the treated subject; the clinical treatment data includes clinical dose data and preset priority levels for the target area and organs at risk in the lesion region of the treated subject; a delineation and segmentation module for automatically segmenting based on the medical imaging data to obtain target segmentation data; wherein the target segmentation data reflects the distribution of the target area and organs at risk in the lesion region of the treated subject; a dose prediction module for predicting the dose distribution based on the target segmentation data to obtain initial dose distribution data; a dose optimization module for iteratively optimizing the initial dose distribution data using the clinical dose data and preset priority levels to obtain target dose distribution data for the treated subject; and a plan generation module for iteratively optimizing the radiotherapy plan based on the target dose distribution data to generate a target radiotherapy plan for the treated subject; wherein the target radiotherapy plan describes the target dosimetric data of the target area and organs at risk in the lesion region.

[0006] Thirdly, this application also provides a computer device including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of any of the above methods.

[0007] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of any of the methods described above.

[0008] Fifthly, the present invention provides a computer program product, including a computer program that, when executed by a processor, implements the steps of any of the above methods.

[0009] In the above embodiments, based on the medical imaging data of the treated patient, automatic segmentation is performed to obtain target segmentation data reflecting the distribution of the target area and organs at risk within the lesion region of the treated patient. Then, dose distribution prediction is performed based on this data to obtain initial dose distribution data. This initial data is then iteratively optimized by combining clinical dose data and clinical treatment data to derive target dose distribution data, ultimately generating a target radiotherapy plan. Thus, by achieving accurate and efficient automatic segmentation of the prostate cancer target area and organs at risk, and through multiple rounds of iterative optimization of dose distribution and radiotherapy planning, a high-quality target radiotherapy plan is ultimately obtained. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the specific embodiments of this application or the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0011] Figure 1 This is a flowchart of an automated segmentation and intelligent planning method suitable for adaptive radiotherapy of prostate cancer, according to an embodiment of this application;

[0012] Figure 2 This is a flowchart of a method for determining target dose distribution data according to an embodiment of this application;

[0013] Figure 3 This is a flowchart of a method for determining intermediate dose distribution data according to an embodiment of this application;

[0014] Figure 4a This is a flowchart of a method for determining intermediate dose distribution data according to yet another embodiment of this application;

[0015] Figure 4b This is a schematic diagram of out-of-target dose drop according to yet another embodiment of this application;

[0016] Figure 5 This is a flowchart of a method for determining a target radiotherapy plan according to an embodiment of this application;

[0017] Figure 6 This is a flowchart of a method for determining a target radiotherapy plan according to another embodiment of this application;

[0018] Figure 7 This is a comparative schematic diagram of segmentation results provided according to another embodiment of this application;

[0019] Figure 8 This is a structural block diagram of an automatic segmentation and intelligent planning device suitable for adaptive radiotherapy of prostate cancer, according to an embodiment of this application;

[0020] Figure 9 This is an internal structural diagram of a computer device according to an embodiment of the present application. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0022] Prostate cancer is a common malignant tumor in men. Data shows that in 2024, prostate cancer accounted for 29% of newly diagnosed cancers in men in the United States. Radiation therapy is the primary treatment for prostate cancer. It's important to understand several key concepts in the field of radiotherapy. Gross tumor volume (GTV) refers to the tumor visible on imaging; clinical target volume (CTV) is the GTV plus any subclinical lesions and potential infiltration areas; and planning target volume (PTV) accounts for the irradiation range formed by various random and systematic errors during radiotherapy. A treatment planning system (TPS) is used to develop radiotherapy plans for patients. Through a built-in accelerator model, the system designs the incident angle and optimization function to ensure the tumor receives the prescribed dose while ensuring that surrounding normal organs do not exceed the dose limit. Normal organs other than the tumor are called organs at risk (OARs). Dose-volume histogram (DVH) is a method / tool ​​used to evaluate radiotherapy plans. It can obtain data such as the dose-volume of the prescribed dose to the target area, the high dose, and the dose-volume and average dose of organs at risk. The development of intensity-modulated radiotherapy (IMRT) technology allows tumors to receive high doses of radiation while minimizing the radiation dose to surrounding normal organs. The key to achieving this goal lies in accurately delineating the target area and surrounding normal organs. Prostate cancer patients are typically older (average 67 years old), have longer radiotherapy cycles (1-2 months), and the lesion is located between the bladder and rectum. The varying degrees of rectal and bladder filling during each radiotherapy session can not only lead to tumor off-target effects but also cause the intestines and bladder to receive more radiation, triggering gastrointestinal and urinary reactions, and in severe cases, even interrupting radiotherapy, affecting treatment efficacy. Adaptive radiotherapy (ART) technology can automatically segment and plan the radiotherapy plan online based on the patient's pre-treatment images, ensuring the tumor receives a sufficient dose while reducing radiotherapy reactions.

[0023] In current clinical practice, the delineation of target areas and organs at risk for prostate cancer radiotherapy still faces numerous challenges. For experienced urological radiotherapy physicians, manually segmenting the target area and organs at risk takes approximately three hours, while less experienced physicians struggle to accurately define the extent of these areas. Although artificial intelligence technology offers some assistance in delineating organs at risk, significantly reducing manual delineation time, the difficulty in delineating organs at risk surrounding pelvic tumors, particularly the segmentation of the small intestine and colon, remains unsatisfactory, requiring several tens of minutes for revisions. Furthermore, target area segmentation still primarily relies on manual operation. These factors not only impact the efficiency of radiotherapy physicians but also significantly hinder the implementation of prostate cancer radiotherapy in second- and third-tier hospitals. Simultaneously, prostate cancer radiotherapy planning involves numerous target areas and organs at risk. The positional relationship between the target area and many surrounding organs at risk determines the high complexity of the radiotherapy plan. Therefore, prostate cancer radiotherapy plan design is a time-consuming and human-intensive task, and the plan's quality is easily influenced by the individual experience of the radiotherapy planner.

[0024] Furthermore, while online adaptive radiotherapy can adjust the radiotherapy plan in real time based on the patient's imaging on the day of treatment, the patient remains fixed on the treatment bed during the procedure. Even if the planned target volume (PTV) is expanded by 5mm, organ movement of the prostate gland will intensify over time, potentially requiring a larger expansion of the planned target volume boundary to ensure the treatment does not miss the target. Moreover, the delineation of the target volume and surrounding organs at risk is labor-intensive, manual radiotherapy planning is time-consuming, and plans based on individual experience vary in quality. These are all key technical problems that urgently need to be solved in the clinical application of online adaptive radiotherapy. Therefore, accurate and rapid automatic segmentation of the prostate cancer target volume and organs at risk, and the realization of intelligent prostate cancer planning, are of great significance. Accurate delineation and automatic planning of automatic prostate cancer segmentation can not only improve the work efficiency of radiation oncologists and planners and enhance the quality of radiotherapy plans, but also promote the homogenization of prostate cancer radiotherapy quality in secondary and tertiary hospitals. It can also shorten the time required for online adaptive radiotherapy, facilitating its application and promotion in clinical practice, thereby benefiting more prostate cancer patients.

[0025] Based on this, according to the embodiments of this application, an embodiment of an automatic segmentation and intelligent plan generation method suitable for adaptive radiotherapy of prostate cancer is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0026] This embodiment provides an automatic segmentation and intelligent plan generation method suitable for adaptive radiotherapy of prostate cancer. Figure 1This is a flowchart of an automatic segmentation and intelligent planning method suitable for adaptive radiotherapy of prostate cancer according to an embodiment of this application, such as... Figure 1 As shown, the process includes the following steps:

[0027] S110. Obtain medical imaging data and clinical treatment data of the patients.

[0028] Medical imaging data refers to visual information about the internal structure and function of the human body obtained from imaging examinations of the pelvic region of the patient. For example, if the patient is diagnosed with prostate cancer and needs to receive radiation therapy, then the medical imaging data could be CT images, ultrasound images, or MRI images of the lesion site obtained through various medical imaging techniques.

[0029] It is important to understand that in clinical practice, the clinical treatment data of the treated patient is also a crucial basis for medical decision-making and clinical research. This clinical treatment data can be formulated by experienced physicians and covers various aspects such as basic patient information, symptoms, examination and test results, treatment plans and their implementation, and records of changes in the patient's condition. It is important to note that clinical treatment data refers to treatment target values ​​based on guidelines for tumor radiotherapy and clinical practice experience. This pre-radiotherapy data belongs to the radiotherapy planning stage and is used to guide the design of the treatment plan; it does not represent the actual treatment outcome data. Specifically, clinical treatment data can include clinical dose data and preset priority levels for the target area and organs at risk within the lesion region of the treated patient. Clinical dose data for the target area can include the required prescription dose, target dose coverage requirements, and dose uniformity indicators; clinical dose data for organs at risk can include the maximum dose limit, average dose limit, and volume limits at different dose levels. Preset priority levels are a hierarchical system set according to the importance and sensitivity of different target areas and organs at risk, used to guide dose optimization and adjust optimization targets. For example, the preset priority level can be based on clinical needs and treatment goals, dividing the optimization direction of different target areas and organs at risk into multiple levels, with different levels corresponding to different optimization goals and dose processing methods. Furthermore, according to different preset priority levels, each target area and organ at risk has different treatment goals set during radiotherapy. For example, for high-priority target areas, the focus may be on ensuring they receive sufficient dose; while for low-priority organs at risk, dose coverage may be sacrificed to some extent to ensure the treatment effect of high-priority areas. Alternatively, for high-priority target areas and organs at risk, their processing order may be superior to low-priority lesion areas to prioritize ensuring the safety threshold of high-risk areas and perform multi-stage optimization process control; or, when multiple optimization strategies for a certain lesion area are mutually exclusive, the conflicting strategies can be arbitrated according to the preset level of the lesion area to maintain the validity of high-priority constraints. S120: Automatic segmentation based on medical image data to obtain target segmentation data.

[0030] The target segmentation data reflects the distribution of target areas and organs at risk within the lesion region of the treated patient. Specifically, after acquiring the patient's medical imaging data, a pre-trained segmentation model can be used to segment the region of interest, thus obtaining the target segmentation data. For example, a batch of medical imaging data from prostate cancer patients whose regions of interest have been labeled by professional physicians can be collected first, and this data can be divided into training, validation, and test sets according to a certain ratio. After data segmentation, preprocessing operations such as cropping and normalization are performed on the data, while the diversity of the training set data is enhanced through methods such as rotation. Next, a deep learning framework is used to build the segmentation model, and relevant parameters such as batch size and loss function are set. During the model training phase, the input image generates a segmentation result through forward propagation. This result is compared with the true labels to calculate the loss value, and then the backpropagation algorithm is used to update the model weights. During this process, the model performance is also evaluated using the validation and test sets to adjust and optimize the model. Ultimately, in practical applications, medical image data is input into this trained segmentation model, which can then automatically perform segmentation operations to obtain target segmentation data that reflects the distribution of target areas and organs at risk within the lesion region of the treated patient.

[0031] S130. Based on the target segmentation data, predict the dose distribution to obtain the initial dose distribution data.

[0032] It's important to understand that dose distribution data, also known as three-dimensional dose distribution data, refers to data that quantitatively describes and assesses the distribution of radiation therapy doses. It primarily describes the distribution pattern of the dose throughout the treatment volume, encompassing information such as dose uniformity and gradient. Initial dose distribution data, on the other hand, refers to data reflecting preliminary spatial distribution predictions of the radiation dose within the lesion region of the treated patient, based on the distribution of the target area and organs at risk. For example, initial dose distribution data may include preliminary estimates of the expected dose received by the target area and organs at risk, such as the dose magnitude and coverage area that different regions may receive.

[0033] Specifically, the initial dose distribution data can be obtained by predicting the dose distribution based on the target segmentation data, which can also be achieved using a pre-trained dose distribution prediction model. For example, firstly, actual radiotherapy dose distribution data corresponding to the target segmentation data is collected, labeled, and divided into training, validation, and test sets according to a certain ratio. Next, a suitable deep learning model (such as a convolutional neural network) is selected and its architecture is designed. Then, training parameters are set, a loss function is defined, and the training data is input into the model for forward propagation. Backpropagation and optimization algorithms are used to update the parameters. During training, the performance is evaluated using the validation set to adjust the model, and the model is evaluated using the test set. Finally, the target segmentation data to be predicted is input into this pre-trained dose distribution prediction model, and the model's output yields the initial dose distribution data.

[0034] S140. The initial dose distribution data is iteratively optimized using clinical dose data and preset priority levels to obtain the target dose distribution data for the treated subjects.

[0035] It needs to be clarified again that the clinical dose data here refers to the dose requirements for the target area and organs at risk within the lesion region of the treated subject, determined based on clinical treatment data. The initial dose distribution data, on the other hand, is a three-dimensional dose distribution data obtained by using a deep learning model to predict the dose distribution based on target segmentation data, providing a preliminary estimate of the expected dose received by the target area and organs at risk.

[0036] In this context, target dose distribution data refers to the ideal dose distribution result that meets clinical treatment needs, obtained after a series of optimization processes and control strategies in a prostate cancer radiotherapy plan. Specifically, to obtain target dose distribution data for the treated patient, it is first necessary to clarify the optimization goals and strategies. Clinical dose data contained in clinical treatment data can be used to obtain specific goals such as the maximum dose limit, average dose limit, and volume limit at different dose levels for the target area and organs at risk. Then, combined with preset priority levels, the optimization order and trade-off principles are clarified. For example, when the dose coverage priority of the target area is higher than the dose limit of organs at risk, the optimization process should prioritize ensuring the dose requirements of the target area while satisfying the dose limits of organs at risk as much as possible. Further, the initial dose distribution data is used as a starting point, and iterative calculations are performed according to the selected optimization algorithm and parameters. In each iteration, the dose distribution parameters are adjusted according to the optimization goals and strategies, such as adjusting the weight, angle, and shape of the irradiation field to change the dose distribution. Finally, after each iteration, evaluation indicators can be used to determine whether the dose distribution obtained in the current round meets the target requirements. Once the target requirement is met or the maximum number of iterations is reached, the iteration stops, and the dose distribution data obtained at this point is the target dose distribution data.

[0037] S150. Based on the target dose distribution data, iteratively optimize the radiotherapy plan to generate the target radiotherapy plan for the patient.

[0038] The targeted radiotherapy plan can be a set of dosimetric data describing the target area and organs at risk within the lesion region. For example, the targeted radiotherapy plan encompasses a range of information throughout the radiotherapy process, from setting radiotherapy equipment parameters to selecting the irradiation mode. Its purpose is to ensure that the target area receives the radiation dose specified by the target dose distribution data during radiotherapy, while maximizing the protection of organs at risk. Specifically, an initial radiotherapy plan can first be constructed based on the target dose distribution data, determining key elements such as radiation energy and dose rate. Then, using this initial radiotherapy plan as the core basis, iterative evaluation and optimization are performed. After each iteration, the variable parameters in the radiotherapy plan are modified, and a new round of evaluation and optimization is restarted. This process is repeated until the radiotherapy plan meets all preset criteria and achieves the clinically required indicators, thus generating a targeted radiotherapy plan suitable for the patient.

[0039] In the above implementation, based on the medical imaging data of the patient, automatic segmentation is performed to obtain target segmentation data reflecting the distribution of the target area and organs at risk within the lesion region. Then, dose distribution prediction is performed based on this data to obtain initial dose distribution data. This initial data is then iteratively optimized by combining clinical dose data and clinical treatment data to derive target dose distribution data, ultimately generating a target radiotherapy plan. Thus, by achieving accurate and efficient automatic segmentation of the prostate cancer target area and organs at risk, and through multiple rounds of iterative optimization of dose distribution and radiotherapy planning, a high-quality target radiotherapy plan is ultimately obtained.

[0040] In some implementation methods, please refer to the appendix. Figure 2 The initial dose distribution data is iteratively optimized using clinical dose data and preset priority levels to obtain the target dose distribution data for the treated subjects, including:

[0041] S210: Obtain multiple preset optimization strategies.

[0042] Pre-defined optimization strategies refer to a series of methods pre-set during the radiotherapy planning process to meet different clinical needs and goals, thereby improving the quality of radiotherapy plans and ensuring treatment efficacy. Specifically, these pre-defined optimization strategies can be developed based on clinical experience and research. For example, pre-defined optimization strategies may be priority control strategies that set specific dose targets for different target areas and organs at risk according to a pre-defined priority order; they may also be target coverage strategies that adjust parameters such as the prescription dose for specific target areas to improve target coverage; or they may be target dose conformity strategies that control the conformity of the prescription dose to the target area, etc.

[0043] S220. Following the execution order of multiple preset optimization strategies, the initial dose distribution data is adjusted sequentially using clinical dose data and preset priority levels to obtain the intermediate dose distribution data of the treated subjects.

[0044] It is important to understand that radiotherapy involves numerous target areas and organs at risk, each with different dose requirements, and different clinical needs and goals may also be mutually restrictive. Therefore, to meet complex clinical needs and avoid conflicts between multiple optimization strategies, the initial dose distribution data obtained from the deep learning model needs to be adjusted according to the execution order of multiple preset optimization strategies. Specifically, firstly, based on clinical treatment data, the specific content of the clinical dose data needs to be clearly defined, including but not limited to detailed information such as the current dose values ​​and dose distribution ranges of each target area and organ at risk. Simultaneously, a preset priority level needs to be accurately obtained from the clinical treatment data so that, during subsequent optimization, this preset priority level can guide the adoption of different optimization target settings for different lesion sites. For example, if the preset priority level for a certain organ at risk is clinical priority, then the clinical dose data in the clinical treatment data will be used as its optimization target during optimization; if the preset priority level for another organ at risk is prediction priority, then the initial dose distribution data obtained from the deep learning model will be used as its optimization target during optimization. Furthermore, based on the optimization objectives corresponding to each lesion site, and according to the execution order of multiple preset optimization strategies, the corresponding dose portion in the initial dose distribution data is initially adjusted to obtain the intermediate dose distribution data of the treated subject.

[0045] S230. Using the intermediate dose distribution data as the initial dose distribution data, repeat the above iterative optimization process until the target dose distribution data of the treated object is obtained.

[0046] Specifically, the intermediate dose distribution data obtained from the previous optimization round is used as the initial dose distribution data for the new round of optimization. Based on preset priority levels and the execution order of multiple preset optimization strategies, the doses to various target areas and organs at risk are adjusted using the new initial dose distribution data. Furthermore, after each iteration, the newly obtained intermediate dose distribution data is comprehensively evaluated, and the evaluation results are compared with preset evaluation criteria. If the evaluation results meet the preset criteria, the prediction target is considered achieved, and the iterative optimization process stops, with the iteration result determined as the target dose distribution data for the treated subject. Conversely, if the evaluation results do not meet the preset criteria, the current intermediate dose distribution data is used as the new initial dose distribution data, and the above iterative optimization and evaluation steps are repeated for another iteration.

[0047] It is understandable that the preset evaluation criteria can take many forms. For example, the preset evaluation criteria can be a maximum number of iterations. Once the number of iterations reaches this set value, the iteration process will terminate regardless of whether the final result meets the accuracy requirements. The preset evaluation criteria can also be various clinical evaluation indicators. When all clinical evaluation indicators of the iteration results meet the expected standards, it is considered that the target dose distribution data of the treated subject has been obtained, and the iterative optimization process stops.

[0048] In the above embodiments, by iteratively optimizing the initial dose distribution data using clinical dose data and preset priority levels, the quality of radiotherapy planning and treatment efficacy are effectively improved. Simultaneously, during radiotherapy, multiple preset optimization strategies based on clinical experience and research are used to sequentially adjust the dose, performing multiple iterative optimizations and making judgments based on preset evaluation criteria. This ensures that the final target dose distribution data is more accurate and reliable, thereby improving the accuracy and safety of radiotherapy, reducing damage to normal tissues, and also increasing the efficiency of radiotherapy planning. This contributes to achieving precise and standardized radiotherapy treatment for diseases such as prostate cancer.

[0049] In some implementations, multiple preset optimization strategies include priority control strategies and auxiliary adjustment strategies, with the priority control strategies executed before the auxiliary adjustment strategies. Following the execution order of each preset optimization strategy, the initial dose distribution data is optimized using clinical dose data and preset priority levels to obtain the intermediate dose distribution data for the treated subject. Please refer to the appendix. Figure 3 ,include:

[0050] S310. By using a priority control strategy, the initial dose distribution data of the lesion area is adjusted using clinical dose data and a preset priority level to obtain the first dose distribution data that matches the preset priority level.

[0051] Priority control strategy refers to a strategy used to selectively adjust the dose distribution of different target areas and organs at risk within a lesion region based on preset priority levels in clinical treatment data. For example, as shown in Table 1 below, the preset priority levels of all parts of the lesion region can be divided into three levels: 0, 1, and ≥2. Through priority control strategy, different optimization targets are set for lesion sites of different levels.

[0052] Table 1 Priority Control Strategy

[0053]

[0054]

[0055] Furthermore, after determining the optimization targets for different lesion sites based on the priority control strategy, the initial dose distribution data of the lesion area can be optimized and adjusted. For example, taking the small intestine, one of the organs at risk, as an example, its priority is set to 1, and based on clinical experience, its maximum dose is limited to less than or equal to 5000.00 cGy. If the clinical dose data in the clinical treatment data is less than this limit, but the initial dose distribution data predicted by the deep learning model shows that its current maximum dose exceeds this limit, then, according to the priority control strategy, the clinical dose data should be used as the predicted target value for the maximum dose. The radiotherapy parameters should be adjusted to make the dose distribution of the small intestine closer to the predicted target value, ultimately obtaining the first dose distribution data that matches the preset priority level.

[0056] S320. By using an auxiliary adjustment strategy, the first dose distribution data is further adjusted to obtain the intermediate dose distribution data.

[0057] In this context, auxiliary adjustment strategies refer to optimization strategies that make a series of supplementary adjustments to the first dose distribution data during the radiotherapy planning optimization process. For example, auxiliary adjustment strategies may include in-target dose control strategies and out-of-target dose reduction strategies, which are used to optimize and adjust for deficiencies or factors that have not been fully considered after the priority control strategy has been adjusted.

[0058] It is important to note that, compared to priority control strategies that determine the overall optimization direction, auxiliary adjustment strategies tend to modify and adjust certain details of the first dose distribution data to obtain more accurate, safer, and more effective intermediate dose distribution data. Therefore, priority control strategies are executed earlier than auxiliary adjustment strategies. Furthermore, after obtaining the first dose distribution data, a series of auxiliary adjustment strategies are used to comprehensively consider various complex situations during radiotherapy, such as target dose uniformity, dose limitations affecting organs at risk, and dose distribution conformity, to further supplement and adjust the details of the first dose distribution data, ultimately yielding the intermediate dose distribution data.

[0059] In the above implementation, the initial dose distribution data is optimized by sequentially executing a priority control strategy and an auxiliary adjustment strategy, significantly improving the quality and accuracy of the radiotherapy plan. The priority control strategy sets clear optimization targets for different target areas and organs at risk, ensuring the dose distribution initially meets basic clinical needs and laying the foundation for subsequent optimization. The auxiliary adjustment strategy further considers various complex factors in radiotherapy, making supplementary adjustments to address any shortcomings after adjusting the priority control strategy. This not only more accurately meets the dose requirements of each target area and organ at risk, reducing radiation damage to normal tissues, but also improves the overall safety and effectiveness of the radiotherapy plan, enhancing radiotherapy outcomes and contributing to precision treatment of diseases such as prostate cancer.

[0060] In some implementations, the auxiliary adjustment strategies include target coverage strategy, target dose conformity strategy, and out-of-target dose drop strategy, and the auxiliary adjustment strategies are executed in the following order: target coverage strategy, target dose conformity strategy, and out-of-target dose drop strategy.

[0061] It's important to understand that the auxiliary adjustment strategy, implemented sequentially as target coverage, target dose conformity, and off-target dose drop, is designed to ensure the safety and effectiveness of the radiotherapy plan. Specifically, target coverage is fundamental to radiotherapy; if the target area cannot be effectively covered, subsequent optimizations of dose conformity and organ-at-risk protection become meaningless. Only by ensuring sufficient target coverage can dose conformity be further optimized, allowing the radiation dose to more precisely match the target area's shape and enhancing tumor-killing power. Once both target coverage and off-target dose drop are guaranteed, the off-target dose drop strategy controls the dose to organs at risk outside the target area, preventing excessive radiation to normal tissues, reducing radiotherapy side effects, and ultimately achieving overall optimization of the radiotherapy plan.

[0062] Based on this, the first dose distribution data was further adjusted using an auxiliary adjustment strategy to obtain the intermediate dose distribution data. Please refer to the appendix. Figure 4a ,include:

[0063] S410. By using a target coverage strategy, the target coverage of the first dose distribution data is adjusted to obtain the second dose distribution data.

[0064] Target coverage strategy refers to an optimization strategy implemented during radiotherapy planning to ensure sufficient dose irradiation to the target area, thereby improving the coverage of the target area by the prescribed dose. Specifically, firstly, based on clinical experience and research, the dose coverage requirements for different target areas are determined. The actual coverage of the corresponding target areas in the first dose distribution data is analyzed to identify areas that do not meet the requirements. Further, based on the location and dose conditions of these areas, the parameters of the first dose distribution data are adjusted to obtain second dose distribution data that meets the target coverage requirements. For example, if an area of ​​the PTV (planning target area) is found to have insufficient dose, the intensity of the incident rays from the corresponding direction can be appropriately increased to improve the dose in that area, thereby enhancing the overall target coverage and ultimately obtaining the second dose distribution data.

[0065] S420. By using a target area dose conformal strategy, the dose field shape of the second dose distribution data is adjusted to obtain the third dose distribution data.

[0066] Target dose conformity strategy refers to an optimization strategy used to control the conformity of the prescribed dose to the target area, ensuring that the distribution shape of the irradiated dose matches the shape of the target area as closely as possible. Specifically, based on clinical experience and research, a target dose conformity index is determined. Then, based on the target volume and irradiated volume, the current dose conformity index is calculated. Further, based on the target dose conformity strategy, the difference between the current dose conformity index and the target dose conformity index is measured, and the dose distribution of the second dose distribution data is adjusted to improve the accuracy of radiotherapy and reduce unnecessary irradiation of normal tissues, resulting in the third dose distribution data. For example, the target dose conformity index can be determined using the following formula:

[0067] CI current =TV RI *TV RI / (V RI *TV)

[0068] In the formula, TV RI V represents the target volume, specifically the planned target volume (PTV); RI CI represents the irradiated volume, i.e., the volume actually irradiated; current The dose conformity index represents the degree of conformity between the current dose distribution and the target area. The closer it is to 1, the more conformal the dose distribution is, meaning the irradiated volume is closer to the target volume.

[0069] Furthermore, the process of adjusting using a target-area dose-conformity strategy can be referenced in the following formula:

[0070] f CIobj =w(CI)current -CI obj ) 2

[0071] In the formula, f CIobj represents the target dose conformity objective function, used to measure the difference between the current dose conformity index (CI) and the target dose conformity index; w represents the weighting factor used to adjust the weights of the objective function; CI current Indicates the current dose conformity index; CI obj The target dose conformity index is used to reflect the degree of conformity between the dose distribution and the target area under ideal conditions.

[0072] S430. By using the out-of-target dose drop strategy, the dose distribution of organs at risk outside the target area in the third dose distribution data is adjusted to obtain intermediate dose distribution data.

[0073] The off-target dose drop strategy refers to the optimization of radiotherapy planning by controlling the dose distribution outside the target area, ensuring that the dose decreases at a reasonable rate outside the target area, thereby reducing the irradiation dose to organs at risk and normal tissues outside the target area. Specifically, if there are important organs at risk near the target area, the off-target dose drop strategy can be used to adjust parameters such as the starting distance of the dose drop, the dose values ​​at different distances, and the drop rate to optimize the third dose distribution data, ensuring that the dose received by these organs at risk is within a safe range, while maintaining the dose in the target area, thus achieving effective control of the off-target dose.

[0074] For example, the key parameters of off-target dose drop can first be determined based on clinical experience and radiotherapy requirements. Please refer to... Figure 4b In the graph, the horizontal axis represents the distance from the target center, and the vertical axis represents the percentage of dose (Rx). Base dose, located at 100 on the vertical axis, represents 100% of the prescribed dose. Start dose, located at the edge of the target area, represents the initial dose at 105 on the vertical axis, indicating slightly above 100% of the prescribed dose. End dose, located at 50 on the vertical axis, represents 50% of the prescribed dose. distStart indicates the distance from the target edge where the dose begins to drop; it is represented by a specific distance value on the graph and marks the starting point of a rapid dose decrease. Lower Fall-off Rate represents a slower dose drop, indicating a gradual but slower dose decrease further from the target center. Higher Fall-off Rate represents a faster dose drop, indicating a rapid dose decrease closer to the target center.

[0075] Furthermore, the dose distribution of organs at risk outside the target area in the third dose distribution data is determined, and the dose distribution is adjusted according to the above parameters. For example, if the dose in a region of an organ at risk that is close to the target area is found to be too high, the dose in that region can be rapidly reduced according to the Higher Fall-off Rate by adjusting the intensity distribution of radiation or increasing shielding measures. For the dose in regions far from the target area, adjustments are made according to the Lower Fall-off Rate to ensure that the dose drops reasonably outside the target area. Finally, after adjusting the third dose distribution data, intermediate dose distribution data that effectively protects organs at risk can be obtained.

[0076] In the above implementation, by sequentially executing a series of auxiliary adjustment strategies—target coverage strategy, target dose conformity strategy, and off-target dose drop strategy—the quality and safety of radiotherapy planning are significantly improved, leading to better treatment outcomes for diseases such as prostate cancer. The target coverage strategy ensures that the target area receives sufficient dose irradiation, increasing the coverage of the target area by the prescribed dose, fundamentally guaranteeing the tumor-killing power of radiotherapy. The target dose conformity strategy, through precise calculation and adjustment of the dose conformity index, ensures that the distribution shape of the irradiated dose closely matches the target area, further improving the accuracy of radiotherapy, reducing unnecessary irradiation of normal tissues, and enhancing the tumor-killing effect while protecting normal tissues. The off-target dose drop strategy focuses on protecting organs at risk outside the target area. By rationally adjusting various key parameters of dose drop, it effectively controls the dose distribution of organs at risk outside the target area, avoiding excessive radiation to normal tissues and reducing radiotherapy side effects. The synergistic effect of these strategies resulted in intermediate dose distribution data that ensured effective treatment of the tumor while minimizing damage to normal tissues. This optimized the overall radiotherapy plan, improved the safety and effectiveness of radiotherapy, and helped enhance the patient's treatment experience and recovery outcomes.

[0077] In some implementation methods, please refer to the appendix. Figure 5 Based on the target dose distribution data, the radiotherapy plan is iteratively optimized to generate the target radiotherapy plan for the patient, including:

[0078] S510. Determine the initial radiotherapy plan based on the target dose distribution data.

[0079] The initial radiotherapy plan refers to a preliminary plan based on target dose distribution data. This plan sets the basic parameters of radiotherapy based on the target dose distribution data, but it has not yet been further optimized and may not fully meet clinical needs. Specifically, after obtaining the target dose distribution data, appropriate radiotherapy equipment and techniques can be selected based on the included information such as the dose distribution of the target area and organs at risk, and the type and energy of the radiation can be determined. Furthermore, based on the shape, location, and size of the target area, the number, size, and direction of the irradiation fields are planned.

[0080] S520. By using a global maximum dose control strategy, the maximum dose to the lesion area in the initial radiotherapy plan is optimized and adjusted to obtain an intermediate radiotherapy plan.

[0081] The global maximum dose control strategy refers to the strategy of controlling and optimizing the maximum dose that may occur in the entire treatment area in the radiotherapy plan. It is used to ensure that the dose at any point during the treatment process will not exceed the preset safety threshold, thereby avoiding excessive damage to normal tissues, while ensuring that the tumor target area can receive sufficient dose irradiation to achieve a balance between treatment effect and safety.

[0082] Specifically, after obtaining the initial radiotherapy plan, a global maximum dose control strategy is used. Based on the clinically defined maximum dose threshold for the entire treatment area, the dose distribution of the lesion area and surrounding normal tissues in the initial radiotherapy plan is analyzed to identify the location and value of the potential maximum dose. If a maximum dose is found to exceed the preset maximum dose threshold, the dose in the corresponding area of ​​the initial radiotherapy plan is optimized and adjusted to reduce the maximum dose value to within a safe range, ultimately resulting in an intermediate radiotherapy plan.

[0083] S530. Using the intermediate radiotherapy plan as the initial radiotherapy plan, repeat the above iterative optimization process until the final radiotherapy plan for the patient is obtained.

[0084] Specifically, after obtaining the intermediate radiotherapy plan, it can be treated as a new initial radiotherapy plan and optimized again using the global maximum dose control strategy. This process is repeated continuously, with further optimization of the dose distribution in each iteration to adjust the parameters of the radiotherapy plan. Simultaneously, the results of each iteration are continuously evaluated. The iteration can be stopped when the maximum dose remains stable below the safety threshold throughout the treatment area, and both the target dose and the dose to organs at risk meet clinical requirements. The resulting radiotherapy plan can then be used as the final radiotherapy plan.

[0085] S540. Determine the target radiotherapy plan for the patient based on the final radiotherapy plan.

[0086] Specifically, after obtaining the final radiotherapy plan, a comprehensive evaluation can be conducted to check whether the dose coverage of the target area is sufficient and uniform, whether the dose to organs at risk is within an acceptable range, and whether the overall dose distribution conforms to clinical guidelines and treatment goals. Once confirmed to be correct, the final radiotherapy plan can be determined as the target radiotherapy plan to guide the actual radiotherapy treatment process for the patient.

[0087] In the above implementation, a global maximum dose control strategy is used to strictly control the maximum dose across the entire treatment area during the optimization process, avoiding excessive radiation damage to normal tissues while ensuring sufficient dose irradiation to the tumor target area, thus balancing therapeutic efficacy and safety. Furthermore, multiple iterative optimizations are performed, continuously adjusting radiotherapy plan parameters to gradually approach the required levels, making the dose distribution more precise and reasonable until the maximum dose stabilizes below the safety threshold. Ultimately, a target radiotherapy plan highly tailored to the actual needs of the treated patient is generated, effectively improving the safety and effectiveness of radiotherapy.

[0088] In some implementation methods, please refer to the appendix. Figure 6 The target radiotherapy plan for the patient is determined based on the final radiotherapy plan, including:

[0089] S610. Based on the dose normalization strategy, the uniformity of dose distribution in the lesion area in the final radiotherapy plan is optimized and adjusted to obtain the optimized radiotherapy plan.

[0090] Dose normalization strategy refers to a strategy used in radiotherapy planning optimization to adjust and standardize the dose distribution in the lesion area, ensuring both uniformity and complete target coverage. Specifically, a unified dose standard or reference value can be pre-defined based on clinical indicators, and a detailed analysis of the dose distribution in the lesion area in the final radiotherapy plan can be conducted. Based on the analysis results, the dose distribution is optimized and adjusted using the dose normalization strategy. If the analysis reveals that the dose in a certain area is too high, the dose in that area can be reduced by decreasing the corresponding radiation weight, adjusting the incident angle of the radiation, or reducing the irradiation time. Conversely, if the dose in a certain area is too low, the opposite approach is taken to increase the dose in that area, ensuring a more uniform dose distribution after adjustment.

[0091] Optionally, after optimization, the dose distribution in the lesion area can be reassessed to check whether the dose normalization requirement has been met. For example, the uniformity of the dose distribution can be measured by calculating indicators such as the dose uniformity index. If the assessment results are unsatisfactory, adjustments are continued until the dose normalization standard is met, thus obtaining an optimized radiotherapy plan.

[0092] S620. Determine the target radiotherapy plan for the patient based on the optimized radiotherapy plan.

[0093] Specifically, after obtaining the optimized radiotherapy plan, a comprehensive review can be conducted, including assessing target volume dose coverage, whether dose restrictions for organs at risk are met, and the feasibility and safety of the radiotherapy plan. This includes checking whether the target volume is completely covered by the prescribed dose and whether the dose distribution is uniform; whether the dose to organs at risk is within a safe range and will not cause serious damage to normal tissues, etc. Furthermore, after a comprehensive review and evaluation, if the optimized radiotherapy plan meets all clinical requirements and standards and no problems or potential risks are found, it can be designated as the patient's target radiotherapy plan to guide the actual radiotherapy treatment process.

[0094] For example, the comparison between the target radiotherapy plan obtained by the above-described automated planning and the manual radiotherapy plan obtained by an experienced physician in actual clinical practice for the same patient can be seen in the following dose advantage table:

[0095] Table 2 Dosage Advantage Table

[0096]

[0097]

[0098] The table provides a detailed comparison of various assessment metrics for targeted radiotherapy plans and manual radiotherapy plans across multiple lesion regions and organs at risk. Target volume-related metrics include PTV45Gy, PTV65Gy, and PTV67.5Gy, representing the planning target volume for different prescription doses; HI stands for Homogeneity Index, measuring the uniformity of dose distribution within the target volume; CI is the Conformity Index, reflecting the degree of fit between the radiation dose distribution and the target volume shape; and coverage represents the percentage of the target volume actually receiving the prescription dose. Organ-at-risk metrics cover multiple sites, such as the bladder, rectum, small bowel, colon, anal canal, penile bulb, spinal cord, and pelvic bone. These indicators include the percentage of the total volume of each organ at risk that received different doses of radiation relative to the corresponding site, as well as the maximum dose Dmax (Gy) and mean dose Dmean (Gy), used to assess the extent of the impact of radiotherapy on these organs. Furthermore, the t-values / z-values ​​in the table measure the significance of the differences between the targeted radiotherapy plan and the manual radiotherapy plan on the corresponding assessment indicators, while the p-value further indicates whether such differences are statistically significant.

[0099] Furthermore, the results in the table clearly show that the dose distribution of the targeted radiotherapy plan and the manual radiotherapy plan are highly similar, with small differences in some key clinical dose parameters, all within clinically acceptable ranges. In some cases, the targeted radiotherapy plan is even more effective than the original manual clinical radiotherapy plan. Simultaneously, the DVH (dose-volume histogram) curves for PTV (target volume) and OARs (organs at risk) in both plans are largely consistent, exhibiting a high degree of agreement. In addition, the average computation time required to generate the targeted radiotherapy plan is only 8 minutes and 51 seconds, far less than the generation time of the manual radiotherapy plan, significantly improving the efficiency of radiotherapy planning.

[0100] In the above implementation, a dose normalization strategy is used to conduct a detailed analysis of the dose distribution in the lesion area within the final radiotherapy plan. Precise adjustments are made to areas with excessively high or low doses to obtain an optimized radiotherapy plan. After obtaining the optimized radiotherapy plan, it is comprehensively evaluated again to ensure that the plan meets clinical requirements and is free of any potential risks, thus finalizing the target radiotherapy plan. This target radiotherapy plan not only accurately guides the actual radiotherapy process for the treated patient, significantly improving the accuracy of radiotherapy, but also minimizes the risk of damage to normal tissues, effectively enhancing radiotherapy outcomes and improving the patient's treatment experience.

[0101] In some implementations, automatic segmentation is performed based on medical image data to obtain target segmentation data, including:

[0102] The 3D U-Net model is used to automatically delineate medical image data to obtain target segmentation data.

[0103] It's important to understand that the U-Net model is an image segmentation model based on a Convolutional Neural Network (CNN), while the 3DU-Net model is an evolution of the U-Net model that can also handle input images of arbitrary sizes. Specifically, it consists of an encoder (downsampling path) and a decoder (upsampling path), forming a U-shaped structure. The encoder extracts high-level features of the image by progressively applying convolutional and pooling layers, gradually reducing the size of the feature maps. For example, its encoder can consist of four convolutional blocks, each containing two 3x3 convolutional layers and a ReLU activation function, followed by a 2x2 max-pooling layer for downsampling after each convolutional block. The decoder, on the other hand, progressively restores the spatial resolution of the image through upsampling operations and combines the feature maps from the encoder with low-level and high-level features via skip connections. For example, its decoder can also consist of four upsampling blocks, each containing a transposed convolutional layer, two 3x3 convolutional layers, and a ReLU activation function. The upsampling block concatenates the output of the previous layer with the output of the corresponding encoder layer, creating a skip connection. This allows the model to fully utilize the detailed information from the encoding process during decoding, thereby improving segmentation accuracy. Furthermore, a bottleneck layer acts as a bridge between the encoder and decoder, containing two 3x3 convolutional layers and a ReLU activation function. Starting with the input image, the model sequentially passes through the encoder, bottleneck layer, and decoder, culminating in a 1x1 convolutional layer that adjusts the number of output channels to the required level, ultimately outputting the segmentation result.

[0104] Understandably, after determining the framework of the segmentation model, it needs to be trained. Specifically, medical image data with regions of interest already labeled by professional doctors can be used as samples, divided into training, validation, and test sets in an 8:1:1 ratio. Next, image preprocessing is performed on the medical image dataset, including cropping images to a uniform size, normalizing image pixel values, and increasing data diversity through operations such as rotation, flipping, translation, and scaling. During training, the 3D U-Net model can be trained using the deep learning framework PyTorch. Its batch size is set to 4, and the cross-entropy loss function is used as its loss function, with the Momentum optimizer (momentum factor of 0.9, weight decay coefficient of 0.0005) as the optimization function. The initial learning rate is set to 0.003, the maximum number of training epochs is 1000, and the number of training iterations is 30, with the learning rate dynamically adjusted using a cosine annealing strategy. Simultaneously, a validation set is used to evaluate model performance. Segmentation effectiveness is measured by calculating metrics such as the Dice coefficient and Intersection over Union (IoU), and the model performing best on the validation set is selected as the segmentation model. In each iteration, the input image is forward-propagated through the UNet network to generate segmentation results, which are then compared with the ground truth labels to calculate the loss function value. The gradient is calculated using the backpropagation algorithm to update the model weights. At the same time, data augmentation techniques are used to generate new training samples to improve training efficiency and stability. Correspondingly, after training, the final model performance can be evaluated on a test set to ensure its effectiveness in practical applications. For example, please refer to... Figure 7 The dashed lines in the figure represent the segmentation data output by the segmentation model, while the solid lines represent the regions of interest marked by physicians. As can be seen from the figure, this segmentation model has a much higher segmentation range and accuracy than manual annotation, and the delineation of organs at risk is also very comprehensive and accurate.

[0105] Understandably, based on the above training process, the trained segmentation model can be used for medical image segmentation tasks, that is, to automatically segment the medical image data of the treated object, and then obtain target segmentation data that reflects the distribution of target area and organs at risk in the lesion area of ​​the treated object.

[0106] Furthermore, dose distribution prediction is performed based on the target segmentation data to obtain initial dose distribution data, including:

[0107] The dose distribution of the target segmentation data is predicted based on the HD U-Net model to obtain the initial dose distribution data.

[0108] The HD U-Net model is also developed based on the U-Net model. It fully leverages the global and local information processing capabilities of U-Net and incorporates the more efficient feature propagation and reuse advantages of Dense Net. In the HD U-Net model, dense convolutions are used, effectively combining rectified linear units (ReLU) for quasi-convolution operations. Dense downsampling is performed through strided convolutions and ReLU, resulting in a new feature set with half the resolution. Subsequently, max pooling is applied to the previous feature set, and the previous feature sets are concatenated to form a new feature set. Finally, U-Net upsampling is performed using upsampling, convolution, and ReLU, connecting the feature sets on the other side of the "U"-shaped structure. For each dense operation, a growth rate is defined to determine the number of new features generated in the convolution step. In this process, the HD U-Net model fully utilizes a hierarchical structure; that is, there are different resolution levels between each max pooling or upsampling operation in U-Net, and convolutional layers are densely connected within each hierarchy, but not connected during upsampling operations between different levels of U-Net. This structural design enables more accurate capture and processing of information at different scales. Understandably, while the HD U-Net architecture has a slightly longer prediction time and requires more GPU random access memory, it exhibits better resistance to overfitting when modeling dose. For example, the 3D-UNet-based neural network model also consists of an encoder and a decoder. Both the encoder and decoder employ modules containing convolutional layers and attention mechanisms (CAD blocks). During the encoding phase, the number of convolutional filters in each module is 32, 64, 128, 256, and 512, respectively. After passing through the Max Pooling layer, the feature map size is reduced to half its original size. During decoding, the upsampling operation doubles the size of the feature map. Except for the output layer, where the convolutional filter size is 1×1×1, the remaining convolutional filters are all 3×3×3 in size. Furthermore, a skip connection mechanism is used between the encoder and decoder to fuse shallow image information with deep semantic information. By combining Group Norm layers and the He_normal method, the generalization ability of this model is significantly improved. The input to this model is three-dimensional contour information (i.e., segmentation data), and the output is three-dimensional dose field information (i.e., dose distribution data).

[0109] Similarly, after determining the framework of the dose prediction model, it is also necessary to train the model. First, clinical treatment data for each patient can be sampled beforehand, and the planned target volume (PTV) and organ at risk (OAR) contour information for each patient can be collected. This contour data will serve as input to the HD U-Net model. The actual dose distribution within the patient's body serves as the ground truth (true value) for training, guiding the model's learning. To ensure the data can be effectively processed by the dose prediction model, preprocessing operations such as normalization and resizing can be performed. Furthermore, during training, the Adam optimizer can be used with an initial learning rate set to 1e-4, a total number of training epochs set to 1000, and a weight decay coefficient of 1e-4. These parameter settings help the model adjust weights more reasonably during training, effectively avoiding overfitting and improving the model's generalization ability. Simultaneously, the mean squared error (MSE) is chosen as the loss function to measure the difference between the model's predicted dose distribution and the actual dose distribution (ground truth). By continuously minimizing this loss function, the model's prediction results gradually approach the true dose distribution. In the specific training implementation, Python 3.8 was used as the programming language, PyTorch 1.13 was used as the deep learning framework, and the training task was executed on an NVIDIA RTX 3090 GPU running a Linux system. Leveraging the powerful computing capabilities of the GPU significantly accelerated the model training speed.

[0110] It is important to note that this deep learning-based dose prediction algorithm primarily trains the model for dose control of organs at risk in radical prostate cancer; the target area prescription dose is mostly predicted and optimized based on actual conditions. Therefore, this model can be applied to multi-prescription planning.

[0111] Furthermore, based on the above training process, the processed target segmentation data can be input into the trained HD U-Net model. The model will perform feature extraction, downsampling, upsampling, and feature fusion on the input data through a series of operations of the encoder and decoder, thereby predicting the dose distribution and finally outputting the predicted three-dimensional dose field, which is the initial dose distribution data.

[0112] In the above implementation, automatic segmentation of medical image data based on the 3D U-Net model can accurately delineate medical images, significantly improving the segmentation range and accuracy of target areas, and providing comprehensive and highly accurate delineation of organs at risk. This not only improves the accuracy of automatic segmentation of organs at risk in the pelvic cavity, but also greatly enhances the work efficiency of radiation oncologists. Furthermore, the introduction of automatic target area segmentation technology fills the gap in automatic segmentation of prostate cancer target areas and contributes to the homogenization of prostate cancer radiotherapy. In the dose distribution prediction stage, dose distribution prediction based on the HD U-Net model integrates the advantages of U-Net and Dense Net, accurately capturing information at different scales using a hierarchical structure. It can quickly and accurately output initial dose distribution data based on target segmentation data, providing a precise and efficient foundation for radiotherapy planning. This powerfully promotes the development of radiotherapy technology towards precision and intelligence, contributing to the application and promotion of online adaptive radiotherapy technology in prostate cancer radiotherapy, improving radiotherapy accuracy, and reducing radiotherapy reactions.

[0113] It should be understood that although the steps in the flowchart above are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowchart above may include multiple steps or stages, which are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages in other steps.

[0114] This specification also provides an automated segmentation and intelligent planning device 800 suitable for adaptive radiotherapy of prostate cancer, such as... Figure 8 As shown, it includes: a data acquisition module 810, a delineation and segmentation module 820, a dose prediction module 830, a dose optimization module 840, and a plan generation module 850, wherein:

[0115] The data acquisition module 810 is used to acquire medical imaging data and clinical treatment data of the treated subject; wherein, the medical imaging data is obtained by imaging examination of the pelvic region of the treated subject; the clinical treatment data includes clinical dose data and preset priority levels of the target area and organs at risk in the lesion area of ​​the treated subject.

[0116] The delineation and segmentation module 820 is used to automatically segment based on medical image data to obtain target segmentation data; among which, the target segmentation data is used to reflect the distribution of target area and organs at risk in the lesion area of ​​the treated object.

[0117] The dose prediction module 830 is used to predict the dose distribution based on the target segmentation data to obtain the initial dose distribution data.

[0118] The dose optimization module 840 is used to iteratively optimize the initial dose distribution data using clinical dose data and preset priority levels to obtain the target dose distribution data for the treated subject.

[0119] The plan generation module 850 is used to iteratively optimize the radiotherapy plan based on the target dose distribution data and generate the target radiotherapy plan for the treated subject; wherein, the target radiotherapy plan is used to describe the target dosimetry data of the target area and organs at risk in the lesion area.

[0120] In some implementations, the dose optimization module 840 is further configured to acquire multiple preset optimization strategies; adjust the initial dose distribution data sequentially using clinical dose data and preset priority levels according to the execution order of the multiple preset optimization strategies to obtain intermediate dose distribution data of the treated subject; use the intermediate dose distribution data as the initial dose distribution data, and repeat the above iterative optimization process until the target dose distribution data of the treated subject is obtained.

[0121] In some implementations, multiple preset optimization strategies include priority control strategies and auxiliary adjustment strategies, and the execution order of the priority control strategies is earlier than the execution order of the auxiliary adjustment strategies. The dose optimization module 840 is further configured to adjust the initial dose distribution data of the lesion area using clinical dose data and preset priority levels through the priority control strategies to obtain first dose distribution data matching the preset priority levels; and to further adjust the first dose distribution data through the auxiliary adjustment strategies to obtain intermediate dose distribution data.

[0122] In some implementations, the auxiliary adjustment strategies include a target coverage strategy, a target dose conformity strategy, and an out-of-target dose drop strategy, and the auxiliary adjustment strategies are executed sequentially in the order of target coverage strategy, target dose conformity strategy, and out-of-target dose drop strategy. The dose optimization module 840 is further configured to adjust the target coverage of the first dose distribution data using the target coverage strategy to obtain second dose distribution data; adjust the dose field shape of the second dose distribution data using the target dose conformity strategy to obtain third dose distribution data; and adjust the dose distribution of the third dose distribution data related to organs at risk outside the target area using the out-of-target dose drop strategy to obtain intermediate dose distribution data.

[0123] In some implementations, the planning generation module 850 is also used to determine an initial radiotherapy plan based on target dose distribution data; optimize and adjust the maximum dose of the lesion area in the initial radiotherapy plan through a global maximum dose control strategy to obtain an intermediate radiotherapy plan; use the intermediate radiotherapy plan as the initial radiotherapy plan and repeat the above iterative optimization process until the final radiotherapy plan of the patient is obtained; and determine the target radiotherapy plan of the patient based on the final radiotherapy plan.

[0124] In some implementations, the planning generation module 850 is also used to optimize and adjust the uniformity of dose distribution in the lesion area in the final radiotherapy plan based on a dose normalization strategy to obtain an optimized radiotherapy plan; and to determine the target radiotherapy plan for the patient based on the optimized radiotherapy plan.

[0125] In some implementations, the delineation and segmentation module 820 is also used to automatically delineate medical image data based on a 3D U-Net model to obtain target segmentation data.

[0126] In some implementations, the dose prediction module 830 is also used to predict the dose distribution of the target segmentation data based on the HD U-Net model to obtain initial dose distribution data.

[0127] Specific limitations regarding the automated segmentation and intelligent planning device for adaptive radiotherapy suitable for prostate cancer can be found in the above-described limitations of the automated segmentation and intelligent planning method for adaptive radiotherapy suitable for prostate cancer, and will not be repeated here. Each module in the aforementioned automated segmentation and intelligent planning device for adaptive radiotherapy suitable for prostate cancer can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0128] In this embodiment, an automatic segmentation and intelligent planning device suitable for adaptive radiotherapy of prostate cancer is presented in the form of functional units. Here, a unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that execute one or more software or fixed programs, and / or other devices that can provide the above functions.

[0129] This application also provides a computer device, which may be a terminal, and its internal structure diagram may be as follows: Figure 9As shown, the computer device includes a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, NFC (Near Field Communication), or other technologies. When executed by the processor, the computer program implements an automatic segmentation and intelligent planning method suitable for adaptive radiotherapy of prostate cancer. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device casing, or an external keyboard, touchpad, or mouse.

[0130] Those skilled in the art will understand that Figure 9 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0131] This application also provides a computer-readable storage medium. The methods described in this application can be implemented in hardware or firmware, or implemented as recordable on a storage medium, or implemented as computer code downloaded over a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and subsequently stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, an optical disk, read-only memory, random access memory, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code, which, when accessed and executed, implements the methods shown in the above embodiments.

[0132] This application provides a computer program product including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the method of any embodiment of this application.

[0133] The above embodiments illustrate an automatic segmentation and intelligent planning method and apparatus suitable for adaptive radiotherapy of prostate cancer. This method and apparatus can be implemented using a computer chip or physical entity, or by a product with certain functions. A typical implementing device is a computer. Specifically, the computer can be, for example, a personal computer, a laptop computer, a cellular phone, or any combination of these devices. For ease of description, the above apparatus is described by dividing it into various functional units. Of course, in implementing this application, the functions of each unit can be implemented in one or more software and / or hardware components.

[0134] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0135] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0136] In the description of this specification, references to terms such as "an embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0137] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0138] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. Since they are fundamentally similar to the method embodiments, the descriptions are relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0139] The above description is merely an embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of this application should be included within the scope of the claims of this application.

[0140] Although embodiments of this application have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of this application, and all such modifications and variations fall within the scope defined by the appended claims.

Claims

1. A method for automatic segmentation and intelligent planning generation suitable for adaptive radiotherapy of prostate cancer, characterized in that, The method includes: Acquire medical imaging data and clinical treatment data of the treated subject; wherein, the medical imaging data is obtained by imaging examination of the pelvic region of the treated subject; the clinical treatment data includes clinical dose data and preset priority levels of the target area and organs at risk in the lesion area of ​​the treated subject; Automatic segmentation is performed based on the medical image data to obtain target segmentation data; wherein, the target segmentation data is used to reflect the distribution of target area and organs at risk in the lesion area of ​​the treated object; Based on the target segmentation data, dose distribution prediction is performed to obtain initial dose distribution data; Multiple preset optimization strategies are obtained; wherein, the multiple preset optimization strategies include priority control strategies and auxiliary adjustment strategies, and the execution order of the priority control strategies is earlier than the execution order of the auxiliary adjustment strategies; the auxiliary adjustment strategies include target coverage strategies, target dose conformity strategies, and out-of-target dose drop strategies, and the execution order of the auxiliary adjustment strategies is as follows: target coverage strategies, target dose conformity strategies, and out-of-target dose drop strategies are executed sequentially. The priority control strategy adjusts the initial dose distribution data of the lesion area using the clinical dose data and the preset priority level to obtain first dose distribution data matching the preset priority level. The preset priority level includes clinical priority and prediction priority. Adjusting the initial dose distribution data of the lesion area using the clinical dose data and the preset priority level includes: if the preset priority level for the organ at risk is clinical priority, then the clinical dose data is used as the optimization target; if the preset priority level for the organ at risk is prediction priority, then the initial dose distribution data is used as the optimization target. The target coverage of the first dose distribution data is adjusted using the target coverage strategy to obtain the second dose distribution data; the dose field shape of the second dose distribution data is adjusted using the target dose conformal strategy to obtain the third dose distribution data; and the dose distribution of the third dose distribution data related to organs at risk outside the target area is adjusted using the out-of-target dose drop strategy to obtain intermediate dose distribution data. The intermediate dose distribution data is used as the initial dose distribution data, and the above iterative optimization process is repeated until the target dose distribution data of the treated object is obtained. Based on the target dose distribution data, the radiotherapy plan is iteratively optimized to generate the target radiotherapy plan for the treated subject; wherein, the target radiotherapy plan is used to describe the target dosimetry data of the target area and organs at risk in the lesion region.

2. The method according to claim 1, characterized in that, The step of iteratively optimizing the radiotherapy plan based on the target dose distribution data to generate the target radiotherapy plan for the patient includes: The initial radiotherapy plan is determined based on the target dose distribution data; By using a global maximum dose control strategy, the maximum dose to the lesion area in the initial radiotherapy plan is optimized and adjusted to obtain an intermediate radiotherapy plan; The intermediate radiotherapy plan is used as the initial radiotherapy plan, and the above iterative optimization process is repeated until the final radiotherapy plan for the patient is obtained. The target radiotherapy plan for the patient is determined based on the final radiotherapy plan.

3. The method according to claim 2, characterized in that, Determining the target radiotherapy plan for the patient based on the final radiotherapy plan includes: The uniformity of dose distribution in the lesion area in the final radiotherapy plan is optimized and adjusted based on the dose normalization strategy to obtain the optimized radiotherapy plan. The target radiotherapy plan for the patient is determined based on the optimized radiotherapy plan.

4. The method according to claim 1, characterized in that, The automatic segmentation based on the medical image data to obtain target segmentation data includes: The medical image data is automatically delineated based on the 3D U-Net model to obtain the target segmentation data; The step of predicting the dose distribution based on the target segmentation data to obtain initial dose distribution data includes: The initial dose distribution data is obtained by predicting the dose distribution of the target segmentation data based on the HD U-Net model.

5. An automated segmentation and intelligent planning device suitable for adaptive radiotherapy of prostate cancer, characterized in that, The device includes: The data acquisition module is used to acquire medical imaging data and clinical treatment data of the treated subject; wherein, the medical imaging data is obtained by imaging examination of the pelvic region of the treated subject; the clinical treatment data includes clinical dose data and preset priority levels of the target area and organs at risk in the lesion area of ​​the treated subject; The delineation and segmentation module is used to automatically segment the medical image data to obtain target segmentation data; wherein, the target segmentation data is used to reflect the distribution of target area and organs at risk in the lesion area of ​​the treated object; The dose prediction module is used to predict the dose distribution based on the target segmentation data to obtain initial dose distribution data. A dose optimization module is used to acquire multiple preset optimization strategies; wherein, the multiple preset optimization strategies include a priority control strategy and an auxiliary adjustment strategy, and the execution order of the priority control strategy is earlier than the execution order of the auxiliary adjustment strategy; the auxiliary adjustment strategy includes a target coverage strategy, a target dose conformity strategy, and an out-of-target dose drop strategy, and the execution order of the auxiliary adjustment strategy is as follows: target coverage strategy, target dose conformity strategy, and out-of-target dose drop strategy are executed sequentially. The priority control strategy adjusts the initial dose distribution data of the lesion area using the clinical dose data and the preset priority level to obtain first dose distribution data matching the preset priority level. The preset priority level includes clinical priority and prediction priority. Adjusting the initial dose distribution data of the lesion area using the clinical dose data and the preset priority level includes: if the preset priority level for the organ at risk is clinical priority, then the clinical dose data is used as the optimization target; if the preset priority level for the organ at risk is prediction priority, then the initial dose distribution data is used as the optimization target. The target coverage of the first dose distribution data is adjusted using the target coverage strategy to obtain the second dose distribution data; the dose field shape of the second dose distribution data is adjusted using the target dose conformal strategy to obtain the third dose distribution data; and the dose distribution of the third dose distribution data related to organs at risk outside the target area is adjusted using the out-of-target dose drop strategy to obtain intermediate dose distribution data. The intermediate dose distribution data is used as the initial dose distribution data, and the above iterative optimization process is repeated until the target dose distribution data of the treated object is obtained. The plan generation module is used to iteratively optimize the radiotherapy plan based on the target dose distribution data to generate the target radiotherapy plan for the treated subject; wherein the target radiotherapy plan is used to describe the target dosimetry data of the target area and organs at risk in the lesion region.

6. A computer device, characterized in that, include: A memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, the processor executing the computer instructions to perform the method of any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing a computer to perform the method of any one of claims 1 to 4.

Images