Agent AI-based full-process intelligent radiotherapy system and working method thereof

Abstract

The invention relates to an Agent AI-based full-process intelligent radiotherapy system and a working method thereof, the system comprises a treatment Agent, a feedback Agent and a QA Agent, the feedback Agent is connected with the treatment Agent, the input of the treatment Agent is multi-modal data, and an initial radiotherapy scheme is dynamically output through deep learning and an evolutionary algorithm; the feedback Agent is connected with the multi-sensor system and used for monitoring user state data of the patient in real time, correspondingly adjusting treatment parameters according to an initial treatment scheme and outputting an optimized treatment scheme; and the QA Agent is used for monitoring key parameters in the treatment process, performing early treatment verification and later quality evaluation, and outputting a quality assurance report. Compared with the prior art, multi-Agent mutual cooperation is designed, multi-modal data are integrated, full-process automation from treatment scheme determination, data acquisition, target region sketching, plan generation, treatment execution to QA is achieved, dependence on manual intervention can be reduced, and treatment individuation and efficiency are improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent medical technology, and in particular to an Agent AI-based full-process intelligent radiotherapy system and its working method. Background Art

[0002] In the field of radiotherapy, with the development of artificial intelligence and automation technology, intelligent radiotherapy systems have been gradually applied in clinical practice. However, existing radiotherapy systems still have significant technical limitations in achieving full-process automation and de-manipulation. Most current radiotherapy planning systems rely on the guidance of doctors, providing the delineation of target areas and organs at risk and prescription doses, and using optimization algorithms or deep learning models to generate treatment plans [1]. Although these systems have certain automation capabilities in data processing and generating preliminary plans, the core decision-making links (such as prescription doses, target area delineation, plan evaluation and optimization) still require manual intervention by experts [2]. It is impossible to achieve full-process intelligent optimization of radiotherapy plans from generation to execution, and it is difficult to get rid of dependence on manual operations. For example, Chinese patent CN114423494A proposes a radiotherapy treatment plan using a differentiable dose function, which uses multimodal imaging information and physiological data for adaptive adjustment and automatically generates partial radiotherapy plans. However, the core of this technology lies in the adjustment and optimization of local plans, and manual decision-making is still required in initial and complex situations.

[0003] In recent years, some radiotherapy systems have introduced deep learning technology for automated target delineation, and have improved the accuracy of delineation to a certain extent [3]. These systems train models with a large amount of imaging data, which can assist doctors in reducing their workload when delineating the target area. However, this type of technology still cannot completely replace manual work, and the evaluation and adjustment of delineation results still require expert intervention, and lacks the ability to make dynamic adjustments during the treatment process [4]. For example, Chinese patent CN117219292A proposes a method and device for automatic dose verification and evaluation before precise radiotherapy for tumor patients. Its AI target delineation system automatically delineates the target area based on a deep learning model, but after generating the initial delineation, manual adjustment is still required to ensure accuracy; US patent US9697602B1 proposes an automatic contour drawing scheme for adaptive radiotherapy, which uses a method combining boundary detection and shape dictionary to automatically delineate and adjust the target area, but this scheme still requires manual intervention for fine-tuning in highly variable cases.

[0004] In addition, some intelligent radiotherapy systems have begun to integrate real-time feedback and adjustment functions, using sensors to monitor the patient's status during treatment and perform simple parameter adjustments [5]. However, the feedback mechanism and adjustment capabilities of such systems rely on preset rules and lack intelligent adaptive adjustment. The decision-making process often requires manual confirmation, making it impossible to autonomously optimize the treatment plan throughout the entire process and unable to achieve de-manual operation.

[0005] In existing studies, adaptive radiotherapy is expected to improve treatment efficacy by making real-time adjustments based on the patient's physiological changes in theory [6]. However, although adaptive radiotherapy has introduced artificial intelligence for assistance, its actual operation is still highly dependent on the participation of doctors. Before each treatment, doctors need to evaluate the latest imaging data and manually adjust the treatment plan, which increases the workload and introduces potential human errors. In addition, the calculation and workflow of adaptive radiotherapy are very complex. Especially during the treatment process, the patient's waiting time on the treatment table is limited. How to complete the adjustment and confirmation of the plan in a short time is a huge challenge. Although AI (Artificial Intelligence) has accelerated this process to a certain extent, manual review and confirmation are still required, which further increases time pressure and operational difficulty. At the same time, the success of adaptive radiotherapy also depends on a large amount of high-quality data and advanced algorithms, but existing AI models are usually trained on limited data sets, and their generalization ability in different data sets and clinical environments is limited. These problems have limited the application of adaptive radiotherapy in large-scale clinical promotion, and the actual efficacy and efficiency are difficult to guarantee.

[0006] In summary, most existing technologies have achieved intelligence and automation in some aspects of the radiotherapy process, but there are still shortcomings in key decision-making and dynamic adjustment.

[0007] Prior art literature:

[0008] [1]Bibault, JE, Giraud, P., Burgun, A. (2017). Big Data and machine learning in radiation oncology: State of the art and future prospects. Cancer Letters, 382(1), pp.110-117.

[0009] [2]Shi, W., Tian, J., Qi, W., et al. (2019). Artificial intelligence in radiotherapy: A review of current applications and challenges. Frontiers in Oncology, 9, pp. 1 - 10.

[0010] [3]Men, K., Zhang, T., Chen, X., et al. (2019). Fully automated radiotherapy treatment planning with deep learning: A clinical study for breast cancer. Frontiers in Oncology, 9, pp. 1 - 8.

[0011] [4]Giger, M. L. (2020). Machine learning in medical imaging. Journal of the American College of Radiology, 17(3), pp. 512 - 520.

[0012] [5]Li, W., Cao, P., Zhao, D., et al. (2021). The application of artificial intelligence in radiation oncology: A systematic review. Cancer Imaging, 21, pp. 1 - 13

[0013] [6]Ogawa, S., Onimaru, R., Oizumi, Y., Hirose, K., Miyasaka, Y., Yotsuya, T., et al. (2023). Prospects for online adaptive radiation therapy (ART) for head and neck cancer. Cancers, 16(6), 1206. Summary of the Invention

[0014] The purpose of the present invention is to overcome the defects of the above-mentioned existing technologies and provide a full-process intelligent radiotherapy system based on Agent AI and its working method. By introducing Agent AI (intelligent body) technology, designing multi-agent collaboration, integrating multimodal data, and realizing full-process automation from determining treatment plans, data collection, target area delineation, plan generation, treatment execution to QA (Quality Assurance), it can reduce dependence on manual intervention and improve the personalization and efficiency of treatment.

[0015] The objectives of the present invention can be achieved through the following technical solutions: a full-process intelligent radiotherapy system based on Agent AI, including a treatment agent, a feedback agent, and a QAAgent. The feedback agent is connected to the treatment agent. The treatment agent inputs multimodal data and dynamically outputs an initial radiotherapy plan through deep learning and evolutionary algorithms.

[0016] The feedback agent is connected to the multi-sensor system to monitor the patient user status data in real time, adjust the treatment parameters accordingly based on the initial treatment plan, and output an optimized treatment plan;

[0017] The QA Agent is used to monitor key parameters during the treatment process, perform early treatment verification, late quality assessment, and output a quality assurance report.

[0018] A working method of a full-process intelligent radiotherapy system based on Agent AI, comprising the following steps:

[0019] S1. Use treatment agents and QA agents to make intelligent treatment decisions and recommend personalized treatment plans;

[0020] S2. Use the treatment agent to collect multimodal data of patient users and perform fusion analysis to generate treatment recommendations;

[0021] S3. Use treatment agents and feedback agents to perform target delineation and dose planning;

[0022] S4. Use the treatment agent to generate an initial treatment plan, and use the feedback agent to optimize and dynamically adjust it to generate an optimized treatment plan;

[0023] S5. Use the treatment agent and feedback agent to perform pre-treatment data calibration and real-time verification;

[0024] S6. Using the treatment agent and feedback agent, the working state of the radiotherapy equipment is controlled accordingly according to the optimized treatment plan, and the state parameters during the treatment process are monitored in real time;

[0025] S7. Use the QA Agent and treatment Agent to analyze and process the post-treatment imaging data to dynamically update and adjust the treatment plan;

[0026] S8. Use QA Agent to perform verification, quality monitoring, and quality assessment for the early stage of treatment, during treatment, and after treatment, and output a quality assurance report.

[0027] Furthermore, the specific process of step S1 is as follows:

[0028] S11. Collect multimodal data of patient users, including imaging data, medical records data, and genetic information;

[0029] S12. Based on the multimodal data, determine whether the patient needs radiotherapy through knowledge graph and intelligent reasoning technology. If radiotherapy is determined to be necessary, execute step S13; otherwise, recommend other treatment methods and end the current process;

[0030] S13. Generate multiple candidate radiotherapy plans, and screen and obtain a recommended radiotherapy plan based on similar cases and multimodal data of patient users;

[0031] S14. Collect the patient's physiological data and, in combination with the recommended radiotherapy plan, determine whether the patient is suitable for adaptive radiotherapy. If so, generate an adaptive radiotherapy plan; otherwise, formulate a standard radiotherapy plan.

[0032] S15. Dynamically set the prescription dose for radiotherapy through historical data and deep learning models, conduct quality review on the generated prescription dose, and recommend corresponding clinical trials based on the individual characteristics of the patient user.

[0033] Furthermore, the specific process of step S2 is as follows:

[0034] S21. Collect the patient's current image data in real time, and synchronously integrate the patient's medical record data and examination data from multiple sources;

[0035] S22. Integrate the current imaging data, integrated medical record data, and examination data to generate a complete portrait of the patient user. By searching the distributed case database and using the federated learning algorithm based on blockchain technology, select the historical cases with the highest similarity and generate corresponding treatment recommendations.

[0036] Furthermore, the specific process of step S3 is as follows:

[0037] S31. Input the patient's image data into the deep learning model to delineate the tumor and surrounding organs at risk (OARs). Correct the boundaries of each target area and OAR through edge detection and morphological matching.

[0038] S32. Generate multiple candidate dose distribution schemes based on the patient's anatomical structure, tumor characteristics, and radiobiological model. The feedback agent uses a multi-objective optimization algorithm to screen the candidate dose distribution schemes, and verifies them based on radiotherapy biological parameters and machine learning models to obtain the optimal dose distribution scheme.

[0039] Furthermore, the specific process of step S4 is as follows:

[0040] S41. Generate an individualized treatment plan based on multimodal data and target delineation results;

[0041] S42. For individualized treatment plans, we use evolutionary algorithms to perform multiple rounds of evolutionary optimization and dynamic adjustment. We simulate the generated treatment plans in a virtual treatment environment, model various types of uncertainties based on Bayesian optimization fault tolerance analysis, and automatically adjust the treatment plans based on the simulation results.

[0042] S43. Use the embedded model verification mechanism to check the key parameters in the treatment plan and generate a test report.

[0043] Furthermore, the specific process of step S5 is as follows:

[0044] S51. Perform real-time multimodal data scanning before treatment begins and match real-time images of different modalities with the original treatment plan to monitor tumor location, organ movement, and morphological changes;

[0045] S52: Predict the patient's physiological dynamics during treatment through a multi-layer AI model, and adjust key parameters in the treatment plan accordingly;

[0046] S53. Verify the adjusted key parameters.

[0047] Furthermore, the specific process of step S6 is as follows:

[0048] S61, the treatment agent controls the radiotherapy equipment to perform radiotherapy operations according to the optimized radiotherapy plan;

[0049] S62. Real-time collection of patient status parameters during treatment to adaptively adjust radiotherapy plans, and detection and early warning of abnormalities during treatment through deep learning algorithms.

[0050] Furthermore, the specific process of step S7 is: after the treatment is completed, the image data of the patient user is collected, and an effect evaluation report is obtained through analysis and processing for use in the adjustment and update of the subsequent treatment plan.

[0051] Furthermore, the specific process of step S8 is as follows:

[0052] S81. Confirm the status of radiotherapy equipment before treatment begins and review the compliance of the treatment plan;

[0053] S82. Real-time monitoring of actual dose and key safety parameters during treatment;

[0054] S83. After the treatment is completed, the treatment effect is evaluated and a quality assurance report is generated.

[0055] Compared with the prior art, the present invention has the following advantages:

[0056] The present invention provides a treatment agent, a feedback agent, and a quality assurance agent (QA agent), connecting the feedback agent to the treatment agent. The treatment agent receives multimodal data as input and dynamically outputs an initial radiotherapy plan through deep learning and evolutionary algorithms. The feedback agent is connected to a multi-sensor system to monitor patient user status data in real time, adjust treatment parameters accordingly based on the initial treatment plan, and output an optimized treatment plan. The QA agent monitors key parameters during the treatment process, performs pre-treatment verification, post-treatment quality assessment, and outputs a quality assurance report. This creates a fully automated intelligent radiotherapy system. Combining the treatment agent, feedback agent, and QA agent, it achieves full automation and intelligent operation, from treatment plan generation, target area delineation, treatment process monitoring, to real-time dynamic adjustment. This ensures treatment effectiveness while significantly reducing the risk of manual operation and human error.

[0057] In this invention, through the collaborative work of the treatment agent, feedback agent, and QAAgent, full automation is achieved, from radiotherapy plan generation, treatment process monitoring, real-time dynamic adjustment, to post-treatment quality assessment. The treatment agent and QAAgent are used to make intelligent treatment decisions and recommend personalized plans; the treatment agent is used to collect multimodal data from patient users and perform fusion analysis; the treatment agent and feedback agent are used to perform target area delineation and dose planning; the treatment agent is used to generate the initial treatment plan, and the feedback agent is used to optimize and dynamically adjust it to generate an optimized treatment plan; the treatment agent and feedback agent are used to perform pre-treatment data calibration and real-time verification; the treatment agent and feedback agent are used to perform fully automatic treatment execution and real-time monitoring; the QA agent and treatment agent are used to analyze and process post-treatment imaging data; and the QA agent is used to perform quality assurance and monitoring throughout the pre-treatment, mid-treatment, and post-treatment stages. Thus, through the mutual collaboration of multiple agents and the integration of multimodal data, the entire process from treatment plan determination, data collection, target area delineation, plan generation, treatment execution, to QA is automated, greatly reducing reliance on manual intervention and significantly improving the personalization and efficiency of treatment. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 Schematic diagram of the method flow of the present invention;

[0059] Figure 2 Schematic diagram of the application process of the embodiment;

[0060] Figure 3 Schematic diagram of the intelligent treatment decision-making and personalized plan formulation process in the embodiment;

[0061] Figure 4 Schematic diagram of the automatic collection and preliminary analysis process of patient data in the embodiment;

[0062] Figure 5 Schematic diagram of the target delineation and dose planning, and automatic generation and optimization process of the treatment plan in the embodiment;

[0063] Figure 6 Schematic diagram of the data collection, analysis and processing process before, during and after treatment in the embodiment;

[0064] Figure 7 Schematic diagram of the working process of QA Agent before, during and after treatment in the embodiment. DETAILED DESCRIPTION

[0065] The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0066] Example

[0067] A full-process intelligent radiotherapy system based on Agent AI, including a treatment agent, a feedback agent, and a QAAgent. The feedback agent is connected to the treatment agent. The treatment agent takes multimodal data as input and dynamically outputs an initial radiotherapy plan through deep learning and evolutionary algorithms.

[0068] The feedback agent is connected to the multi-sensor system to monitor the patient's user status data in real time, adjust the treatment parameters accordingly based on the initial treatment plan, and output the optimized treatment plan;

[0069] QA Agent is used to monitor key parameters during the treatment process, conduct pre-treatment verification, post-treatment quality assessment, and output quality assurance reports.

[0070] The working method of the above-mentioned full-process intelligent radiotherapy system based on Agent AI is as follows: Figure 1 As shown, the following steps are included:

[0071] S1. Use treatment agents and QA agents to make intelligent treatment decisions and recommend personalized treatment plans;

[0072] S2. Use the treatment agent to collect multimodal data of patient users and perform fusion analysis to generate treatment recommendations;

[0073] S3. Use treatment agents and feedback agents to perform target delineation and dose planning;

[0074] S4. Use the treatment agent to generate an initial treatment plan, and use the feedback agent to optimize and dynamically adjust it to generate an optimized treatment plan;

[0075] S5. Use the treatment agent and feedback agent to perform pre-treatment data calibration and real-time verification;

[0076] S6. Using the treatment agent and feedback agent, the working state of the radiotherapy equipment is controlled accordingly according to the optimized treatment plan, and the state parameters during the treatment process are monitored in real time;

[0077] S7. Use the QA Agent and treatment Agent to analyze and process the post-treatment imaging data to dynamically update and adjust the treatment plan;

[0078] S8. Use QA Agent to perform verification, quality monitoring, and quality assessment for the early stage of treatment, during treatment, and after treatment, and output a quality assurance report.

[0079] This embodiment applies the above solution and builds a fully automated intelligent radiotherapy system based on Agent AI technology. It combines treatment agents, feedback agents, and QA agents to achieve full-process automation and intelligent operation from treatment plan generation, target area delineation, treatment process monitoring to real-time dynamic adjustment. Specifically:

[0080] 1. Treatment Agent: This agent uses multimodal data input, deep learning, and evolutionary algorithms to generate personalized radiotherapy plans. Unlike existing technologies that require physician involvement in target delineation and preliminary plan adjustments, the treatment agent possesses adaptive learning capabilities and can iteratively optimize treatment plans based on historical data and real-time feedback, thereby eliminating human error and improving the accuracy of personalized treatment.

[0081] 2. Feedback Agent: During treatment, the Feedback Agent monitors the patient's status through a multi-sensor system (such as real-time imaging and vital sign data) and, combined with adaptive algorithms, automatically adjusts treatment parameters. Unlike traditional feedback mechanisms that rely on preset rules, the Feedback Agent not only fine-tunes treatment plans based on real-time data but also dynamically updates treatment plans to address physiological changes during treatment, ensuring continuous optimization of treatment.

[0082] 3. QA Agent: The QA Agent focuses on quality assurance throughout the entire treatment process through an independent quality monitoring and evaluation system. Unlike traditional QA processes that rely on manual inspections and fixed rules, the QA Agent, based on multimodal data analysis and deep learning algorithms, can monitor key parameters during the treatment process in real time, including equipment calibration, dose distribution, and treatment accuracy. This agent not only has the ability to automatically identify potential risks, but can also dynamically adjust quality control standards based on actual conditions to ensure consistency between treatment plans and actual implementation. In addition, the QA Agent uses continuous data feedback and historical learning to self-iterate and optimize the quality control process, significantly reducing the impact of human intervention on the safety and effectiveness of radiotherapy, thereby achieving fully automated quality assurance.

[0083] This system significantly reduces manual intervention and potential decision-making conflicts through a fully automated treatment process. Existing technologies often lead to inconsistent treatment plans due to differences in physician experience and inconsistent standards. This solution, through a standardized intelligent decision-making system, ensures consistent and coherent treatment plans.

[0084] The application process of this embodiment is as follows Figure 2 As shown, it mainly includes:

[0085] Step 1: Intelligent treatment decision-making and personalized plan formulation (such as Figure 3 shown)

[0086] 1. Intelligent Determination of Patient Treatment Options: Before making a radiotherapy decision, the system first comprehensively assesses the patient's overall health status, including tumor type, medical history, genetic profile, and immune status, using an embedded large language model and distributed knowledge base. The system automatically analyzes multimodal data (imaging, medical records, and genetic information) and uses knowledge graphs and intelligent reasoning technology to determine whether the patient needs radiotherapy. If radiotherapy is not necessary, the system recommends alternative treatment options, such as surgery or chemotherapy.

[0087] 2. Personalized treatment recommendations: After determining that a patient requires radiotherapy, the system automatically generates multiple candidate options, including different radiotherapy modalities (IMRT, VMAT, SBRT, or radiotherapy combined with immunotherapy). Using a federated learning algorithm, the system extracts the most effective treatment options from similar cases and compares them with the patient's individual characteristics. During this process, the system not only considers the radiotherapy modality but also predicts which treatment combination will be most effective based on the biological characteristics of the patient's tumor, gene expression, and immune response.

[0088] 3. Adaptive radiotherapy need determination: The system automatically determines whether a patient is suitable for adaptive radiotherapy based on the patient's real-time physiological data (such as breathing patterns and organ displacement) and dynamic changes in tumor location. If the system identifies a patient who may experience significant physiological changes during treatment, adaptive radiotherapy is recommended and treatment is administered within the optimal time window.

[0089] 4. Prescription Dose and Clinical Trial Recommendations: The system dynamically sets the prescribed radiotherapy dose using historical data and deep learning models, ensuring a balance between therapeutic efficacy and safety. Simultaneously, a QA agent conducts a quality review of the generated prescription dose to ensure the safety and reliability of all parameters. The system then assesses whether the patient is eligible for a specific clinical trial and recommends appropriate clinical trials based on the patient's individual characteristics. The system comprehensively considers the patient's health status, treatment goals, and the risk-benefit ratio of the trial to ensure that the recommended trial best meets the patient's needs.

[0090] Step 2: Automatic collection and preliminary analysis of patient data (e.g. Figure 4 shown)

[0091] 1. Automatic Image Data Acquisition: Upon arrival at the radiotherapy center, imaging equipment (such as CT and MRI) automatically captures the patient's latest imaging data and uploads it to the central data processing system via the DICOM standard. During data acquisition, the system automatically assesses image quality and adjusts scanning parameters in real time to ensure that the acquired imaging data meets diagnostic standards. Once acquired, the imaging data is uploaded to the central data processing system via the high-speed DICOM transmission protocol.

[0092] 2. Multi-source Synchronous Integration of Medical Records and Examination Data: The system seamlessly integrates with HER (Electronic Human Resource Management) to automatically extract a patient's lifecycle health data, including medical history, laboratory tests, genetic data, medication usage, and more. This process integrates and cleanses data using an intelligent interface based on the FHIR standard, ensuring seamless integration of data from different sources. In this embodiment, the system's built-in natural language processing (NLP) module automatically identifies and structures unstructured text information in medical records, enabling real-time integration and analysis of multi-source data.

[0093] 3. Multimodal Data Fusion and Preliminary Analysis: This embodiment designs a data fusion module that fuses multimodal (CT, MRI, PET) images with medical records to generate a complete multimodal patient portrait. The treatment agent utilizes an enhanced deep learning network (e.g., an adaptive transfer learning model) to automatically delineate the tumor region and generate preliminary target region definitions and tumor feature analysis results (e.g., volume, morphology, and location).

[0094] 4. Intelligent Case Comparison and Treatment Recommendations: The system automatically searches a distributed case database and uses a blockchain-based federated learning algorithm to select the most similar historical cases. Through cross-institutional data sharing, the system provides highly personalized treatment recommendations, including optimal treatment options, expected efficacy, and possible side effects, while protecting privacy.

[0095] Step 3: Accurate target delineation and dose planning (eg Figure 5 shown)

[0096] 1. Multi-round Target Delineation and Organs at Risk (OAR) Labeling: The system utilizes a self-supervised learning algorithm to automatically delineate tumors and surrounding organs at risk (OARs) without human intervention. Based on a multi-round, iterative active learning mechanism, the model continuously updates and adjusts during treatment to adapt to tumor changes at different stages. The feedback agent performs self-calibration after each delineation round, ensuring the precise boundaries of each target volume and OAR through high-precision image comparison (such as edge detection and morphology matching).

[0097] 2. Automatic Dose Distribution Generation: The system automatically generates multiple candidate dose distributions based on patient anatomy, tumor characteristics, and radiobiological models. In this embodiment, the system's built-in dose optimization algorithm also considers spatiotemporal dose control (e.g., adjusting dose gradients within different treatment cycles), providing refined treatment strategies for complex tumors.

[0098] 3. Multi-objective optimization-based solution screening: The feedback agent uses advanced multi-objective optimization algorithms (such as evolutionary multi-objective optimization, EMO) to screen solutions and conducts multiple verifications based on radiotherapy biological parameters and machine learning models to ensure that the final selected solution achieves the optimal balance in multiple dimensions such as tumor coverage, healthy tissue protection, and treatment time.

[0099] Step 4: Automatic generation and optimization of treatment plans (e.g. Figure 5 shown)

[0100] 1. Dynamic Generation of Individualized Treatment Plans: The system automatically generates a treatment plan based on multimodal data and target delineation, including beam angle, radiation energy, beam shape (such as dynamic multileaf grating control), and fractionation scheme. This process utilizes an intelligent strategy based on reinforcement learning to automatically adjust the beam path when dealing with complex anatomical structures, ensuring target coverage while minimizing exposure to normal tissue.

[0101] 2. Multi-round evolutionary optimization and dynamic adjustment: After the treatment plan is generated, the system initiates an evolutionary algorithm, performing multiple rounds of iterative optimization using simulated annealing, genetic algorithms, and other methods. Each round of optimization considers not only physical dose parameters but also the patient's physiological state (such as respiration and blood flow) and device status (such as radiation output stability), achieving highly personalized dynamic optimization.

[0102] 3. Virtual Treatment Simulation and Fault-Tolerance Verification: The system simulates the generated treatment plan in a virtual treatment environment, including the impact of various possible patient states (such as respiration, body position, and organ movement) on treatment outcomes. The feedback agent models various uncertainties through fault-tolerance analysis based on Bayesian optimization and automatically adjusts the treatment plan based on the simulation results, ensuring robustness in a real-world treatment environment.

[0103] 4. Plan Generation and Automatic Verification: The final treatment plan undergoes a multi-level automated verification process. The system utilizes an embedded model verification mechanism to check key parameters, such as dose coverage, beam path redundancy, and device tolerances, to ensure that the plan is free of safety hazards during execution. The Feedback Agent acts as a supervisor during the verification process, continuously analyzing whether various indicators meet clinical safety requirements and conducting safety and feasibility assessments. All verification data is automatically generated into a report for system archiving and subsequent audits.

[0104] Step 5: Automatic calibration and real-time verification (such as Figure 6 shown)

[0105] 1. Real-time scanning and data matching before treatment: Before each treatment, the system automatically triggers imaging equipment (such as CBCT, MRI, PET) to perform multimodal real-time scanning. The image processing module uses AI-based cross-modal fusion and registration algorithms to accurately match real-time images of different modalities with the original plan, ensuring comprehensive monitoring of tumor location, organ movement, and morphological changes. The system uses deep learning models to automatically detect abnormal changes (such as tumor morphological mutations or organ displacement) and automatically issues adjustment instructions when necessary.

[0106] 2. Adaptive parameter fine-tuning: The system not only makes adjustments based on real-time scan data but also uses a multi-layered AI model to predict the patient's physiological dynamics during treatment (such as changes in respiratory rate). Based on these predictions, the system proactively adjusts key parameters such as beam path, dose distribution, and irradiation time to ensure that the treatment plan remains accurate across a wider range of patient conditions. The feedback agent monitors these adjustments in real time and dynamically updates the adjustment strategy based on pre-set rules.

[0107] 3. Intelligent Self-Verification and Dynamic Closed-Loop Feedback: After each calibration, the feedback agent automatically generates a dynamic self-verification report based on multiple historical data and intelligently verifies the adjusted parameters. The system combines data accumulated from previous treatments to perform trend analysis on the current verification results, ensuring that deviations from the original treatment target remain within the optimal range. In this embodiment, all verification data is intelligently encrypted and archived to ensure data security and integrity, while also providing a foundation for future QA analysis and system upgrades.

[0108] Step 6: Fully automatic treatment execution and real-time monitoring (such as Figure 6 shown)

[0109] 1. Seamless execution of treatment plans: The treatment agent fully controls the radiotherapy equipment and automatically executes radiotherapy according to the optimized plan. The system schedules beam path, energy intensity, and timing parameters in real time, adjusting them based on the patient's condition to ensure high-precision irradiation of the tumor area.

[0110] 2. Real-time Monitoring and Dynamic Adjustment: The system integrates multimodal sensors and imaging equipment to monitor the patient's breathing, heart rate, movement, and other dynamic parameters in real time. The feedback agent uses this data to make multi-dimensional adjustments, enabling adaptive radiotherapy and ensuring accuracy and stability during treatment.

[0111] 3. Automatic Anomaly Detection and Self-Adjustment: The system uses deep learning algorithms to detect anomalies during treatment, such as equipment failure, patient misalignment, and unstable radiation. During treatment, the system's built-in anomaly detection module continuously monitors equipment operation and patient status. If any equipment anomaly or patient discomfort occurs, it automatically issues an alert and performs self-adjustments to ensure a safe and stable treatment process.

[0112] Step 7: Post-treatment data analysis and plan adjustment (e.g. Figure 6 shown)

[0113] 1. Automatic analysis of post-treatment imaging data: After each treatment, the system automatically collects and analyzes imaging data, generating reports such as tumor change curves and dose-effect relationship diagrams. The AI ​​model uses this data to determine treatment efficacy, including tumor reduction, target area changes, and healthy tissue protection, and generates a detailed efficacy evaluation report.

[0114] 2. Dynamically adjust subsequent treatment plans: The system automatically updates and adjusts subsequent treatment plans based on the effect evaluation report, including adjusting target area boundaries, redistributing doses, optimizing irradiation paths, etc., to adapt to the dynamic changes of the tumor and ensure the continued effectiveness and accuracy of the treatment.

[0115] 3. Personalized long-term follow-up and health monitoring: After treatment is completed, the system enters intelligent follow-up mode, regularly collecting imaging and biomarker data for remote health monitoring. The system triggers recurrence warnings based on review results and generates follow-up treatment recommendations based on real-time data, ensuring long-term tracking and management of treatment outcomes.

[0116] Step 8: Automated quality assurance (QA) and safety monitoring (e.g. Figure 7 shown)

[0117] 1. Early treatment verification

[0118] Before each treatment begins, QA Agents perform comprehensive equipment and system checks to verify the operational status and safety of all radiotherapy equipment. This includes system self-tests, checking the intensity and accuracy of radiation sources, and confirming the proper functioning of all safety gates and emergency stop systems. Furthermore, QA Agents conduct a comprehensive review of treatment plans to ensure accurate dose calculations and that the plan complies with the clinical pathway and patient-specific needs.

[0119] 2. Quality monitoring during treatment implementation

[0120] Unlike the dynamic adjustments made by the feedback agent, the QA agent focuses on real-time monitoring of the consistency between the actual administered dose and the planned dose during treatment, ensuring accurate dose delivery. The QA agent also monitors key safety parameters during treatment, such as radiation leakage and abnormal equipment operation, to ensure the safety of both patients and operators.

[0121] 3. Post-quality evaluation and feedback

[0122] After the treatment is completed, the QA Agent conducts an in-depth analysis of the treatment results, assessing the tumor response and the impact on surrounding healthy tissues. By comparing pre- and post-treatment imaging data, the QA Agent assesses the degree of tumor shrinkage and possible side effects. Based on these analyses, the QA Agent generates a detailed treatment quality report that records all key data and any deviations from the planned schedule, providing support for clinical decision-making.

[0123] 4. Continuous optimization of system performance and security

[0124] The QA Agent is responsible for continuously tracking and analyzing long-term treatment data, leveraging big data and machine learning to continuously optimize treatment strategies. In this embodiment, based on quality assurance reports and technological developments, the QA Agent also recommends system hardware and software updates and parameter adjustments, ensuring the system remains at the forefront of technological advancements while meeting new clinical needs.

[0125] It should be noted that, when actually applying this solution, a variety of changes and adjustments can be made to adapt to different clinical needs and application scenarios. For example, this embodiment uses deep learning and reinforcement learning algorithms for intelligent control, but in some special cases, other optimization methods such as genetic algorithms and simulated annealing algorithms can also be used to provide more flexible treatment plans. Secondly, in terms of data processing, the system can not only integrate current multimodal imaging and medical record data, but can also be expanded to process more complex data sources, such as three-dimensional reconstruction data or distributed data sharing based on privacy protection. In addition, the system's architectural design has a high degree of device compatibility and can be connected to a variety of radiotherapy equipment as needed, and even supports the integration of new equipment such as proton therapy. Finally, the system's feedback mechanism can be customized according to the actual application scenario, and the safety and reliability of the treatment process can be ensured by introducing expert systems or cross-institutional collaboration mechanisms. In other words, the flexibility of this solution makes it widely applicable, and it can meet diverse clinical needs through the combination and adjustment of different modules.

[0126] Compared with existing technologies, the advantages of this solution are mainly reflected in the following aspects:

[0127] 1. Solve the problem of insufficient personalized regulation of traditional artificial intelligence in radiotherapy

[0128] Traditional AI systems typically rely on large amounts of training data, generating standardized radiotherapy plans through fixed models. This approach is insufficient for individualized treatment needs. Intelligent agents, however, go a step further, accumulating experience and learning through real-world applications, enabling real-time learning and adjustment. For example, when determining prescribed doses, traditional systems typically only provide generalized recommendations based on preset templates, failing to fully consider individual patient differences. However, intelligent agents can dynamically adjust prescribed doses during treatment based on the patient's real-time response and condition, ensuring that the plan better meets the patient's specific needs. This flexible, personalized control capability effectively improves the accuracy and personalization of treatment.

[0129] 2. Address the problem of insufficient individualization of treatment plans and reliance on manual intervention in the current radiotherapy decision-making process

[0130] The current radiotherapy plan development process relies primarily on manual operations and decision-making by doctors and medical technicians based on imaging data and the radiotherapy planning system. Due to differences in doctors' experience levels and judgment criteria, plan development often lacks consistency, making it difficult to accurately match the needs of individual patients. This inconsistency directly affects the effectiveness of radiotherapy and may even lead to treatment failure. This solution introduces agent AI technology, which can significantly improve the individualization and consistency of radiotherapy plans through intelligent data analysis and automated decision-making. The agent system can fully consider individual patient differences, reduce errors caused by manual operations, and ensure that each patient receives the most appropriate treatment plan.

[0131] 3. Solving the problem of real-time dynamic adjustment during radiotherapy

[0132] Existing radiotherapy systems generally lack the ability to monitor and adjust the patient's condition in real time during treatment, and are unable to effectively respond to the patient's physiological changes during treatment. This lack of real-time dynamic adjustment may affect the overall treatment effect, especially when the patient's condition changes unexpectedly, it is difficult to respond in time. To solve this problem, the Agent AI system in this solution integrates a multimodal data perception module, which can obtain a variety of information such as the patient's imaging data and vital signs data in real time, and dynamically adjust it in combination with deep learning algorithms. In this way, the system can continuously optimize treatment parameters during radiotherapy, ensure that the treatment process is always in the best state, and improve the efficacy.

[0133] 4. Improving the efficiency and automation of radiotherapy plan optimization

[0134] The existing radiotherapy plan optimization process typically relies on repeated adjustments based on expert experience. This is not only time-consuming and labor-intensive, but also susceptible to human interference, resulting in unstable plan quality. To improve optimization efficiency and automation, this solution, based on the feedback mechanism and optimization algorithm of the Agent AI system, significantly accelerates the generation and optimization of radiotherapy plans through continuous model updates and automated processes. This system not only rapidly generates optimized treatment plans but also possesses the ability to continuously learn and improve, thereby continuously improving overall system performance and reducing reliance on expert manual intervention.

Claims

1. A full-process intelligent radiotherapy system based on Agent AI, characterized by: The system includes a treatment agent, a feedback agent, and a QAAgent. The feedback agent is connected to the treatment agent. The treatment agent takes multimodal data as input and dynamically outputs an initial radiotherapy plan through deep learning and evolutionary algorithms. The feedback agent is connected to the multi-sensor system to monitor the patient user status data in real time, adjust the treatment parameters accordingly based on the initial treatment plan, and output an optimized treatment plan; The QAAgent is used to monitor key parameters during treatment, perform pre-treatment verification, post-treatment quality assessment, and output quality assurance reports.

2. A working method of a full-process intelligent radiotherapy system based on Agent AI, applied to the full-process intelligent radiotherapy system based on Agent AI as claimed in claim 1, characterized in that: The following steps are involved: S1. Use treatment agents and QA agents to make intelligent treatment decisions and recommend personalized treatment plans; S2. Use the treatment agent to collect multimodal data of patient users and perform fusion analysis to generate treatment recommendations; S3. Use treatment agents and feedback agents to perform target delineation and dose planning; S4. Use the treatment agent to generate an initial treatment plan, and use the feedback agent to optimize and dynamically adjust it to generate an optimized treatment plan; S5. Use the treatment agent and feedback agent to perform pre-treatment data calibration and real-time verification; S6. Using the treatment agent and feedback agent, the working state of the radiotherapy equipment is controlled accordingly according to the optimized treatment plan, and the state parameters during the treatment process are monitored in real time; S7. Analyze and process the post-treatment imaging data using QAAgent and treatment agent to dynamically update and adjust the treatment plan. S8. Use QAAgent to perform verification, quality monitoring, and quality assessment before treatment, during treatment, and after treatment, and output quality assurance reports.

3. The working method of the full-process intelligent radiotherapy system based on Agent AI according to claim 2 is characterized in that: The specific process of step S1 is: S11. Collect multimodal data of patient users, including imaging data, medical records data, and genetic information; S12. Based on the multimodal data, determine whether the patient needs radiotherapy through knowledge graph and intelligent reasoning technology. If radiotherapy is determined to be necessary, execute step S13; otherwise, recommend other treatment methods and end the current process; S13. Generate multiple candidate radiotherapy plans, and screen and obtain a recommended radiotherapy plan based on similar cases and multimodal data of patient users; S14. Collect the patient's physiological data and, in combination with the recommended radiotherapy plan, determine whether the patient is suitable for adaptive radiotherapy. If so, generate an adaptive radiotherapy plan; otherwise, formulate a standard radiotherapy plan. S15. Dynamically set the prescription dose for radiotherapy through historical data and deep learning models, conduct quality review on the generated prescription dose, and recommend corresponding clinical trials based on the individual characteristics of the patient user.

4. The working method of the full-process intelligent radiotherapy system based on Agent AI according to claim 2 is characterized in that: The specific process of step S2 is: S21. Collect the patient's current image data in real time, and synchronously integrate the patient's medical record data and examination data from multiple sources; S22. Integrate the current imaging data, integrated medical record data, and examination data to generate a complete portrait of the patient user. By searching the distributed case database and using the federated learning algorithm based on blockchain technology, select the historical cases with the highest similarity and generate corresponding treatment recommendations.

5. The working method of the full-process intelligent radiotherapy system based on Agent AI according to claim 2 is characterized in that: The specific process of step S3 is: S31. Input the patient's image data into the deep learning model to delineate the tumor and surrounding organs at risk (OARs). Correct the boundaries of each target area and OAR through edge detection and morphological matching. S32. Generate multiple candidate dose distribution schemes based on the patient's anatomical structure, tumor characteristics, and radiobiological model. The feedback agent uses a multi-objective optimization algorithm to screen the candidate dose distribution schemes, and verifies them based on radiotherapy biological parameters and machine learning models to obtain the optimal dose distribution scheme.

6. The working method of the full-process intelligent radiotherapy system based on Agent AI according to claim 2 is characterized in that: The specific process of step S4 is as follows: S41. Generate an individualized treatment plan based on multimodal data and target delineation results; S42. For individualized treatment plans, we use evolutionary algorithms to perform multiple rounds of evolutionary optimization and dynamic adjustment. We simulate the generated treatment plans in a virtual treatment environment, model various types of uncertainties based on Bayesian optimization fault tolerance analysis, and automatically adjust the treatment plans based on the simulation results. S43. Use the embedded model verification mechanism to check the key parameters in the treatment plan and generate a test report.

7. The working method of the full-process intelligent radiotherapy system based on Agent AI according to claim 2 is characterized in that: The specific process of step S5 is as follows: S51. Perform real-time multimodal data scanning before treatment begins and match real-time images of different modalities with the original treatment plan to monitor tumor location, organ movement, and morphological changes; S52: Predict the patient's physiological dynamics during treatment through a multi-layer AI model, and adjust key parameters in the treatment plan accordingly; S53. Verify the adjusted key parameters.

8. The working method of the full-process intelligent radiotherapy system based on Agent AI according to claim 2 is characterized in that: The specific process of step S6 is as follows: S61, the treatment agent controls the radiotherapy equipment to perform radiotherapy operations according to the optimized radiotherapy plan; S62. Real-time collection of patient status parameters during treatment to adaptively adjust radiotherapy plans, and detection and early warning of abnormalities during treatment through deep learning algorithms.

9. The working method of the full-process intelligent radiotherapy system based on Agent AI according to claim 2, characterized in that: The specific process of step S7 is: after the treatment is completed, the patient's image data is collected, and an effect evaluation report is obtained through analysis and processing, which is used for the adjustment and update of the subsequent treatment plan.

10. The working method of the full-process intelligent radiotherapy system based on Agent AI according to claim 2, characterized in that: The specific process of step S8 is as follows: S81. Confirm the status of radiotherapy equipment before treatment begins and review the compliance of the treatment plan; S82. Real-time monitoring of actual dose and key safety parameters during treatment; S83. After the treatment is completed, the treatment effect is evaluated and a quality assurance report is generated.

Images