Radiotherapy plan re-optimization method and system based on target inheritance and flow reset

Abstract

The invention discloses a radiotherapy plan re-optimization method and system based on target inheritance and flow reset, and relates to the technical field of medical physics and radiotherapy, and the method comprises the steps: obtaining an initial radiotherapy plan for a patient in a treatment plan system; and extracting a patient specific target set from the initial radiotherapy plan to form a configuration file independent of the initial radiotherapy plan, and forbidding modification of the content of the configuration file in subsequent optimization to complete target freezing and obtain a frozen target set. According to the method, through a strategy of combining target inheritance with freezing and photon flow resetting, an optimizer can get rid of the constraint of historical parameters of an initial plan under the completely same clinical intention, and global search is restarted from a blank state. According to the method, local optimal locking caused by path dependence in traditional reverse optimization is fundamentally broken through, the endangered organ stress can be systematically and remarkably reduced on the premise that the target dose reaches the standard, and deterministic improvement of plan quality is achieved.

Description

Technical Field

[0001] This invention relates to the fields of medical physics and radiotherapy technology, specifically to a method and system for re-optimizing radiotherapy plans based on target inheritance and flow reset. Background Technology

[0002] Radiation therapy planning, particularly volumetric intensity-modulated rotational radiotherapy (VMAT) and intensity-modulated radiotherapy (IMRT), relies on solving a highly non-convex, multi-peak inverse optimization problem. The solution process exhibits significant path dependence: the optimization algorithm is extremely sensitive to the initial photon flux distribution and iteration history. Once preliminary deliverable parameters (such as the number of monitoring units and the position of the multi-leaf collimator blades) are established, subsequent searches are easily confined to the neighborhood of this local solution, making escape difficult. This results in clinically acceptable plans often being only locally optimal solutions, even under the same patient anatomy, prescription, and optimization objectives. Their dosimetric quality, especially protection of organs at risk, may have untapped potential for improvement.

[0003] Currently, whether it's manual weight tuning based on experience or automated planning based on knowledge bases or deep learning, the underlying principle is the same: adjusting the optimization objective to find new solutions. However, this approach has inherent flaws: first, it alters the clinically validated original optimization intent, potentially introducing uncertainty and deviation; second, and more importantly, it doesn't address the fundamental problem of path dependence. Adjusting the objective weights within the attraction domain of existing local optima makes it difficult to guide the optimizer to achieve global escape and reconstruction, easily leading to bottlenecks in optimization performance.

[0004] Therefore, there is an urgent need in the field for a re-optimization method specifically designed to break path dependence caused by historical deliverable parameters without altering any established clinical objectives. Existing technologies lack a platform-neutral, general-purpose solution that directly intervenes in the optimization process, systematically mining the solution space potential under fixed objectives by resetting history and restarting the search. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for re-optimizing radiotherapy plans based on target inheritance and flow reset, in order to address the shortcomings of the prior art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a radiotherapy planning re-optimization method based on target inheritance and flow reset, comprising: S1. Obtain the initial radiotherapy plan for the patient from the treatment planning system; S2. Extract the patient-specific target set from the initial radiotherapy plan to form a configuration file independent of the initial radiotherapy plan, and prohibit modification of the contents of the configuration file in subsequent optimizations to complete the target freezing and obtain the frozen target set; S3. Create a replica plan based on the initial radiotherapy plan, and perform a photon flow reset operation on the replica plan to make it in a blank deliverable state; S4. On the replication plan, call the reverse optimizer and load the frozen target set to perform re-optimization and generate a re-optimized soft plan; S5. Process the re-optimized soft plan to generate a deliverable re-optimized radiotherapy plan; S6. Compare the optimized radiotherapy plan with the initial radiotherapy plan, and output a recommended plan based on preset conditions.

[0007] In a preferred embodiment, the initial radiotherapy plan in S1 includes at least: The patient's imaging and structural information, field or arc geometry parameters, prescription dosage information, and the patient-specific target set used by the inverse optimizer; The patient-specific target set defines multiple target functions that are associated with anatomical structures and include dose or volume constraint parameters.

[0008] In a preferred embodiment, the replicated plan in S3 inherits all the imaging, structural, geometric, and prescription settings of the initial radiotherapy plan; The photon flow reset operation includes: clearing the existing monitoring unit allocation on all control points of the field or arc, clearing the existing photon flow intensity distribution of all fields or arcs, and resetting the position of the multi-leaf collimator blades. The blank deliverable state refers to the state in which, after the clearing and resetting operations are completed, the replicated plan does not contain any executable irradiation parameters from the initial radiotherapy plan.

[0009] In a preferred embodiment, S4 specifically includes: Load the reverse optimizer of the same type as the initial radiotherapy plan onto the copy plan which is in a blank deliverable state; The frozen target set is provided as the sole optimization input to the inverse optimizer, driving it to recalculate the photon flow distribution and control point parameters from scratch.

[0010] In a preferred embodiment, step S5 specifically includes: performing dose calculation and dose normalization on the re-optimized soft plan, and solving for the final gantry parameters, dose rate, and multi-leaf collimator blade motion sequence based on the calculation results to form the deliverable re-optimized radiotherapy plan.

[0011] In a preferred embodiment, the preset conditions in S6 specifically include: The target dose coverage index of the re-optimized radiotherapy plan is not lower than that of the initial radiotherapy plan, and it is superior to the initial radiotherapy plan in terms of dose index of at least one organ at risk.

[0012] In a preferred embodiment, the step further includes the following step after S1 and before S2: Assess whether the initial radiotherapy plan meets all pre-set clinical dose constraints, and mark cases that do not meet all constraints as priority cases for further optimization; S2 to S6 are performed preferentially on the preferential re-optimization cases.

[0013] In a preferred embodiment, the method includes independently repeating S3 to S5 at least twice for the same case to obtain multiple re-optimized soft plans; In S5, the median values ​​of the multiple re-optimized soft plans on key dosimetric indicators are taken to generate the final deliverable re-optimized radiotherapy plan.

[0014] In a preferred embodiment, the method further includes the following step before step S6: The planning complexity of the re-optimized radiotherapy plan is evaluated, and the planning complexity indicators include the total number of monitoring units or the total travel of the multi-leaf collimator blades; The subsequent comparison and output steps are performed only when the change in the complexity of the re-optimized radiotherapy plan is within a preset threshold.

[0015] This invention also provides a radiotherapy planning re-optimization system based on target inheritance and flow reset, comprising: The data interface module is used to read initial plan data from the treatment planning system; The target freezing module is used to extract and lock the target set in the initial plan and generate a frozen target set file; The plan reset module is used to create a copy plan and perform a photon flow reset operation, clearing all deliverable irradiation parameters; The optimization driver module is used to load the frozen target set file onto the reset replication plan and drive the built-in optimizer of the treatment plan system to perform re-optimization; The plan generation module is used to calculate the dosage and solve for executable parameters based on the re-optimization results. The evaluation and decision-making module is used to compare the dosimetric indicators of the initial and re-optimized plans and output a recommended plan based on preset rules.

[0016] The technical effects and advantages provided by the present invention in the above technical solution are as follows: This invention does not rely on any external prediction model. Its implementation is entirely based on the inverse optimizer and dose calculation engine embedded in commercial treatment planning systems. It can be called through standardized scripts or application programming interfaces, thereby achieving seamless integration with treatment planning system platforms from different manufacturers.

[0017] This invention employs a strategy combining target inheritance with freezing and photon flow reset, enabling the optimizer to break free from the constraints of initial plan history parameters and restart the global search from a blank state under identical clinical intent. This fundamentally breaks the local optimum locking caused by path dependence in traditional inverse optimization. Experimental data shows that it can systematically and significantly reduce the dose to organs at risk while maintaining target dose compliance, achieving a deterministic improvement in plan quality.

[0018] This invention, through its automated post-processing module, can be directly embedded into existing clinical workflows without altering the initial planning process. It achieves significant improvements in plan quality with limited computational cost, making it particularly effective as a rescue tool for challenging cases. Furthermore, by integrating mechanisms such as multiple optimizations to obtain the median value and plan complexity verification, the stability and executability of the output results are further ensured, giving it significant clinical application value. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0020] Figure 1 This is a flowchart of the method of the present invention.

[0021] Figure 2 This is a system block diagram of the present invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Example 1, please refer to Figure 1 As shown in this embodiment, the radiotherapy planning re-optimization method based on target inheritance and flow reset includes: S1. Obtain the initial radiotherapy plan for the patient from the treatment planning system; S2. Extract the patient-specific target set from the initial radiotherapy plan to form a configuration file independent of the initial radiotherapy plan, and prohibit modification of the contents of the configuration file in subsequent optimizations to complete the target freezing and obtain the frozen target set; S3. Create a replica plan based on the initial radiotherapy plan, and perform a photon flow reset operation on the replica plan to make it in a blank deliverable state; S4. On the replication plan, call the reverse optimizer and load the frozen target set to perform re-optimization and generate a re-optimized soft plan; S5. Process the re-optimized soft plan to generate a deliverable re-optimized radiotherapy plan; S6. Compare the optimized radiotherapy plan with the initial radiotherapy plan, and output a recommended plan based on preset conditions.

[0024] In one possible implementation, step S1 specifically includes: S11. Interface connection and data location: The system establishes a connection through the application programming interface (API) or dedicated scripting interface provided by the treatment planning system (such as the scripting API of Varian Eclipse or the Python scripting API of RayStation). Users specify the patient ID and plan name (e.g., patient Zhang San, plan nasopharyngeal carcinoma VMAT protocol 1) via a graphical interface or command line, and the system uses this information to search the treatment planning system database and locate the target plan.

[0025] S12. Structured Data Reading and Parsing: The system reads and parses the complete data packet of the initial plan, which is typically stored internally in a structured format (such as the DICOM-RT standard for medical digital imaging and radiotherapy, or a vendor-specific format). At least the following core elements are extracted: Patient's imaging and structural information: Imaging: Reading the patient's localized CT image sequence, including pixel data, spatial resolution (e.g., pixel pitch 0.9766mm × 0.9766mm), slice thickness (e.g., 3.0mm), and coordinate system information. For example, the CT sequence contains 120 slices for accurate dose distribution calculation.

[0026] Structure Set: Reads all drawn outline structures, including their names, 3D coordinate point sets, and structure types. Key structures include: Target area: Planned target area (Prescription 70 Gy), irradiation prevention area (Prescription 60 Gy).

[0027] Organs at risk: spinal cord, brainstem, left parotid gland, right parotid gland, left lens, etc.

[0028] Field or arc geometry parameters: For the volumetric intensity-modulated rotational radiotherapy plan in this embodiment, the geometric definitions of two complete arcs are read: Arc 1: Frame rotation start angle = 181°, end angle = 179°, rotation direction = clockwise, collimator angle = 5°, bed angle = 0°.

[0029] Arc 2: Frame rotation start angle = 179°, end angle = 181°, rotation direction = counterclockwise, collimator angle = 355°, bed angle = 0°.

[0030] Simultaneously read the settings of the dose calculation grid, such as grid resolution = 2.5mm.

[0031] Prescription dosage information: Read the prescription associated with the plan, such as the planned target area. The prescribed dose is 70 Gy, to be administered in 33 divided doses.

[0032] Patient-specific target set used by the inverse optimizer: The system uses an API to gain deep access to the optimization engine's configuration layer, extracting all the objective functions and their parameters actually used in the reverse optimization process to generate the plan. These parameters define the clinical intent of the optimization.

[0033] For example, the extracted target set might contain the following entries (example): Structure: Planned target area Type: Minimum dose, Dose value: 70 Gy, Weight: 100; Structure: Planned target area Type: Uniform dose, Dose value: 70 Gy, Weight: 50; Structure: Spinal cord, Type: Maximum dose, Dose value: 45 Gy, Weight: 80; Structure: Left parotid gland; Type: Mean dose; Dose value: <26 Gy; Weight: 70; Structure: Left lens, Type: Maximum dose, Dose value: 8 Gy, Weight: 60; (...the rest of the objective functions are omitted); This set of objectives fully defines all dosimetric goals that the program aims to pursue and avoid in its optimization process.

[0034] S13. Data encapsulation and verification: All the extracted information is encapsulated within the system into a structured initial plan object, and a consistency check is performed (e.g., confirming that all referenced structure names exist in the structure set and that the field parameters are complete). Once the check passes, this object becomes the input basis for subsequent steps, marking the completion of step S1.

[0035] In one possible implementation, step S2 specifically includes: S21. Deep extraction of target set structure data: The system calls the target freeze module to perform a deep parsing of the initial plan object obtained in S1. Instead of simply reading surface parameters, the module accesses and extracts the complete, original set of objective functions used by the inverse optimizer to generate the plan through the treatment plan system's optimization engine interface. Each objective function is parsed into a series of machine-readable key-value pair attributes, including at least: Related structures: such as planned target area ,spinal cord.

[0036] Target type: such as minimum dose, maximum dose, average dose, dose-volume (e.g., 30% volume dose <20Gy).

[0037] Constraint parameters include dose values ​​(e.g., 70, in Gy), volume percentages (e.g., 30%), and inequality directions (e.g., ≦, ≧).

[0038] Optimization weight: Represents the relative importance of the objective in the global optimization (e.g., 100).

[0039] Priority (if supported by the system): A hierarchical parameter used to handle target conflicts.

[0040] For example, in the case of nasopharyngeal carcinoma, a typical objective function entry extracted can be expressed as: {Structure: Left parotid gland, Type: Mean dose, Parameter: <26, Weight: 70}; S22. Generate a separate configuration file: All extracted objective function entries are serialized according to a preset, standardized data format (such as JSON, XML, or YAML) to generate a single physical file. This file is completely separate from the specific treatment plan system project file and patient database.

[0041] S23. Implement a freezing mechanism: To prevent modification, the system employs software-level mandatory constraints: Access control: Once the configuration file is generated, immediately set its operating system attributes to read-only.

[0042] Memory Loading and Verification: In subsequent optimization processes, when the target set needs to be loaded, the system reads this configuration file into a protected area in memory. Before and after loading, the hash value of the file (such as the result of MD5 hash algorithm or SHA-256 secure hash algorithm) is calculated and compared to ensure that the content has not been tampered with during transmission and storage.

[0043] Interface isolation: When invoking the reverse optimizer, the system injects the frozen target set data directly into the optimization engine through a wrapped interface, bypassing and disabling the channels in the treatment plan system's graphical interface that typically allow users to interactively adjust target weights and parameters. The optimizer's input is strictly limited to this dataset.

[0044] S24. Form a frozen target set: Through the above steps, the original objective function configurations, which may have been scattered throughout the treatment planning system, are transformed and solidified into a separate, content-protected frozen objective set. This set fully represents the clinical optimization intent of the initial plan and serves as the sole, immutable input criterion for the inverse optimizer in subsequent step S4, thereby ensuring that the re-optimization process only resets the solution path while fully inheriting the original clinical objectives.

[0045] In one possible implementation, step S3 specifically includes: S31. Create a deep replication plan: The system calls the plan reset module to perform a deep copy operation through the treatment plan system's application programming interface.

[0046] Operation instructions: The module sends instructions to the treatment planning system, such as creating a plan copy (source plan identifier = original plan, new plan name = re-optimization and cleanup plan).

[0047] Copy Content: The new plan (re-optimized cleanup plan) generated by this operation exists as a completely independent plan object within the treatment plan system. It inherits all the static settings of the initial plan, including: Patient CT images and the outlines of all associated anatomical structures (such as the planning target area and organs at risk).

[0048] All geometric parameters of the irradiation technique (such as the gantry start and end angles, collimator angle, bed angle, and dose calculation grid for the two complete arcs of volumetric intensity-modulated rotational radiotherapy).

[0049] Prescription dosage information.

[0050] Exclusions: Meanwhile, this replication plan does not inherit any dynamic parameters related to specific optimization results and machine execution, preparing for subsequent resets.

[0051] S32, Perform photon stream reset operation: After the replication plan is created, the plan reset module immediately performs a series of atomic operations on it to clear all deliverable historical parameters: Clear the assigned number of monitoring units: Operation: Traverse every control point in all shooting fields or arcs in the plan (e.g., 178 control points in two arcs), and force the number of monitored units attribute to 0 by calling the application programming interface.

[0052] Example: For example, the number of monitoring units for control point #1 (arc 1, rack angle 181°) is set from the original 12.5 to 0; For example, the number of monitoring units for control point #2 is set from 10.8 to 0, until the number of monitoring units for all control points is reduced to zero.

[0053] Clearing photon flow / intensity distribution: Operation: For the photon flux intensity map associated with each control point, call the function set to uniform flux or a similar function to reset it to a uniform, unmodulated intensity distribution (e.g., all pixels have a value of 1 or 0).

[0054] Technical essence: This is equivalent to clearing the initial intensity modulation shape of each field direction and restoring it to a blank state.

[0055] Reset the position of the multi-leaf collimator blades: Operation: For each control point, call the reset multi-leaf collimator position function to reset all blade positions to fully open (i.e., the blades are at the maximum allowable opening, such as from -200mm to +200mm) or a neutral, undefined initial state.

[0056] Example: At control point #1, the originally complex sequence of blade concave and convex shapes (such as blade A: 12.5mm, blade B: -5.3mm...) is uniformly set to the maximum opening position (blade A: 150.0mm, blade B: -150.0mm).

[0057] S33. Verify and achieve a blank deliverable state: After completing the above reset operation, the system performs an internal verification of the replication plan.

[0058] State Definition Verification: The so-called blank deliverable state is technically defined here as follows: in the program object, there are no non-zero monitoring unit values, no non-uniform photon flux intensity modulation, and the multi-leaf collimator does not form any meaningful obstruction shape. Its dose calculation result will be 0 or a meaningless uniform low dose distribution.

[0059] Verification Example: The system reads the re-optimization cleanup plan, confirms that its total number of monitored units is 0, and cannot generate a clinically meaningful dose distribution via the dose calculation button in the treatment plan system (the system may indicate that the intensity map is invalid). In this case, the plan cannot be physically executed by the linear accelerator because it does not contain any valid control commands.

[0060] In one possible implementation, step S4 specifically includes: S41. Environment Initialization and Optimizer Configuration: In a copy plan (re-optimization cleanup plan) environment that is in a blank deliverable state, the system calls the reverse optimizer engine of the exact same type and version as the initial plan (e.g., the photon optimizer version 15.6 of the Varian Eclipse system) through the optimization driver module.

[0061] Key configuration: The module resets all internal states and caches of the optimizer and explicitly sets its initial solution parameters to start from zero or no initial flux, ensuring that the optimizer does not attempt to find or reuse any implicit initial solutions.

[0062] S42, Force loading a frozen target set as the sole input: The optimization driver module maps and injects the contents of the frozen target set configuration file (re-optimize_frozen target.json) generated in step S2 into the target function settings of the optimizer via API.

[0063] Write and Lock: The system executes a batch setting command, writing each objective function entry (e.g., structure: spinal cord, type: maximum dose, dose value: 45.0 Gy, weight: 80) from the configuration file one by one and accurately into the optimizer's objective function list. This process bypasses the graphical user interface and disables the automatic weight adjustment or user-interactive modification functions that the optimizer typically allows during iteration. The optimizer's objective function domain is strictly limited to the range defined by the frozen objective set.

[0064] S43. Perform iterative optimization from scratch: Once the configuration is complete, the system sends a command to the optimizer to begin iteration.

[0065] Iterative process: The optimizer, based on its built-in algorithm (such as gradient-based sequential quadratic programming), starts from an initial state with zero monitoring units, fully open multi-leaf collimators, and a uniform intensity map. It attempts to adjust the photon flow distribution at each control point (i.e., the weights of each pixel in the intensity map) to minimize the global cost function defined by the frozen target set. For example, it may attempt to increase the coverage of the planned target area. The flux to the region is set to meet the minimum dose constraint, while flux toward the spinal cord is strictly suppressed to comply with the maximum dose constraint.

[0066] Process monitoring: In this embodiment, the optimization process lasted for approximately 1520 iterations. The system recorded the decline curve of the objective function value and the satisfaction of key constraints in real time. For example, after approximately 800 iterations, the planned target area... The dose coverage target has been largely met, and subsequent iterations will focus on further reducing the average dose to the left and right parotid glands.

[0067] S44, Generate a re-optimized soft plan: The iteration automatically terminates when the optimizer reaches the preset convergence criterion (e.g., the objective function value improves by less than 0.1% after 50 consecutive iterations).

[0068] Output: The optimizer outputs the final optimization result, which is a complete new set of control point parameters, including: The new non-zero monitoring unit value for each control point (for example, the new monitoring unit number for control point 1 is 10.2, which is different from the initial plan of 12.5).

[0069] The photon flow intensity map corresponding to each control point is newly modulated (showing a modulation pattern that is different from the initial plan and is newly created to meet the freezing target).

[0070] A preliminary sequence of multi-leaf collimator blade positions matched to the new photon flow.

[0071] State Definition: The plan generated at this stage is called the re-optimized soft plan. It contains all the necessary intensity modulation information, but has not yet undergone precise dose deposition calculations by the final dose calculation engine, nor has it been transformed into a machine-executable sequence of control points that takes into account practical mechanical constraints. It is a mathematical solution to an optimization problem, not an executable physical plan.

[0072] In one possible implementation, step S5 specifically includes: S51. Perform high-precision dose calculation: The system invokes the dose calculation and segmentation module to initiate a completely new and full three-dimensional dose calculation in the replicated plan environment (the re-optimized cleanup plan now includes photon flow data from the soft plan).

[0073] Engine and parameter consistency: To ensure comparable results, the dose calculation algorithm and parameter configuration must be exactly the same as the initial plan. For example, specify the use of the AcurosXB algorithm (version 15.6), set the dose calculation grid resolution to 2.5 mm, and strictly align it with the coordinate system of the planned CT.

[0074] Calculation Process: The dose calculation engine reads the new photon flux intensity distribution at each control point in the re-optimized soft plan. Based on these distributions and the corresponding preliminary positions of the multi-leaf collimators, it simulates the transmission, scattering, and energy deposition processes of photons as they pass through heterogeneous human tissue described in the patient's CT images. This process is computationally intensive; in this embodiment, the calculation of the dual-arc volumetric modulated arc radiotherapy plan takes approximately 8 minutes. Finally, a three-dimensional dose matrix covering all CT slices and containing the absorbed dose value (in Gy) for each voxel is generated.

[0075] S52. Implement clinical dose normalization: The calculated original dose distribution needs to be adjusted to meet clinical prescribing standards.

[0076] Normalization reference: The system selects a key target region as the normalization reference based on a preset clinical protocol. In this embodiment, the planned target region is selected. (Prescription 70Gy) as the normalized structure.

[0077] Normalization: The system calculates the D95 (the lowest dose received by 95% of the target volume) for that region. Assuming the calculated original D95 is 68.5 Gy, the system calculates a normalization factor: 70.0 Gy / 68.5 Gy ≈ 1.0219. Subsequently, all voxel dose values ​​in the entire three-dimensional dose matrix are multiplied by this factor.

[0078] Verification: After normalization, the system immediately verifies the planned target area. To ensure that the D95 accurately reaches 70.0 Gy, and at the same time check other target areas (such as the planned target area)... This step ensures that the re-optimized plan is at the exact same absolute dose level as the initial plan, making subsequent dosimetric comparisons fair.

[0079] S53. Generate an executable control sequence: This is a crucial step in translating dose distribution into machine instructions.

[0080] Call the sequence generation engine: The system calls the sequence generation or leaf sequence optimization engine built into the treatment plan system (such as the post-processing stage of the progressive resolution optimizer 3 in the Eclipse system, or a separate photon stream to leaf sequence converter).

[0081] Inputs and Constraints: The engine's inputs are the normalized, precise dose distribution and the photon flow in the re-optimized soft plan. Simultaneously, the system loads the physical and mechanical constraint parameters of the linear accelerator (such as the Varian TrueBeam linear accelerator), including but not limited to: the maximum speed of the multi-leaf collimator blades (e.g., 2.5 cm / s), the limit on the position difference between adjacent blades, the dose rate variation range, and the gantry rotation speed limit.

[0082] Solution and Output: Under the premise of satisfying all machine hard constraints, the engine solves for a set of executable parameters that most accurately reproduces the target dose distribution at delivery. Its output is a complete deliverable plan containing the following information: Control point sequence: final rack angle, dose rate, and cumulative number of monitored units for each control point.

[0083] Multi-leaf collimator blade sequence: The precise and smooth motion trajectory position of each pair of multi-leaf collimator blades at each control point (accuracy typically down to 0.1 mm), ensuring that the blade movement speed is within physical limits.

[0084] Overall plan parameters: total number of jumps planned (Total number of monitoring units, for example, 595 monitoring units in this embodiment), estimated treatment time, etc.

[0085] S54, Final Verification and Encapsulation: The system automatically verifies the generated deliverable plan, checking for any parameters exceeding machine safety limits (such as blade collision risk or dose rate exceeding limits), and confirming the consistency of its dose distribution with the optimization target set. Once verification is successful, the plan is saved in the treatment planning system as a formal, parallel to the initial plan, re-optimized radiotherapy plan available for clinical review, transmission to the treatment machine, and final execution; for example, it may be named the final re-optimized plan.

[0086] In one possible implementation, step S6 specifically includes: S61. Automatically calculate key dosimetric assessment indicators: The system invokes the evaluation and decision-making module to automatically perform a comprehensive dosimetric analysis on the original plan and the final re-optimization plan.

[0087] Target coverage and uniformity indicators: The system calculates the dose-volume histogram for each major target region and extracts: D98 (dose covering 98% of volume): Used to assess the robustness of dose coverage. For example, for planning the target volume. The initial plan for D98 was 69.2 Gy, and the optimized plan was 69.5 Gy.

[0088] D2 (dose covering 2% of volume): Used to assess high-dose (hotspot) levels. For example, the planned target area. The initial plan for D2 was 73.8 Gy, and the optimized plan was 73.5 Gy.

[0089] Uniformity index: Calculated by (D2-D98) / D50, the smaller the value, the more uniform the uniformity.

[0090] At-risk organ dose indices: The system calculates the key dose-volume parameters for each at-risk organ. Average dose It is crucial for parallel organs (such as the parotid gland and lungs). For example, the left parotid gland... The initial plan was 28.5 Gy, and the optimized plan was 25.1 Gy.

[0091] Maximum dose It is crucial for sequential organs such as the spinal cord and brainstem. For example, the spinal cord... The initial plan was 43.2 Gy, and the optimized plan was 42.7 Gy.

[0092] Volumetric dose VxGy (volume receiving a dose of xGy or more): e.g., V30Gy for the left parotid gland (percentage of volume receiving more than 30Gy of irradiation).

[0093] As a comprehensive evaluation, the system calculates plan quality metrics. PQM . PQM It is a scalar score ranging from 0 to 100, with higher scores indicating better overall plan quality. The calculation method is as follows: Indicator selection and normalization: A set of core dosimetric indicators were selected, including target coverage index (e.g., D98 of the planned target area), target homogeneity index, and dose indicators of key organs at risk (e.g., average dose to the parotid gland). Maximum dose of spinal cord Each indicator is assigned a target value based on its clinical importance.

[0094] Fraction calculation: PQM Calculated using the following formula: PQM = 100； in: This represents the actual value planned for this indicator.

[0095] This is the clinical target value for this indicator.

[0096] This is the penalty tolerance value for this indicator (i.e., the maximum range of deviation from the target that is allowed).

[0097] The normalized weights of this indicator (satisfying) =1).

[0098] Example: For a nasopharyngeal carcinoma treatment plan, assume only three metrics are considered: the planned target volume PTV_D98 (target value T=70Gy, tolerance value P=2Gy, weight w=0.4), and the average dose to the left parotid gland. (Target value T = 26 Gy, tolerance value P = 4 Gy, weighting value w = 0.3), maximum spinal cord dose (Target value T = 45 Gy, tolerance value P = 2 Gy, weight value w = 0.3). If the actual values ​​of a certain plan are 70.2 Gy, 25.1 Gy, and 42.7 Gy respectively, then its PQM The calculation is as follows: [0.4 (1-|70.2-70| / 2)+0.3 (1-|25.1-26| / 4)+0.3 (1-|42.7-45| / 2)] 100 = ≈ 66.75 points; S62. Comparison and judgment based on preset conditions: The evaluation module compares the calculated index values ​​pairwise and automatically makes judgments based on strict preset conditions programmed into the system. These preset conditions typically include two levels: Mandatory non-inferiority condition (for target areas): All key performance indicators of major target areas in the re-optimization plan must not be worse than those of the initial plan beyond the allowable tolerance range.

[0099] Example rule: D98 must not decrease by more than 1% (0.7 Gy), and D2 must not increase by more than 1% (0.7 Gy).

[0100] This example determines the target area of ​​the re-optimization plan. D98 (69.5 Gy) is better than the initial plan (69.2 Gy), and D2 (73.5 Gy) is better than the initial plan (73.8 Gy). The non-inferiority condition is met.

[0101] Optimize benefit criteria (for organs at risk): The re-optimization program should demonstrate clinically meaningful improvement in at least one key indicator for organs at risk.

[0102] Example rule: Average dose to any organ at risk Reduce by ≥2 Gy, or the maximum dose Reduce by ≥1 Gy.

[0103] This example determines the average dose of the left parotid gland. The mean dose to the right parotid gland was reduced by 3.4 Gy (28.5 → 25.1 Gy). The maximum dose to the spinal cord was reduced by 2.5 Gy (26.8 → 24.3 Gy). The reduction was 0.5 Gy. The improvement in both the left and right parotid glands significantly exceeded the threshold, meeting the conditions for optimized benefit.

[0104] S63. Comprehensive Evaluation and Decision Output: Decision logic: The system determines the re-optimization plan as the recommended plan only when both the non-inferiority condition and the optimization benefit condition are met simultaneously.

[0105] Execution Action: In this embodiment, since all conditions are met, the evaluation and decision-making module performs the following operations: Marking the plan: Within the treatment planning system, the final re-optimization plan is marked as recommended, and key improvement summaries (such as a 3.4 Gy reduction in the average dose to the left parotid gland) are automatically recorded in the plan description.

[0106] Report generation: Automatically generate a structured comparison report, which visually displays the differences between the initial plan and the recommended plan across all indicators in the form of tables and dose-volume histogram overlays.

[0107] Output and Prompt: A clear prompt appears on the system interface for the user (physicist): Optimization is complete and a recommended plan has been generated. Key Improvement: Average parotid dose significantly reduced. Final clinical review recommended.

[0108] S64, Complexity Verification: Before the final output, the system will perform the steps defined above to verify that the complexity of the recommended plan is within clinically acceptable limits.

[0109] Calculation metrics: For example, the total number of monitoring units in the final re-optimization plan is calculated to be 595, an increase of 2.6% compared to the initial plan of 580 monitoring units.

[0110] Threshold judgment: The increase is lower than the system's preset complexity change threshold (e.g., 10%). Therefore, the system confirms that the plan poses no additional risk to its feasibility and maintains its recommended status.

[0111] The preset threshold (e.g., ±10% change in total monitor units) is not arbitrarily set, but determined based on a statistical analysis of the complexity indicators of over 50 historical high-quality volumetric intensity-modulated radiotherapy (IMRT) plans at our institution. The analysis shows that over 95% of clinically approved plans have natural fluctuations in the total monitor units or total multi-leaf collimator blade travel between different feasible plans for the same case that are less than this threshold. Therefore, this threshold can effectively distinguish between normal physical variability of the plan and abnormal increases in complexity that may affect execution reliability or treatment efficiency, ensuring that recommended re-optimized plans improve dosimetry without sacrificing clinical feasibility.

[0112] Example 2: Further optimization of difficult cases and enhanced outcome stability; This embodiment demonstrates the application of the method of the present invention to difficult cases and shows the specific operation of improving the stability of the results by optimizing the median value multiple times.

[0113] The system read an initial plan for volumetric intensity-modulated rotational radiotherapy (IMRT) for prostate cancer. Assessment revealed that the rectal V65Gy (rectal volume receiving doses of 65Gy or higher) was 18%, exceeding the institution's clinical constraint limit (15%). Therefore, this case was marked as a priority for re-optimization. Subsequently, the system automatically executed steps S2 to S6 as described in claim 1. After freezing all initial target sets and resetting the photon flow, the re-optimized plan reduced the rectal V65Gy to 13.5%, successfully meeting clinical constraints, while maintaining non-inferiority in target dose (planned target D98) (change <0.5%). This case demonstrates the value of this invention as a clinical rescue tool.

[0114] To further improve the robustness of optimization results for complex cases or organs that are extremely sensitive to dose (such as the rectum in this case), the system executed the methods described above: After completing S2 (target freezing), the system independently and repeatedly executed steps S3 (copying and resetting) to S5 (plan generation) three times based on the same initial plan. Each execution started from a completely blank deliverable state, and the optimizer's random number seed was different, thus generating three re-optimized soft plans with slight differences in dose distribution and subsequent deliverable plans.

[0115] The system reads the rectal V65Gy values ​​from these three plans, which are 13.5%, 12.8%, and 14.1%, respectively. According to the preset rules, the median value of 13.5% is taken as the indicator value for the final output plan.

[0116] The system selects the re-optimization plan corresponding to the median value (13.5%) as the final re-optimization plan for output and comparison.

[0117] This method effectively smooths out output fluctuations that may occur in a single optimization due to random initial conditions, ensuring the stability and reliability of the recommended plan on key dose indicators, avoiding the output of occasional poor results, and improving the clinical applicability of the method.

[0118] Example 3, please refer to Figure 2 As shown, the radiotherapy planning re-optimization system based on target inheritance and flow reset described in this embodiment includes: The data interface module is used to read initial plan data from the treatment planning system; This module acts as a bridge between the system and commercial treatment planning systems (such as Varian Eclipse). It locates and reads the specified initial radiotherapy plan by calling specific application programming interface functions provided by the treatment planning system (e.g., methods such as Open Plan and Get Structure Set using the scripting application programming interface in Eclipse).

[0119] This module can parse complex, structured data objects within the treatment planning system, accurately extracting four core categories of data: patient imaging and structural information (CT sequences, contours), field / arc geometry parameters, prescription dosage information, and, most importantly, the raw parameters of the patient-specific target set used by the inverse optimizer. This data is converted into a unified intermediate representation within the system, laying the foundation for subsequent processing.

[0120] The target freezing module is used to extract and lock the target set in the initial plan and generate a frozen target set file; This module is responsible for implementing the logic of target inheritance. It receives the raw target set data extracted by the data interface module and performs in-depth analysis and reconstruction.

[0121] This module iterates through each objective function, serializing its structural relationships, objective type (such as minimum, maximum, and average dose), dose / volume constraints, weights, and other parameters into a separate standard format file (such as a JSON-formatted frozen objective file). After file generation, the module immediately sets its attributes to read-only via operating system commands. In all subsequent processes, when the objective set needs to be called, the module calculates the hash value of the file for integrity verification and directly and in batches injects it into the optimizer through a encapsulated interface. Simultaneously, it hides the controls in the treatment planning system's graphical interface used to modify objectives, thus achieving software-level freezing.

[0122] The plan reset module is used to create a copy plan and perform a photon flow reset operation, clearing all deliverable irradiation parameters; This module is responsible for performing the physical operation of resetting the photon stream. It first creates a complete copy of the initial plan (the new plan object) within the treatment planning system, and then zeroes the parameters of that copy.

[0123] This module is invoked via an application programming interface (API). First, it creates a replication plan. Then, it performs three atomic operations on the replication plan: (1) Traverse all control points and use the function to set the number of monitoring units to zero; (2) Call the function to reset flux to uniformity to reset the photon flux intensity map of each control point to a uniform distribution; (3) Use the function to set the multi-leaf collimator position to open to reset all multi-leaf collimator blades to the fully open position. After the operation is completed, the module performs a status verification to ensure that the total number of planned monitoring units is 0 and no meaningful dose can be calculated, thereby confirming that it has reached the blank deliverable state.

[0124] The optimization driver module is used to load the frozen target set file onto the reset replication plan and drive the built-in optimizer of the treatment plan system to perform re-optimization; This module is the engine that initiates a new round of optimization search. It configures and drives the treatment plan system's built-in inverse optimizer to solve the problem within a blank replication plan environment.

[0125] The key operations performed by this module include: initializing the optimizer instance and setting its initial conditions to start from zero. Then, it reads the configuration file generated by the target freezing module, forces the frozen target set as the sole optimization target to be loaded into the optimizer by functions such as loading the optimization objective function, and disables dynamic weight adjustment during the optimization process. Finally, it calls the start optimization method to initiate iterative computation. Throughout the optimization process, the module monitors the iteration progress and convergence status until it generates the final re-optimization soft plan (i.e., a plan that includes the new photon flux distribution but does not perform final dose calculation and serialization).

[0126] The plan generation module is used to calculate the dosage and solve for executable parameters based on the re-optimization results. This module is responsible for transforming the optimized mathematical solution into a clinically usable physical plan. It comprises two core subprocesses: dose calculation and sequence generation.

[0127] This module invokes the same dose calculation engine (such as the AcurosXB algorithm) as the initial plan, using the same grid settings, to perform a complete three-dimensional dose calculation on the re-optimized soft plan, obtaining the original dose distribution. Subsequently, dose normalization is automatically performed according to the clinical prescription (such as achieving a specified target D95 value).

[0128] The module then calls the sequence generation engine of the treatment planning system (such as the progressive resolution optimizer 3) to combine the normalized dose distribution and photon flow data with the actual physical constraints of the linear accelerator (such as the maximum speed of the multi-leaf collimator and the dose rate limit) to solve a set of final control point sequences that can be accurately executed by the treatment machine. This includes the gantry angle, dose rate, number of monitoring units, and detailed multi-leaf collimator blade positions for each control point.

[0129] The evaluation and decision-making module is used to compare the dosimetric parameters of the initial and re-optimized plans and output a recommended plan based on preset rules. This is the system's quality control and decision-making exit point. It performs an automated and quantitative comparison between the initial plan and the final re-optimized plan, and makes recommendations based on rules.

[0130] The module automatically calculates key dosimetric parameters for both plans (such as D98 and D2 at the target site, and the average dose to organs at risk). Maximum dose The system includes criteria for evaluating the treatment plan's effectiveness (such as the number of monitored units) and the complexity of the plan. It incorporates pre-defined rules for assessment (e.g., non-inferiority of the target dose and improvement in the dose to at least one organ at risk exceeding a threshold). The module matches the results against these rules: if all rules are met, the re-optimized plan is automatically marked as a recommended plan in the treatment planning system, and an evaluation report is generated including a dose-volume histogram comparison and a list of indicators; if not, the reasons are recorded and a message is displayed indicating that the re-optimization did not produce a satisfactory improvement.

[0131] Experimental Example 1: Comparison and verification of the method of this invention with traditional re-optimization strategies; To objectively evaluate the effectiveness of the target inheritance and photon flow reset combination method described in this invention, a comparative experiment was designed.

[0132] Experimental setup: Dataset: Clinical volumetric intensity-modulated rotational radiotherapy plans of 30 nasopharyngeal carcinoma patients admitted to our hospital were selected as the test sample.

[0133] Experimental group: The complete method of this invention is adopted, which combines target freezing with photon flow resetting (denoted as FMRO method).

[0134] Control group A: Simulates the conventional manual optimization approach, only freezing the target, but allows the optimizer to perform iterative fine-tuning based on the initial planned photon flow and control point parameters (denoted as F-only method).

[0135] Control group B: Simulates a possible automated strategy that only resets the photon flow, but does not freeze the objective set during re-optimization. Instead, it uses the same objective function type as the initial plan, but the weights are automatically assigned by the system according to empirical rules (denoted as R-only method).

[0136] Evaluation indicators: The primary assessment focuses on the average dose to the left and right parotid glands. .

[0137] The absolute value of the consistency deviation between the secondary assessment plan and the initial plan in target dose coverage (planned target area D98).

[0138] Record the improvement score of the overall quality indicator (PQM) of the plan.

[0139] Experimental results: Table 1 presents the statistical analysis results (mean ± standard deviation) of the improvement in key indicators:

[0140] Experimental conclusions; Compared to the conventional fine-tuning approach of freezing only the target (F-only method), this invention (FMRO method) achieves significantly superior dosimetric improvements in parotid gland protection. Specifically, the average dose reduction to the parotid gland brought about by this invention is approximately 2.6 Gy higher than that of the F-only method, representing an improvement of approximately 8-12% compared to the F-only method. This demonstrates that simply keeping the target unchanged cannot effectively overcome path dependence; photon flow resetting is a key step in unlocking new search spaces and achieving breakthrough improvements.

[0141] Compared to approaches that merely reset the photon flow but alter the weights (R-only method), this invention (FMRO method) maintains the original clinical significance. Figure 1 It has an absolute advantage in consistency (minimal target deviation) and more stable dose improvement (smaller standard deviation). This demonstrates that target freezing is crucial for maintaining the clinical reliability and predictability of re-optimization plans.

[0142] Target freezing and photon flow reset are both indispensable, and their combination produces a significant synergistic effect. While ensuring the consistency of clinical goals, it achieves systematic and breakthrough optimization of doses to organs at risk, verifying the unique technical value of this invention in solving path dependence problems under fixed goals.

[0143] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A radiotherapy planning re-optimization method based on target inheritance and flow reset, characterized in that, include: S1. Obtain the initial radiotherapy plan for the patient from the treatment planning system; S2. Extract the patient-specific target set from the initial radiotherapy plan to form a configuration file independent of the initial radiotherapy plan, and prohibit modification of the contents of the configuration file in subsequent optimizations to complete the target freezing and obtain the frozen target set; S3. Create a replica plan based on the initial radiotherapy plan, and perform a photon flow reset operation on the replica plan to make it in a blank deliverable state; S4. On the replication plan, call the reverse optimizer and load the frozen target set to perform re-optimization and generate a re-optimized soft plan; S5. Process the re-optimized soft plan to generate a deliverable re-optimized radiotherapy plan; S6. Compare the optimized radiotherapy plan with the initial radiotherapy plan, and output a recommended plan based on preset conditions.

2. The radiotherapy planning re-optimization method based on target inheritance and flow reset according to claim 1, characterized in that: The initial radiotherapy plan in S1 includes at least the following: The patient's imaging and structural information, field or arc geometry parameters, prescription dosage information, and the patient-specific target set used by the inverse optimizer; The patient-specific target set defines multiple target functions that are associated with anatomical structures and include dose or volume constraint parameters.

3. The radiotherapy planning re-optimization method based on target inheritance and flow reset according to claim 1, characterized in that: The replication plan in S3 inherits all the imaging, structural, geometric, and prescription settings of the initial radiotherapy plan; The photon flow reset operation includes: clearing the existing monitoring unit allocation on all control points of the field or arc, clearing the existing photon flow intensity distribution of all fields or arcs, and resetting the position of the multi-leaf collimator blades; The blank deliverable state refers to the state in which, after the clearing and resetting operations are completed, the replicated plan does not contain any executable irradiation parameters from the initial radiotherapy plan.

4. The radiotherapy planning re-optimization method based on target inheritance and flow reset according to claim 1, characterized in that: S4 specifically includes: Load the reverse optimizer of the same type as the initial radiotherapy plan onto the copy plan which is in a blank deliverable state; The frozen target set is provided as the sole optimization input to the inverse optimizer, driving it to recalculate the photon flow distribution and control point parameters from scratch.

5. The radiotherapy planning re-optimization method based on target inheritance and flow reset according to claim 1, characterized in that: S5 specifically includes: performing dose calculation and dose normalization on the re-optimized soft plan, and solving for the final gantry parameters, dose rate and multi-leaf collimator blade motion sequence based on the calculation results to form the deliverable re-optimized radiotherapy plan.

6. The radiotherapy planning re-optimization method based on target inheritance and flow reset according to claim 1, characterized in that: The preset conditions mentioned in S6 specifically include: The target dose coverage index of the re-optimized radiotherapy plan is not lower than that of the initial radiotherapy plan, and it is superior to the initial radiotherapy plan in terms of dose index of at least one organ at risk.

7. The radiotherapy planning re-optimization method based on target inheritance and flow reset according to claim 1, characterized in that: The steps following S1 and before S2 include: Assess whether the initial radiotherapy plan meets all pre-set clinical dose constraints, and mark cases that do not meet all constraints as priority cases for further optimization; S2 to S6 are performed preferentially on the preferential re-optimization cases.

8. The radiotherapy planning re-optimization method based on target inheritance and flow reset according to claim 1, characterized in that: The method includes independently repeating S3 to S5 at least twice for the same case to obtain multiple re-optimized soft plans; In S5, the median values ​​of the multiple re-optimized soft plans on key dosimetric indicators are taken to generate the final deliverable re-optimized radiotherapy plan.

9. The radiotherapy planning re-optimization method based on target inheritance and flow reset according to claim 1, characterized in that: Prior to S6, the following is also included: The planning complexity of the re-optimized radiotherapy plan is evaluated, and the planning complexity indicators include the total number of monitoring units or the total travel of the multi-leaf collimator blades; The subsequent comparison and output steps are performed only when the change in the complexity of the re-optimized radiotherapy plan is within a preset threshold.

10. A radiotherapy planning re-optimization system based on target inheritance and flow reset, used to implement the radiotherapy planning re-optimization method based on target inheritance and flow reset as described in any one of claims 1-9, characterized in that, include: The data interface module is used to read initial plan data from the treatment planning system; The target freezing module is used to extract and lock the target set in the initial plan and generate a frozen target set file; The plan reset module is used to create a copy plan and perform a photon flow reset operation, clearing all deliverable irradiation parameters; The optimization driver module is used to load the frozen target set file onto the reset replication plan and drive the built-in optimizer of the treatment plan system to perform re-optimization; The plan generation module is used to calculate the dosage and solve for executable parameters based on the re-optimization results. The evaluation and decision-making module is used to compare the dosimetric indicators of the initial and re-optimized plans and output a recommended plan based on preset rules.

Images