A radiotherapy pathway planning method based on surface contour monitoring

Abstract

This invention discloses a radiotherapy pathway planning method based on surface contour monitoring, specifically relating to the field of medical image data processing. It addresses the issues of low reusability and intelligence in existing medical radiotherapy pathway planning methods. By collecting labeled target area feature parameters and incident point distribution from historical cases, a distribution prediction model for radiotherapy incident points is constructed. Three-dimensional segmentation of the target area and organ structures is performed in the current patient's CT images, establishing a unified voxel coordinate system. Based on the prediction results, an array of incident points covering the target area surface is constructed, generating a candidate radiotherapy pathway set, and risk organ identification is performed for each pathway. Furthermore, the spatial distance between the pathway trajectory and the risk organ contour surface is combined to calculate a comprehensive risk score for the candidate pathways. Finally, dose simulation is performed on the candidate pathways according to the score order, outputting radiotherapy pathway planning results that meet dose and safety requirements, thus improving the accuracy and clinical applicability of pathway selection.

Description

Technical Field

[0001] This invention relates to the field of medical image data processing technology, and more specifically, to a radiotherapy pathway planning method based on surface contour monitoring. Background Technology

[0002] In high-precision radiotherapy, the planning of the beam incidence path directly affects target coverage and irradiation avoidance of adjacent normal organs. This is especially true in anatomically complex areas such as the head and neck, chest, abdomen, and pelvis, where the target area often borders multiple high-risk organs, such as the optic nerve, spinal cord, lung tissue, mesentery, and bladder. Even slight deviations in the beam path can lead to excessive exposure to non-target doses. However, current initial selection of radiotherapy pathways generally relies on the experience of physicists, manually simulating and tentatively setting up multi-angle models on the three-dimensional structure of the image. This lack of automated initial selection guidance mechanisms results in low pathway generation efficiency, leading to significantly different incidence strategies for similar lesions in different patients. The initial pathway planning lacks consistency and reusability. Furthermore, current radiotherapy pathway simulations often employ anatomical structure-based occlusion judgment strategies to screen candidate pathways, using whether the beam path crosses important organs as the basis for pathway usability. However, this method often uses binary occlusion / non-occlusion results as the standard, lacking detailed assessments of the spatial proximity, crossing length, or risk level between the pathway and organs, making it difficult to achieve refined ranking and optimization of multiple pathway schemes. Meanwhile, the existing process relies heavily on manual specification or line-by-line simulation, lacking an automated path selection mechanism. This makes it difficult to quickly screen out high-risk paths and determine the optimal solution under multiple path options, severely restricting the efficiency and quality of path planning. Summary of the Invention

[0003] In order to overcome the above-mentioned defects of the prior art, embodiments of the present invention provide a radiotherapy path planning method based on surface contour monitoring to solve the problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides the following technical solution:

[0005] A radiotherapy pathway planning method based on surface contour monitoring includes the following steps:

[0006] S1: Collect target area characteristic parameters and radiotherapy incident point distribution of patients in historical radiotherapy cases, and establish a distribution prediction model of the target area corresponding to the radiotherapy incident point;

[0007] S2: Perform structural segmentation on the 3D CT images of the patient's tumor region and construct a voxel structure containing the target area and organ contours;

[0008] S3: Input the current patient target area feature parameters into the distribution prediction model and output the incident point planning area;

[0009] S4: Based on the array of incident points within the planned area of ​​the incident points, generate a set of radiotherapy candidate paths covering the surface of the target area, and mark the risk organs penetrated by each candidate path;

[0010] S5: Based on the spatial distance between the candidate path and the contour surface of the risk organ, and combined with the risk organ type, calculate the risk score of the candidate path;

[0011] S6: Perform dose distribution simulation verification on candidate pathways according to risk score order, and output the final radiotherapy pathway planning results.

[0012] In a preferred embodiment, in S1, collecting target area characteristic parameters and radiotherapy incident point distribution of patients in historical radiotherapy cases, and establishing a distribution prediction model of the target area corresponding to the radiotherapy incident point specifically includes:

[0013] Obtain CT image data with completed structural annotation from historical radiotherapy cases, and extract the location, volume, and boundary curvature of the radiotherapy target area as target area feature parameters;

[0014] Extract the actual radiotherapy incident point location corresponding to each radiotherapy target area, establish a mapping sample group of target area feature parameters and corresponding incident points, and normalize the sample group.

[0015] Gradient enhancement trees were selected as the regression model. Supervised training was performed based on the target region feature parameters and incident point positions in the normalized sample group to establish a radiotherapy incident point distribution prediction model.

[0016] In a preferred embodiment, in S2, structural segmentation is performed on the three-dimensional CT image of the patient's tumor region to construct a voxel structure containing the target area and organ contours, specifically including:

[0017] The CT scan slice sequence of the patient's tumor area is reconstructed into a multi-level three-dimensional volumetric image according to the acquisition order, and a three-dimensional coordinate system is established based on the preset voxel physical space size.

[0018] In the three-dimensional volumetric image, the target area and organ structure are segmented. The target area and organ contour are identified by the connectivity between gray-scale distribution and CT scan slices, and the corresponding three-dimensional voxel mask is generated.

[0019] The target area mask and the voxel mask of the organ outline are integrated into a structural mask set, and all voxels contained in the structural mask set are uniformly labeled with the name of the structure to which they belong.

[0020] In a preferred embodiment, the step of identifying the target region and organ contour by the connectivity between grayscale distribution and CT scan slices, and generating a corresponding three-dimensional voxel mask, specifically includes:

[0021] Based on the range of CT grayscale values ​​of the target area and organs in historical radiotherapy cases, all voxels in the three-dimensional volumetric image were screened for grayscale.

[0022] After screening, the voxel sets are subjected to upper and lower layer connectivity determination according to the CT slice direction. Voxel clusters with cross projection in continuous slices are retained, and discrete voxels with no connectivity and isolated layers are removed.

[0023] Spatial boundary closure is determined for connected voxel clusters, and target regions and organ contour voxel masks with complete structures are screened and constructed.

[0024] In a preferred embodiment, in S3, the current patient target area feature parameters are input into the distribution prediction model, and the output incident point planning region specifically includes:

[0025] Obtain the three-dimensional voxel mask of the target area of ​​the current patient and perform spatial registration to extract target area feature parameters;

[0026] The extracted target area feature parameters are standardized according to the normalization rules in the training process of the distribution prediction model, and then input into the distribution prediction model to output the matching radiotherapy incident point prediction results.

[0027] Based on the predicted radiotherapy incident points, a continuous and uniform array of incident points is constructed on the outer surface of the skin at the corresponding site as the incident point planning area.

[0028] In a preferred embodiment, in S4, a set of radiotherapy candidate pathways covering the target area surface is generated based on the array of incident points within the incident point planning region, and the risk organs penetrated by each candidate pathway are specifically marked as follows:

[0029] Obtain the array of incident points in the planned area of ​​the patient's outer skin surface, and label the spatial coordinates of each incident point according to the three-dimensional coordinate system of voxel mapping;

[0030] Starting from each incident point in the array, ray continuation is performed on the surface of the target area. The incident paths reaching the surface of the target area are selected and marked as candidate paths. The starting point coordinates, direction vectors and the sequence of intersection points of the paths with the target area are recorded.

[0031] The overlap detection between the incident path trajectory and the structural mask set is performed on the candidate path set to identify the organ structures passed through by the incident path, the corresponding organ structures are marked as risk organs, and the risk organ type is recorded.

[0032] In a preferred embodiment, in S5, the risk score calculation for the candidate path based on the spatial distance between the candidate path and the contour surface of the risky organ, combined with the risky organ type, specifically includes:

[0033] Candidate paths that do not penetrate the risk organ are selected from the candidate path set. In the three-dimensional coordinate system of voxel mapping, the shortest spatial distance between the candidate path trajectory and the voxel of the risk organ structure contour surface is calculated.

[0034] Based on the radiation resistance of each risk organ, a corresponding risk weight factor is set, and a weighted comprehensive calculation is performed based on the spatial distance between each candidate path trajectory and all risk organs, combined with the corresponding risk weight factor.

[0035] The weighted and combined calculation results of all candidate paths are converted into risk scores on a uniform scale.

[0036] In a preferred embodiment, in S6, the candidate pathways are simulated and validated according to risk score order, and the final radiotherapy pathway planning result is output, specifically including:

[0037] The candidate paths are sorted in ascending order of risk score to form a queue of candidate paths to be verified.

[0038] Candidate paths in the queue are selected sequentially, and a radiation beamline model is constructed based on the intersection sequence of the path's starting coordinates, direction vector, and corresponding target area, and the beam simulation parameters are initialized.

[0039] Beam energy attenuation calculations and dose deposition simulations were performed on the target mask, and the first qualified candidate path was selected from the verified candidate paths as the final radiotherapy path planning result.

[0040] The technical effects and advantages of the radiotherapy path planning method based on surface contour monitoring of this invention are as follows:

[0041] By constructing a predictive model between target region features and incident point distribution, the radiotherapy pathway planning has shifted from experience-driven to data-driven, improving the accuracy and efficiency of initial pathway screening. A three-dimensional voxel structure is used to represent the spatial relationship between the target region and at-risk organs in a unified coordinate system, providing structured support for pathway selection. Introducing an incident point planning region limits the pathway generation range, reducing the computational burden of invalid pathways while strengthening geometric constraints. Risk labeling of candidate pathways considers not only the type of penetrating organ but also incorporates spatial distance and organ biological attributes to construct a risk score, making the pathway assessment results more clinically relevant. Finally, dose distribution simulation is performed based on the risk score to screen the final pathway, effectively balancing treatment intensity and normal tissue protection, improving the safety and controllability of radiotherapy pathways. This approach boasts technical advantages such as clear structure, reusable model, and physically feasible pathway output. Attached Figure Description

[0042] Figure 1 This is a schematic diagram of a radiotherapy path planning method based on surface contour monitoring according to the present invention. Detailed Implementation

[0043] 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, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0044] Example 1

[0045] Figure 1 This invention presents a radiotherapy path planning method based on surface contour monitoring, which includes the following steps:

[0046] S1: Collect target area characteristic parameters and radiotherapy incident point distribution of patients in historical radiotherapy cases, and establish a distribution prediction model of the target area corresponding to the radiotherapy incident point;

[0047] S2: Perform structural segmentation on the 3D CT images of the patient's tumor region and construct a voxel structure containing the target area and organ contours;

[0048] S3: Input the current patient target area feature parameters into the distribution prediction model and output the incident point planning area;

[0049] S4: Based on the array of incident points within the planned area of ​​the incident points, generate a set of radiotherapy candidate paths covering the surface of the target area, and mark the risk organs penetrated by each candidate path;

[0050] S5: Based on the spatial distance between the candidate path and the contour surface of the risk organ, and combined with the risk organ type, calculate the risk score of the candidate path;

[0051] S6: Perform dose distribution simulation verification on candidate pathways according to risk score order, and output the final radiotherapy pathway planning results.

[0052] In S1, target area characteristic parameters and radiotherapy incident point distribution of patients in historical radiotherapy cases are collected to establish a distribution prediction model of the target area corresponding to the radiotherapy incident point.

[0053] This dataset compiles 3D CT images of patients who have undergone structural segmentation and annotation during previous radiotherapy. Each dataset includes the tumor target structure confirmed by the physician, along with the corresponding treatment path planning record. The image data used has been constructed using DICOM 3D image volume, with a slice spacing of no more than 2mm and a uniform image matrix size of 512×512×N, where N is the number of patient scan layers, all meeting the spatial accuracy requirements for subsequent structural feature extraction. For each dataset, the voxel mask structure of the tumor target region in the image volume is extracted by reading its structural annotation information (see step S2 in the instruction manual for the construction process). This indicates whether each voxel in the image volume belongs to tumor tissue. Spatial geometric processing is performed on the non-zero voxel coordinate set in the mask structure to calculate the position of its geometric center in the 3D coordinate system (see step S2 in the instruction manual for the coordinate system construction process), which is used as the positional feature parameter of the target region. Simultaneously, the number of non-zero voxels in the mask is counted, and combined with the actual spatial size corresponding to each voxel (default setting is 2mm³), the actual volume value of the tumor is obtained, serving as the volume feature parameter. For boundary curvature extraction, surface reconstruction is first performed based on the voxel surface boundary to extract the closed surface model. Then, the average Gaussian curvature of each vertex is calculated on the 3D mesh model, and the mean and variance of the overall curvature distribution are statistically analyzed as morphological complexity indicators. If structural burrs or non-closed structures exist in the curvature extraction region, abnormal surface points are removed and bidirectional triangular mesh smoothing is performed to ensure the stability and usability of the boundary curvature extraction results. Finally, the position coordinates, volume values, and boundary curvature indicators are combined to form a unified target area feature parameter vector, which serves as the input variable for subsequent modeling. All extraction results are stored in a unified structured data format to ensure batch reading and comparison during the model training phase.

[0054] After extracting the feature parameters of each target area, the actual radiotherapy path information used by each patient is further obtained based on historical records, with a focus on extracting the radiotherapy incident point of the main path. This incident point is defined as the location where the treatment path first penetrates the patient's body surface, and is recorded in the radiotherapy plan document in the form of human body location annotations. To ensure data consistency, the incident point needs to be registered to a three-dimensional coordinate system consistent with the CT image. The registration operation is completed based on the image origin positioning parameters and coordinate transformation matrix, with the error controlled within 2mm. For each sample, the extracted target area feature parameters and the corresponding incident point coordinates are combined to form an input-output pair, i.e., the feature vector is used as input, and the three-dimensional position of the incident point is used as the target output, establishing a supervised learning sample group. In terms of sample size, data from no fewer than 500 patients with tumors from different sites are selected, covering multiple target area types such as lung cancer, liver cancer, and brain tumors. Each type of sample is labeled to facilitate subsequent analysis of generalization ability. To improve the numerical convergence efficiency during model training and reduce the bias caused by differences in different feature dimensions, normalization processing is performed on all input and output variables in the sample group. The input vectors are normalized using a min-max normalization method, mapping each dimension to the [0,1] interval based on its maximum and minimum values ​​across the entire sample. Output coordinates are normalized along each axis based on the effective beam incidence range determined by the patient's body contour boundaries, preventing coordinate values ​​from falling into abnormal regions. All normalization parameters are calculated at fixed intervals during dataset construction to maintain consistency between training and prediction. Finally, standardized sample sets of feature-target pairs are constructed.

[0055] Based on the established normalized sample set, the gradient boosting tree algorithm was selected as the regression model to learn the nonlinear mapping relationship between target region feature parameters and the spatial coordinates of the actual incident point. The gradient boosting tree model was chosen because it has high expressive power for small-to-medium scale structured data, can capture complex combinations of target region morphological features and spatial incident preferences, and possesses strong anti-overfitting ability and feature importance interpretation capability. In terms of modeling structure, three independent regressors were constructed to predict the x, y, and z coordinate dimensions of the incident point respectively (x, y, and z are detailed in step S2 of the instruction manual), ensuring that the fitting process for each dimension is independent and the gradient path is stable. During training, the input is the normalized target region feature vector, and the output is the corresponding normalized coordinate value of the incident point. The mean squared error index was used as the loss function, the number of iterations was set to 300, the minimum number of samples per leaf node was set to 10, and the learning rate was fixed at 0.05. To avoid training anomalies caused by sample order or data bias, a five-fold cross-validation mechanism was used during the training phase, with stratified sampling by case number to evaluate the error convergence stability of the model on the validation set in each round. After model convergence, the output coordinates are inversely normalized according to the original normalization parameters to restore the actual incident point coordinates. The Euclidean distance between these coordinates is then compared to the true incident point coordinates, serving as an evaluation metric for model prediction accuracy. In the full test set, over 82% of the predictions had an error less than 5 mm (this can be dynamically adjusted based on accuracy requirements), indicating that the model has good spatial fitting capabilities. Finally, the model is named the Incident Point Distribution Prediction Model and deployed. It accepts target area features from any new patient in radiotherapy pathway planning, quickly outputting the planned incident point region, providing preliminary constraints for candidate pathway selection and dose distribution simulation.

[0056] In S2, the patient's tumor region CT 3D image is structurally segmented to construct a voxel structure containing the target area and organ contour.

[0057] The patient's CT scan image data was processed. The input data consisted of multiple consecutive DICOM format 2D grayscale images, each corresponding to one scan layer. The image sequence was sorted along the z-axis and hierarchically encoded. Each image had a uniform resolution of 512×512 pixels, with pixel spacing explicitly marked in the image metadata. The inter-slice spacing was set to no more than 2mm in the acquisition protocol, using an isometric scanning mode (all CT image-related parameters were set consistently with those in step S1). After sorting the image sequence by slice number, it was stacked sequentially according to its image matrix size and inter-slice spacing to form a 3D image volume. The data structure of this 3D image volume was a 3D grayscale voxel array, with each voxel retaining its corresponding CT value. To ensure a consistent spatial reference system for subsequent structural analysis, path penetration determination, and dose simulation, a physically mapped 3D coordinate system was established on the image volume. This coordinate system had its origin at the top-left corner (layer 0 pixel), with the x-axis defined along the row and column directions, the y-axis along the column direction, and the z-axis perpendicular to the slice direction. The coordinate unit was millimeters, consistent with the CT parameters. The voxel spatial position index is directly mapped to a three-dimensional coordinate system (x, y, z). The calculation method uses a linear mapping, where the increment in the x-direction is the pixel width, the increment in the y-direction is the pixel height, and the increment in the z-direction is the slice layer spacing. In this embodiment, the standard physical size of the voxel is set to 1mm × 1mm × 2mm, and a voxel resampling operation is performed on the original image to ensure consistent spatial resolution in the three-dimensional direction of the reconstructed image volume. This ensures accurate coordinate representation during subsequent voxel-level structure annotation, ray path construction, and dose mapping calculations. The final three-dimensional volume image and its corresponding coordinate system will serve as the spatial reference framework for downstream target recognition and structural segmentation.

[0058] Based on a unified coordinate system of three-dimensional image volume, segmentation of tumor target areas and organ structures is performed. In this embodiment, the structure recognition process combines initial screening based on grayscale thresholds with spatial connectivity enhancement recognition. First, statistical analysis is performed on the CT grayscale values ​​in the image volume. Known target tissues in the sample set typically exhibit density differences compared to adjacent tissues in terms of CT values; for example, soft tissue tumors often show a grayscale range between lung parenchyma and bone density. An initial screening threshold for candidate target areas is set. Voxel regions with grayscale values ​​between 40 and 80 (the specific threshold range is determined based on a comprehensive analysis of target area and organ grayscale values ​​in historical CT images) are extracted as preliminary candidate voxel sets. Then, intra-slice regional connectivity analysis is performed on each candidate voxel set, retaining only voxel clusters that are closed regions within a slice layer and have vertical structural continuity across multiple consecutive slices. The continuity between upper and lower slices is determined by setting a maximum connectivity gap within two layers along the Z-axis and requiring that the centroid distance between voxel regions in adjacent layers does not exceed 5 mm. For target areas with blurred edges or small structures, a three-dimensional morphological closing operation is further introduced to fill holes and repair pseudo-cracks in the segmentation results, ensuring the spatial coherence and integrity of the target area structure. Organ structure recognition is performed using a matching method based on standard anatomical location templates. For example, high-risk organs such as the spinal cord, heart, and lungs have relatively stable spatial distributions among different patients. After locating based on the bony reference line of the thoracic cage, the approximate center of the organ can be quickly located, and grayscale threshold extraction and connectivity enhancement operations are performed in the central region. Finally, each identified structure is represented as a voxel mask, which is a three-dimensional array with the same volume dimension as the image. Voxels with a value of 1 in the mask represent voxels within the structure, and the remaining voxels are set to 0. The output results include target area masks and multiple organ masks, each mask corresponding to a unique structure type.

[0059] After generating voxel masks for each individual structure, the spatial representations of all structures are unified and integrated into a structured mask set. In this embodiment, a 3D volumetric image with a volume size of 512×512×N is used as a unified reference dimension. A structure label array is defined, which is a 3D integer array with an initial value of 0, representing an unassigned state. Subsequently, a voxel-by-voxel scan operation is performed on the target area mask array. For voxels with a mask value of 1, the corresponding voxel in the structure label array is assigned a preset number 1, identifying it as a tumor target area. Next, the organ structure masks are scanned sequentially, and the structure labels are written into the structure label array according to the structure. For example, the heart is assigned number 2, the spinal cord is assigned number 3, the left lung is assigned number 4, and the right lung is assigned number 5. The numbering order is fixed to prevent identification conflicts when structures overlap. During the writing process, for points where a non-zero number already exists at the voxel position, such as overlapping areas between organs, the structure number with the smaller number is retained, and these are considered higher-risk structures. To establish a mapping between structural labels and clinical terminology, each structural number is fixedly mapped to a specific name in the configuration file; for example, number 2 corresponds to "heart" and number 3 corresponds to "spinal cord." Ultimately, each voxel in the structural mask set has a unique structural identifier, and the entire mask set forms a spatial structural annotation map that can be used for path penetration determination, dose distribution mapping, and identification of organs at risk. This structural label map not only has spatial localization capabilities but also anatomical semantic expression capabilities, serving as a core reference structure in the radiotherapy pathway planning and candidate pathway risk analysis stages.

[0060] In S3, the current patient target area feature parameters are input into the distribution prediction model, and the incident point planning area is output.

[0061] If the original acquisition direction of the patient image differs from the direction used during the training of the distribution prediction model (e.g., head-to-toe reversal, left-to-right flipping), a direction matrix transformation is performed based on DICOM positioning to ensure that the mask structure maintains spatial consistency with historical model samples. After registration, target region feature parameters are extracted from the registered voxel set, using the same method as in step S1 of the instruction manual. These feature parameters include position coordinates, volume values, and boundary curvature indices, formatted in the same way as the historical training sample structure in step S1 of the instruction manual, and are used as input data for the distribution prediction model. The extracted target region feature parameters are standardized according to the normalization rules used during the distribution prediction model training process. After normalization, the input is the standardized feature vector, and the output is the predicted radiotherapy incident point, a three-dimensional spatial coordinate value representing the spatial location of the optimal radiotherapy incident point corresponding to the current patient's target region on the skin surface. This prediction result is the main path incident starting point learned by the model by combining historical experience path preferences with the current target region spatial configuration. The model structure employs a gradient boosting tree, and independent regression paths are established for the x, y, and z coordinate dimensions during the training phase. Therefore, the inference output consists of three continuous coordinate values. After the prediction results are generated, they are mapped back to the three-dimensional coordinate system corresponding to the CT image to ensure consistency with the image volume, structural mask, and other path planning data.

[0062] After obtaining the predicted coordinates of the radiotherapy incident point, a certain range of outward expansion processing is constructed around the location of these coordinates on the body surface to define the incident point planning region, forming the basis for the generation of subsequent candidate paths. Specifically, the predicted coordinates are first located in the 3D image volume, and it is determined whether they are located at the image boundary. If they are close to the image boundary, the center of the region is adjusted to within 10mm (or the 3D coordinate system is extended according to the voxel size) to ensure that the subsequent array can be generated completely. Then, the skin region is searched and constructed. The method is to perform ray tracing outward from the predicted incident point in the normal direction, identify the first non-in vivo voxel as the skin surface point, and then extract the surface segment where the point is located as the construction basis. Based on this point as the center, an orthogonal coordinate patch is constructed along the skin surface to form a 2D unfolded planar region. The physical size of this region is set to 40mm×40mm, divided into a 10×10 grid, and all grid intersections are defined as candidate incident point arrays. Each incident point records its spatial 3D coordinates and normal direction vector. This array of points is called the incident point planning region, representing all possible incident positions within an area centered on the predicted incident point and possessing reasonable anatomical structure. During the path generation phase, all points in this array are considered potential path starting points and ray-extended to expand the path candidate set and improve coverage accuracy. By planarizing and unfolding the surface and then mapping it to three-dimensional coordinates, a high-quality set of starting points is constructed that conforms to the skin surface morphology and has uniform spatial density, providing a clear regional input for subsequent path optimization and obstacle avoidance judgment.

[0063] In S4, a set of radiotherapy candidate paths covering the target area surface is generated based on the array of incident points within the incident point planning area, and the risk organs penetrated by each candidate path are marked.

[0064] After constructing the incident point planning region, the system performs 3D coordinate assignment and structured output for the incident point array within that region. This array is obtained by uniformly dividing a 2D grid, with each grid intersection representing an incident point. Since the incident points are located on the outer surface of the skin, to accurately obtain their physical spatial positions, coordinate mapping is performed using a 3D voxel coordinate system established from the patient's CT images. First, the 2D unfolded reference plane containing the incident point planning region is obtained and reprojected onto the corresponding curved surface position of the outer skin surface in 3D space. The 3D coordinates of each incident point are calculated using voxel indexing and the physical spacing between voxels. If the patient images are acquired using non-uniform spacing, different axial pixel sizes are introduced to convert the corresponding axial voxel coordinates. To ensure data structure consistency, the spatial coordinates of all incident points are recorded in a unified unit and stored as 3D vectors. The incident point array ultimately forms a set of points containing coordinate vectors and surface normal vectors. The normal vectors are used to determine the subsequent path extension direction, without introducing subjective bias. A set of initial coordinates for radiotherapy incident points with physical spatial consistency, based on the patient's actual skin surface, is established. Subsequent path construction tasks are performed based on this set. Using the obtained incident point array, a path extension operation towards the target area is initiated point by point. Each incident point is assigned its surface normal vector as the initial extension direction, thus constructing a ray trajectory that penetrates the voxel structure, which is formally a spatial straight line segment extending from the starting point on the skin surface into the internal space. To avoid path deviation from the target area space, voxel-level spatial tracking is performed on each ray during the extension process. This is done by progressively calculating the sequence of voxels traversed by the ray and detecting whether it intersects with the target area mask. Once a ray overlaps with a voxel in the target area mask, it is considered to have successfully penetrated the target structure, and the path is marked as a valid incident path and added to the candidate path set. Simultaneously, three key parameters of the path are recorded: first, the path starting point, i.e., the coordinates of the incident point; second, the direction vector, derived from the surface normal but normalized; and third, the intersection sequence, i.e., the continuous set of spatial positions formed by the path traversing the target area structure voxels, reflecting the contact trajectory of the path within the target area surface. Ray paths that fail to enter the target area are directly discarded and not recorded. After the above processing, a set of candidate radiotherapy paths that truly penetrate the target area is formed. Each path has a complete structure, clear geometric information, and incident direction, providing accurate spatial input data for subsequent risk structure detection and dose calculation.

[0065] After obtaining the candidate path set, the focus is on identifying the human structures penetrated by each path to determine if any important organs are passed through. Specifically, the process involves traversing the ray path from the starting point to the target area intersection of each path in the candidate path set, and progressively comparing its corresponding continuous spatial position sequence in the voxel structure with the structural mask set. The structural mask set contains multiple marker layers, recording the voxel position distribution and name information of different organ structures. The rule for determining path-mask overlap detection is: if the voxel position traversed by the path intersects with the mask voxel set of a certain organ, then the path is considered to have passed through that organ structure. To improve detection accuracy, each path undergoes voxel-level cross-checking, rather than only detecting the starting and ending points. Furthermore, the target area structure itself is excluded during overlap determination to avoid self-penetration and repeated marking. The name of each overlapping organ structure is extracted from the mask set and appended to the path's structural penetration list. If the penetrated organ is a medically defined important organ, such as the lung, liver, heart, or kidney, the path is marked as a path with a risky organ. In this step, organ names are recorded as a classification criterion, providing semantic information support for subsequent risk weight assignment and pathway priority ranking. After this detection operation is completed, each candidate pathway possesses complete spatial geometric information and risk structure markers. The pathway set has been transformed from a purely geometric structure into a set of clinically interpretable functional pathways, fulfilling the prerequisites for implementing pathway screening and dosage adjustment.

[0066] In S5, risk scores are calculated for candidate paths based on the spatial distance between the candidate path and the contour surface of the risk organ, combined with the risk organ type.

[0067] After the candidate path set is constructed and the organ penetration information of each path is labeled, a subset of paths that do not penetrate any high-risk organs are selected as the objects of distance evaluation. "Not penetrating a high-risk organ" means that the candidate path trajectory does not overlap with any voxel region defined as a high-risk organ in the structural mask set. Such paths are usually superior in spatial planning and have high clinical adoptability, but their proximity to high-risk organs still needs further evaluation. Therefore, in a three-dimensional voxel coordinate system, the shortest spatial distance between each path trajectory and the voxel boundary surfaces of all high-risk organs is calculated organ-by-organ. Each path trajectory is defined by the sequence of intersections of the initial incident point, the extension direction vector, and the target area, and is represented in space as a straight line segment structure penetrating the target area. Distance measurement is performed based on the three-dimensional voxel boundary voxel set corresponding to each organ structure and the path trajectory. The voxel boundary voxel set consists of the edge voxels of all adjacent non-structural voxels in the organ mask body, thus forming the contour surface of the organ. For each boundary voxel, the shortest spatial distance between its center point and the path trajectory is calculated. A minimum value selection method is used to obtain the minimum distance value between each path and each organ. Since all distance calculations are performed in a unified three-dimensional voxel coordinate system and the voxel size is standardized, the obtained distances have physical spatial consistency. The shortest spatial distance between each candidate path and each at-risk organ is stored in matrix form, providing complete input data for subsequent risk scoring calculations. This processing avoids the discontinuity problem of path penetration detection and compensates for the deficiency in traditional radiotherapy path evaluation that over-relies on penetrating structures while ignoring nearby hazards.

[0068] After constructing the shortest distance matrix, a comprehensive risk score for each candidate path is calculated based on a joint assessment mechanism of organ radiosensitivity and path geometric distance. First, based on clinical data and a radiation biology knowledge base, the radioresistance levels of common vital organs are defined, and a fixed risk weight factor is assigned to each risk organ accordingly. This factor reflects the organ's tolerance to radiation damage; the weaker the organ's cell regeneration capacity, the higher the corresponding weight factor value. For example, critical structures such as the spinal cord, optic nerve, and kidneys are assigned higher risk factors (e.g., 1.0), while organs with relatively strong regenerative capacity, such as the liver and lungs, are assigned lower factors (e.g., 0.3-0.5). These risk factors are pre-set by domain professionals based on the patient's actual organ function and health, and maintain a consistent standard throughout the entire process to ensure the scoring results are comparable. The weighted distance score between each candidate path and all risk organs is calculated. Specifically, for each risk organ, a distance scoring function (e.g., inverse distance, logarithmic function, etc.) is applied to the path-organ shortest distance, and then multiplied by the corresponding organ's risk weight factor to obtain the path's risk contribution value to that organ. The contribution values ​​of all at-risk organs are summed to form the raw total risk score for the pathway. To ensure a consistent output scale across different pathways, all raw score results are normalized to fall within a fixed range (e.g., 0 to 1), resulting in a standardized risk score. A higher score indicates that the pathway is spatially closer to highly sensitive organs and that the overall radiation tolerance of the involved organs is lower, thus posing a greater clinical risk; conversely, a lower score indicates a safer pathway.

[0069] In S6, the candidate pathways are simulated and validated according to the risk score order, and the final radiotherapy pathway planning results are output.

[0070] After calculating the risk scores of candidate paths and obtaining standardized scores on a unified scale, a sorting operation is performed based on the scores. The candidate paths are stored in ascending order as a linear path queue to construct a queue of candidate paths to be verified, which serves as the input source for the subsequent beam simulation verification module. This path queue logically represents the order of path selection from theoretically safest to relatively high-risk, and its core purpose is to prioritize the evaluation of the optimal path with the goal of minimizing potential secondary damage.

[0071] In the sorted candidate path queue, candidate path entries are read sequentially starting from the head of the queue. Each entry contains three key geometric parameters: the three-dimensional coordinates of the path's starting point, the path's direction vector, and the sequence of intersections formed when the path intersects with the target voxel boundary during its penetration of the target area. These parameters fully describe the propagation behavior and incident mode of the radiation beam in three-dimensional space. Based on these three parameters, a radiation beam geometry model is constructed, which serves as the path basis for subsequent dose simulation and energy deposition. When constructing the beam model, the path's starting point is used as the simulated particle source, the direction vector defines the beam direction, and the sequence of intersections is used to define the boundary of the target area's energy deposition region. During the simulation parameter initialization phase, the radiation source energy level (e.g., 6MV or 10MV X-rays), beam aperture size (e.g., adding a 1cm safety boundary to the target area boundary), energy propagation spacing (e.g., one dose unit is projected per millimeter), and dose attenuation model (e.g., exponential decay function or Monte Carlo approximation range) are set. All parameters are defined in real physical units and conform to industry standards. After each initialization, a simulated beam model for that path is constructed, establishing a mapping relationship from the incident point to each energy distribution unit in the target area within a three-dimensional voxel structure. All path construction processes are static, unaffected by target area deformation or dynamic respiration-induced offsets, thus exhibiting high controllability. After completing the beamline model construction and parameter initialization, beam energy propagation and dose deposition simulation operations are initiated based on the current candidate path. First, the penetration area is traversed voxel by voxel along the beam direction, and the beamline unit energy intensity is updated in real time according to the energy attenuation model. An approximate physical model is used to simulate the energy loss behavior of X-rays in different tissues. The remaining dose intensity in each voxel is calculated based on the path traversal length and tissue density, and accumulated to form a dose distribution map.

[0072] After the dose distribution map is constructed, it is mapped to the target mask area to evaluate the match between the actual deposited dose and the clinically prescribed dose. Evaluation criteria include, but are not limited to, the following three: whether the average dose within the target area reaches the preset lower limit of the prescribed dose; whether there are significantly under-dose voxels in the target area edge region; and whether there are over-dose areas overflowing the target boundary into non-target tissues. Quantitative thresholds are used for the evaluation criteria; for example, at least 95% of the target area voxels must meet at least 90% of the prescribed dose. If the dose simulation results corresponding to the candidate path fully meet the above clinical dose coverage criteria, the path is marked as validated and output as the final radiotherapy path, terminating the subsequent path validation process. If the current path does not meet the requirements, it automatically moves to the next path and repeats the above simulation and validation process. Through this strategy, the path with the lowest risk score is prioritized without compromising dose adaptability, forming an optimal "low-risk + high-dose matching" comprehensive path scheme. This process avoids manual intervention and trial-and-error in pathway planning, significantly improving the efficiency and clinical usability of pathway planning. At the same time, it provides a clear logical chain and evaluation criteria for pathway selection, which aligns with the engineering development trend of modern intelligent radiotherapy.

[0073] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters and thresholds in the formulas are set by those skilled in the art according to the actual situation.

[0074] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. The semiconductor medium can be a solid-state drive.

[0075] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0076] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and modules described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0077] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or modules may be electrical, mechanical, or other forms.

[0078] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0079] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0080] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0081] 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.

[0082] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A radiotherapy path planning method based on surface contour monitoring, characterized in that, Includes the following steps: S1: Collect target area characteristic parameters and radiotherapy incident point distribution of patients in historical radiotherapy cases, and establish a distribution prediction model of the target area corresponding to the radiotherapy incident point; S2: Perform structural segmentation on the 3D CT images of the patient's tumor region and construct a voxel structure containing the target area and organ contours; S3: Input the current patient target area feature parameters into the distribution prediction model and output the incident point planning area; S4: Based on the array of incident points within the planned area of ​​the incident points, generate a set of radiotherapy candidate paths covering the surface of the target area, and mark the risk organs penetrated by each candidate path; S5: Based on the spatial distance between the candidate path and the contour surface of the risk organ, and combined with the risk organ type, calculate the risk score of the candidate path; S6: Perform dose distribution simulation verification on candidate pathways according to risk score order, and output the final radiotherapy pathway planning results; In S1, target area characteristic parameters and radiotherapy incident point distribution of patients in historical radiotherapy cases are collected, and a distribution prediction model of the target area corresponding to the radiotherapy incident point is established, specifically including: Obtain CT image data with completed structural annotation from historical radiotherapy cases, and extract the location, volume, and boundary curvature of the radiotherapy target area as target area feature parameters; Extract the actual radiotherapy incident point location corresponding to each radiotherapy target area, establish a mapping sample group of target area feature parameters and corresponding incident points, and normalize the sample group. Gradient enhancement tree was selected as the regression model. Supervised training was performed based on the target region feature parameters and incident point positions in the normalized sample group to establish a radiotherapy incident point distribution prediction model. In S3, the current patient target area feature parameters are input into the distribution prediction model, and the output incident point planning region specifically includes: Obtain the three-dimensional voxel mask of the target area of ​​the current patient and perform spatial registration to extract target area feature parameters; The extracted target area feature parameters are standardized according to the normalization rules in the training process of the distribution prediction model, and then input into the distribution prediction model to output the matching radiotherapy incident point prediction results. Based on the predicted radiotherapy incident points, a continuous and uniform array of incident points is constructed on the outer surface of the skin at the corresponding site as the incident point planning area.

2. The radiotherapy path planning method based on surface contour monitoring according to claim 1, characterized in that, In S2, structural segmentation is performed on the 3D CT images of the patient's tumor region to construct a voxel structure containing the target area and organ contours. Specifically, this includes: The CT scan slice sequence of the patient's tumor area is reconstructed into a multi-level three-dimensional volumetric image according to the acquisition order, and a three-dimensional coordinate system is established based on the preset voxel physical space size. In the three-dimensional volumetric image, the target area and organ structure are segmented. The target area and organ contour are identified by the connectivity between gray-scale distribution and CT scan slices, and the corresponding three-dimensional voxel mask is generated. The target area mask and the voxel mask of the organ outline are integrated into a structural mask set, and all voxels contained in the structural mask set are uniformly labeled with the name of the structure to which they belong.

3. The radiotherapy path planning method based on surface contour monitoring according to claim 2, characterized in that, The process of identifying target regions and organ contours through the connectivity between grayscale distribution and CT scan slices, and generating corresponding three-dimensional voxel masks, specifically includes: Based on the range of CT grayscale values ​​of the target area and organs in historical radiotherapy cases, all voxels in the three-dimensional volumetric image were screened for grayscale. After screening, the voxel sets are subjected to upper and lower layer connectivity determination according to the CT slice direction. Voxel clusters with cross projection in continuous slices are retained, and discrete voxels with no connectivity and isolated layers are removed. Spatial boundary closure is determined for connected voxel clusters, and target regions and organ contour voxel masks with complete structures are screened and constructed.

4. The radiotherapy path planning method based on surface contour monitoring according to claim 1, characterized in that, In S4, based on the array of incident points within the incident point planning area, a set of radiotherapy candidate pathways covering the target area surface is generated, and the specific risk organs penetrated by each candidate pathway are marked: Obtain the array of incident points in the planned area of ​​the patient's outer skin surface, and label the spatial coordinates of each incident point according to the three-dimensional coordinate system of voxel mapping; Starting from each incident point in the array, ray continuation is performed on the surface of the target area. The incident paths reaching the surface of the target area are selected and marked as candidate paths. The starting point coordinates, direction vectors and the sequence of intersection points of the paths with the target area are recorded. The overlap detection between the incident path trajectory and the structural mask set is performed on the candidate path set to identify the organ structures passed through by the incident path, the corresponding organ structures are marked as risk organs, and the risk organ type is recorded.

5. The radiotherapy path planning method based on surface contour monitoring according to claim 1, characterized in that, In S5, the risk score calculation for candidate paths, based on the spatial distance between the candidate path and the contour surface of the risky organ, combined with the risky organ type, specifically includes: Candidate paths that do not penetrate the risk organ are selected from the candidate path set. In the three-dimensional coordinate system of voxel mapping, the shortest spatial distance between the candidate path trajectory and the voxel of the risk organ structure contour surface is calculated. Based on the radiation resistance of each risk organ, a corresponding risk weight factor is set, and a weighted comprehensive calculation is performed based on the spatial distance between each candidate path trajectory and all risk organs, combined with the corresponding risk weight factor. The weighted and combined calculation results of all candidate paths are converted into risk scores on a uniform scale.

6. The radiotherapy path planning method based on surface contour monitoring according to claim 1, characterized in that, In S6, the candidate pathways are simulated and validated according to their risk scores, and the final radiotherapy pathway planning results are output, including: The candidate paths are sorted in ascending order of risk score to form a queue of candidate paths to be verified. Candidate paths in the queue are selected sequentially, and a radiation beamline model is constructed based on the intersection sequence of the path's starting coordinates, direction vector, and corresponding target area, and the beam simulation parameters are initialized. Beam energy attenuation calculations and dose deposition simulations were performed on the target mask, and the first qualified candidate path was selected from the verified candidate paths as the final radiotherapy path planning result.

Images