Detailed Description
In order to more clearly illustrate the technical solutions of the embodiments of the present specification, the drawings that are required to be used in the description of the embodiments will be briefly described below. The drawings do not represent all embodiments.
It will be appreciated that "system," "apparatus," "unit" and/or "module" as used herein is one method for distinguishing between different components, elements, parts, portions or assemblies at different levels. Other words may be substituted for the words by other expressions if the words achieve the same purpose.
The terms "a," "an," "the," and/or "the" are not specific to the singular, but may include the plural, unless the context clearly indicates otherwise. In general, the terms "comprises" and "comprising" merely indicate that the steps and elements are explicitly identified, and they do not constitute an exclusive list, as other steps or elements may be included in a method or apparatus.
In the embodiments of the present disclosure, when the operations are performed according to the step descriptions, the order of the steps may be changed, the steps may be omitted, and other steps may be included in the operation process unless otherwise specified.
FIG. 1 is a system schematic diagram of a radiation therapy collision avoidance system based on three-dimensional scanning and real-time simulation, shown in accordance with some embodiments of the present description.
In some embodiments, the radiation therapy anti-collision system 100 based on three-dimensional scanning and real-time simulation includes a model generation module 110 and a simulation modeling module 120.
The model generation module is configured to generate an initial three-dimensional model based on body surface data of the patient.
The simulation modeling module is configured to perform a simulation of an anti-collision during radiation therapy based on the initial three-dimensional model, and to determine a first planned path during radiation therapy.
In some embodiments, the radiation therapy anti-collision system 100 based on three-dimensional scanning and real-time simulation further includes a processor and memory.
The processor is configured to process data from at least one module or external data source of the radiation therapy anti-collision system 100 based on three-dimensional scanning and real-time simulation. In some embodiments, the processor includes a central processing unit, an application specific integrated circuit, an image processing unit, a controller, or the like, or any combination thereof.
The memory is configured to store data, instructions, and/or any other information. In some embodiments, the memory includes mass memory, removable memory, or the like, or any combination thereof.
For the foregoing detailed description, reference may be made to the associated descriptions of fig. 2-5.
It should be appreciated that the radiation therapy anti-collision system and its modules shown in fig. 1, which are based on three-dimensional scanning and real-time simulation, can be implemented in a variety of ways. It should be noted that the above description of the system and its modules is for convenience of description only and is not intended to limit the present description to the scope of the illustrated embodiments. It will be appreciated by those skilled in the art that, given the principles of the system, various modules may be combined arbitrarily or a subsystem may be constructed in connection with other modules without departing from such principles. In some embodiments, the model generation module 110 and the simulation module 120 disclosed in fig. 1 may be different modules in a system, or may be one module to implement the functions of two or more modules. For example, each module may share one memory module, or each module may have a respective memory module. Such variations are within the scope of the present description.
Fig. 2 is an exemplary flow diagram of a radiation therapy anti-collision method based on three-dimensional scanning and real-time simulation, according to some embodiments of the present description. In some embodiments, the process 200 is performed by a radiation therapy collision avoidance system (hereinafter referred to as a collision avoidance system) based on three-dimensional scanning and real-time simulation.
In some embodiments, the anti-collision system generates an initial three-dimensional model 220 based on the patient's body surface data 210 and performs a simulation of an anti-collision during radiation therapy based on the initial three-dimensional model 220, determining a first planned path 230 during radiation therapy.
For the relevant content of the anti-collision system, see the corresponding description of fig. 1.
Body surface data refers to data related to the body surface of a patient. For example, at least one of the body shape, height, size of each part, posture, and the like of the patient. The patient includes the person in need of radiation therapy.
In some embodiments, the collision avoidance system is in communication with the data acquisition device and acquires body surface data via the data acquisition device.
In some embodiments, the data acquisition device may be deployed around the patient or in any feasible location, including at least one of a 3D scanning device, an optical camera, an ultrasound scanning device, and the like. The 3D scanning device includes a laser or a photosensitive sensor, etc.
Body surface data acquired by the 3D scanning device are represented by three-dimensional space point clouds. The body surface data acquired by the optical camera is represented in the form of an image. The body surface data acquired by the ultrasonic scanning device is represented by reflection data of ultrasonic waves.
The initial three-dimensional model is a three-dimensional model obtained by modeling based on body surface data. The initial three-dimensional model includes three-dimensional information of the patient, and the three-dimensional information refers to information corresponding to body surface data in the three-dimensional model. For example, the three-dimensional information includes the size of a three-dimensional model corresponding to the body type of the patient, and the like.
In some embodiments, the collision avoidance system generates an initial three-dimensional model based on body surface data by any feasible modeling method (e.g., factorization, neural network, etc.) and/or external modeling software.
In some embodiments, the collision avoidance system may also acquire structural data of a support structure supporting the patient and generate an initial three-dimensional model based on the body surface data and the structural data.
A support structure refers to a structure used to support or contact the body of a patient during radiation therapy. For example, a patient receives a bed of treatment, a device coupled to a body surface of the patient, etc.
Structural data refers to data related to the physical dimensions of the support structure. For example, at least one of the size of the support structure, the relative position to the patient, etc.
In some embodiments, the collision avoidance system obtains structural data via a data acquisition device. The data form of the structure data is similar to that of the body surface data.
In some embodiments, the collision avoidance system generates an initial three-dimensional model based on the structural data and the body surface data by any feasible modeling method and/or external modeling software. If the anti-collision system generates an initial three-dimensional model based on the structure data and the body surface data, the three-dimensional information comprises information corresponding to the structure data in the three-dimensional model. For example, the size of a three-dimensional model corresponding to the support structure, etc.
In some embodiments of the present disclosure, the initial three-dimensional model is constructed with consideration given to the support structure associated with the patient, so that a more accurate initial three-dimensional model is constructed, which is beneficial to improving the authenticity and reliability of the subsequent simulation.
Planning a path refers to indicating the path of movement of the treatment device. The first planned path refers to a planned path set before radiation therapy begins. The treatment apparatus refers to an apparatus (e.g., CT machine, etc.) that performs radiation therapy. The anti-collision system is in communication with the treatment device.
In some embodiments, the anti-collision system generates motion control instructions based on the first planned path and transmits the motion control instructions to the treatment device to control the treatment device to move in the first planned path after radiation treatment begins to deliver radiation treatment to the patient.
In some embodiments, the anti-collision system uses the reference path as the first planned path by querying the reference path corresponding to the body surface data and the structure data in the path preset table based on the body surface data and the structure data. The path preset table is preset based on historical data and comprises a plurality of groups of body surface data, structure data and reference paths corresponding to each group of data.
The anti-collision system takes historical body surface data and historical structure data in the historical radiation treatment process as a group of data into a path preset table, and takes a first planning path which does not collide in the historical radiation treatment process as a reference path into a path planning table. Wherein the reference path includes coordinates of a plurality of spatial points arranged in sequence.
In some embodiments, the anti-collision system performs a simulation of anti-collision during radiation therapy based on the initial three-dimensional model, determining the first planned path.
Simulation refers to a simulation of the movement of the treatment device.
In some embodiments, to prevent a patient and/or support structure from colliding with the treatment device during radiation treatment, the collision avoidance system performs simulation by simulation software based on the plurality of candidate first paths, and selects any one of the candidate first paths that do not collide in the simulation as the first planned path. The simulation software comprises any feasible simulation software such as simulation software built in a system or external simulation software (such as ADAMS, solidWorks and the like).
The candidate first path refers to a planned path to be determined as the first planned path. In some embodiments, the collision avoidance system obtains the plurality of candidate first paths by way of user input or the like. For example, the candidate first paths include paths formed by connecting a plurality of spatial points in sequence, and the spatial points in different candidate first paths are different in position or partially the same. The user includes a doctor, etc. The user may manually set a plurality of planned paths as candidate first paths based on the patient's treatment needs. The treatment needs to include at least one of the amount of radiation or the duration of radiation required for each site, etc. at the site where radiation treatment is required.
In some embodiments, the process of the anti-collision system performing simulation based on a single candidate first path comprises the steps that the anti-collision system sets an initial three-dimensional model in simulation software, sets a motion path of virtual equipment based on the candidate first path, enables the virtual equipment to perform virtual motion in the simulation software, and records whether the virtual equipment collides with the initial three-dimensional model in the virtual motion process. The virtual device refers to a model in simulation software for simulating the treatment device.
In some embodiments, the collision avoidance system may further perform at least one simulation based on the initial three-dimensional model and the target path library, and determine the first planned path based on simulation results of the at least one simulation. See fig. 5 and its associated description for an explanation of this section.
In some embodiments of the present disclosure, a three-dimensional model capable of accurately reflecting body surface data of a patient is constructed based on complete body surface data of the patient, so that error influence of the body surface data can be effectively eliminated, and accuracy and efficiency of path planning are further effectively improved. Through simulation, the possible collision of the treatment equipment in the motion process can be effectively predicted, so that the motion path of the treatment equipment is adjusted, and the possibility of real collision with a patient is reduced.
In some embodiments, the anti-collision system may also acquire real-time scan data of the patient during the treatment, update the initial three-dimensional model based on the real-time scan data, obtain a real-time three-dimensional model, and perform a simulation based on the real-time three-dimensional model.
The real-time scan data refers to body surface data of the patient acquired during radiation therapy. When the posture of the patient changes during the radiotherapy, etc., the real-time scan data is inconsistent with the body surface data acquired before the radiotherapy is started.
In some embodiments, the real-time scan data may also include structural data acquired during radiation therapy.
In some embodiments, the data acquisition device may continuously acquire body surface data and structural data as real-time scan data during radiation therapy and send to the anti-collision system.
The real-time three-dimensional model is a three-dimensional model obtained after the initial three-dimensional model is updated.
In some embodiments, the collision avoidance system updates the initial three-dimensional model based on the real-time scan data via a real-time three-dimensional reconstruction technique (e.g., bundleFusion, etc.) to obtain a real-time three-dimensional model.
For example, for real-time scan data acquired by an optical camera, the anti-collision system may pre-process the real-time scan data, extract information related to the patient and/or the support structure in the image by means of image recognition and processing techniques, etc., convert the obtained information into three-dimensional information, and update the initial three-dimensional model by using a real-time three-dimensional reconstruction technique (such as BundleFusion, etc.), so as to obtain a real-time three-dimensional model. The preprocessing includes at least one of removing noise, adjusting resolution, cropping a region of interest, and the like. The region of interest includes the region of the patient where radiation therapy is desired, and the like.
In some embodiments, after the anti-collision system obtains the real-time three-dimensional model, the real-time three-dimensional model can be updated again through the method for updating the initial three-dimensional model based on the real-time scanning data acquired later.
In some embodiments, after the real-time three-dimensional model is obtained, the anti-collision system may also re-perform simulation based on the real-time three-dimensional model to determine whether a collision of the patient and/or support structure with the treatment device is likely.
In some embodiments of the present disclosure, by monitoring relevant information of a patient and/or a support structure in real time and updating an initial three-dimensional model based on the obtained information, accuracy of the three-dimensional model can be effectively improved, and further accuracy of simulation is improved, which is beneficial to judging subsequent operations.
In some embodiments, the anti-collision system may further determine a first collision risk based on the real-time three-dimensional model and the first planned path, generate early warning information and issue an early warning in response to the first collision risk meeting the early warning condition.
The first collision risk refers to a collision risk corresponding to the first future time. The risk of collision is used to characterize the likelihood of collision with the patient and/or the support structure while the treatment apparatus is in motion. In some embodiments, the collision risk is represented by a numerical value or the like, the greater the numerical value, the higher the collision risk.
The first future time refers to a future time after the real-time three-dimensional model is obtained.
In some embodiments, the first future time includes at least one future point in time. The number of future points in time is related to the movement characteristics of the treatment device.
The movement characteristics are used to characterize the movement of the treatment device. In some embodiments, the motion characteristics include at least one of a motion velocity, a motion acceleration, and the like. The number of future points in time may be positively correlated to the movement speed or movement acceleration, the greater the number of future points in time.
In some embodiments of the present disclosure, the number of future time points is related to the motion characteristics of the treatment device, and more future time points can be determined when the treatment device is running faster, so as to predict the risk of collision for a longer period of time in the future, which is beneficial for improving the safety of radiation treatment.
In some embodiments, the collision avoidance system determines the first collision risk in a plurality of ways based on the real-time three-dimensional model and the first planned path. For example, the anti-collision system executes simulation again based on the real-time three-dimensional model and the first planned path, counts the number of collisions occurring in the simulation process, and the greater the number of collisions occurring, the greater the first collision risk.
For another example, the anti-collision system performs feature vector conversion and/or fusion on the real-time three-dimensional model and the first planning path to obtain a path vector, clusters the path vector as a clustering center in a plurality of historical path vectors, and determines the first collision risk based on a vector set corresponding to the clustering center. The anti-collision system can extract a preset number of historical path vectors in the vector set, obtain historical collision results corresponding to the historical path vectors from the historical data, calculate the ratio of the number of times of collision in the historical collision results to the preset number, and take the ratio as a first collision risk. The preset data is preset based on historical experience.
The historical path vector is a characteristic vector obtained by performing characteristic vector conversion and/or fusion on the historical real-time three-dimensional model and the historical first planning path by the anti-collision system. The historical collision result corresponding to the historical path vector refers to a collision result actually occurring in the historical radiotherapy process corresponding to the historical path vector. The collision results include collision and non-collision.
In some embodiments, the collision avoidance system may further determine a first collision risk for each future point in time in the first future time by the risk prediction model based on the real-time three-dimensional model and the first planned path. See fig. 3 and its associated description for this section.
The early warning condition refers to a condition for judging whether early warning information is generated and giving out early warning. In some embodiments, the pre-warning conditions are preset based on historical experience. For example, the first collision risk is greater than the risk threshold. The risk threshold is preset based on historical experience.
In some embodiments, if the collision avoidance system determines the first collision risk for each of the first future time points by the risk prediction model, the pre-warning condition includes an average of the first collision risks for the plurality of future time points being greater than a risk threshold, and so on.
The early warning information refers to information related to the content from which the early warning is issued. For example, the user is alerted that the first collision risk is high or the patient is alerted not to tamper. In some embodiments, the manner in which the alert is issued includes any feasible manner such as a warning tone.
In some embodiments of the present disclosure, the collision risk at the future time is dynamically calculated through the real-time three-dimensional model and the first planned path, so that accuracy and timeliness of collision risk assessment are improved, early warning is sent to help timely take measures for avoiding collision, and safety of the patient and the treatment device is guaranteed.
In some embodiments, the anti-collision system may further correct the first planned path based on the real-time three-dimensional model to obtain a second planned path, and determine a second collision risk based on the real-time three-dimensional model and the second planned path.
In some embodiments, in response to the second collision risk meeting the pre-warning condition, the anti-collision system generates pre-warning information and issues pre-warning, and/or updates the second planned path.
The second planned path refers to a path indicating movement of the treatment device after the real-time three-dimensional model is obtained.
In some embodiments, the anti-collision system generates motion control instructions based on the second planned path and transmits the motion control instructions to the treatment device to control the treatment device to move in the second planned path to deliver radiation therapy to the patient.
In some embodiments, the anti-collision system corrects the first planned path based on the real-time three-dimensional model to obtain a plurality of candidate second paths, and performs simulation on each candidate second path to obtain a second planned path. The candidate second path refers to a planned path to be determined as the second planned path. The collision avoidance system may select any one of the candidate second paths in the simulation that do not collide as the second planned path.
In some embodiments, the collision avoidance system modifies the first planned path in a number of ways. For example, the collision avoidance system obtains the corrected first planned path (i.e., the candidate second path) via user input. The user can manually revise the first planning path, generate a plurality of candidate second paths and input the candidate second paths into the anti-collision system.
For another example, the anti-collision system may divide the first planned path into a plurality of spatial points based on the preset distance, and randomly adjust coordinates of one or more of the spatial points, thereby completing correction of the first planned path and obtaining a plurality of candidate second paths. The preset distance is preset based on historical experience.
The second collision risk refers to a collision risk corresponding to the second future time. The second future time refers to a future time after the second planned path is obtained. In some embodiments, the second future time is later than or equal to the first future time. The second future time includes at least one future point in time.
In some embodiments, the method of determining the second collision risk by the anti-collision system is similar to the method of determining the first collision risk, and a method of achieving this may be referred to as the method of determining the first collision risk.
In some embodiments, the pre-warning condition further includes the second collision risk being greater than a risk threshold, and so on. And responding to the second collision risk meeting the early warning condition, generating early warning information by the anti-collision system, and sending out early warning and/or updating the second planning path. The anti-collision system obtains an updated second planning path, and can further continuously determine a second collision risk corresponding to the updated second planning path, and update the second planning path again in response to the new second collision risk meeting the early warning condition.
In some embodiments, the collision avoidance system updates the second planned path in a number of ways. For example, the anti-collision system updates the second planned path in the manner described above to determine the second planned path. For another example, the anti-collision system directionally adjusts the second planned path based on the second planned path, and performs simulation on the adjusted second planned path, if no collision occurs in the simulation, the adjusted second planned path is used as a new second planned path, and if a collision occurs in the simulation, the second planned path is continuously directionally adjusted.
In some embodiments, directionally adjusting the second planned path includes the anti-collision system dividing the second planned path into a plurality of spatial points based on the preset distance, and adjusting coordinates of the spatial points that are prone to collision in a direction away from the patient, thereby obtaining an adjusted second planned path. Wherein the magnitude of the adjustment is preset based on historical experience. The anti-collision system may use a plurality of spatial points that are too close to the real-time three-dimensional model or that overlap as spatial points that are prone to collision.
In some embodiments of the present disclosure, the first planned path is modified based on the real-time three-dimensional model to obtain the second planned path, so as to dynamically adjust the movement path of the treatment device in the radiotherapy process, thereby improving accuracy, safety and intelligentization level of the radiotherapy process.
FIG. 3 is an exemplary schematic diagram of a risk prediction model shown in accordance with some embodiments of the present description.
In some embodiments, the collision avoidance system determines the first collision risk 350 via the risk prediction model 340 based on the real-time three-dimensional model 310 and the first planned path 230. For a description of the real-time three-dimensional model, the first planned path and the first collision risk, see fig. 2 and its associated description.
The risk prediction model refers to a model for determining a first collision risk, and in some embodiments, the risk prediction model may be a machine learning model. For example, the risk prediction model may include any one or combination of a convolutional neural network (Convolutional Neural Networks, CNN) model, a neural network (Neural Networks, NN) model, or other custom model structure, etc.
In some embodiments, the input of the risk prediction model comprises a real-time three-dimensional model and the first planned path, and the output comprises a first collision risk for each of the first future points in time.
In some embodiments, the collision avoidance system trains the risk prediction model by gradient descent or the like based on a plurality of first training samples with first tags. The first training samples include a sample real-time three-dimensional model and a sample first planned path, and the first label of the first training samples may be a collision risk of each sample at a future time point in a sample first future time corresponding to the first training samples.
In some embodiments, the first training sample and the first tag are determined based on historical data. For example, the anti-collision system counts a historical real-time three-dimensional model and a historical first planning path in the historical radiotherapy process as a first training sample, for the first training sample which does not actually collide, the labels of all time points corresponding to the first training sample are set to 0, for the first training sample which actually collides, the label of the time point which collides is set to 1, and the label of the time point which does not collide is set to a numerical value of 0-1. Wherein, the closer the time point of collision is, the closer the label thereof is to 1.
In some embodiments, the risk prediction model may be trained by inputting a plurality of first training samples with first labels into an initial risk prediction model, constructing a loss function from the first labels and a prediction result of the initial risk prediction model, updating the initial risk prediction model based on iteration of the loss function, and completing the risk prediction model training when the loss function of the initial risk prediction model satisfies a preset condition. The preset condition may be that the loss function converges, the number of iterations reaches a set value, etc.
In some embodiments, the input to the risk prediction model further includes a number of data dead zones 320 and a site activity 330 corresponding to each data dead zone.
The data blind area is a part of the body of the patient where the data acquisition device cannot acquire the body surface data. For example, a patient's body part that is occluded by the treatment device during movement of the treatment device.
In some embodiments, the collision avoidance system determines a data blind zone based on the real-time scan data. For example, if the data acquisition device is an optical camera, the anti-collision system determines body parts of the patient that are occluded by the treatment device based on the real-time scan data by means of image recognition and processing techniques, and determines such body parts as data blind areas. For another example, if the data acquisition device is a 3D scanning device, the anti-collision system may directly determine body parts of the patient that are occluded by the treatment device based on real-time scan data and determine such body parts as data dead zones due to differences in the three-dimensional point clouds obtained by scanning the human body and scanning the treatment device.
The site activity refers to the activity level of the site corresponding to the blind data zone of the patient. And the activity of the corresponding part of each data blind area is independently determined. See fig. 4 and the associated description for an illustration of how site liveness may be determined.
In some embodiments, if the input of the risk prediction model further includes the number of data dead zones and the portion liveness corresponding to each data dead zone, the first training sample further includes the number of sample data dead zones and the sample portion liveness corresponding to each sample data dead zone.
In some embodiments of the present disclosure, the position liveness of the data blind area is helpful to reflect the movement tendency of the position corresponding to the data blind area of the patient under the condition that no data support is provided, and the position liveness of the data blind area is added in the input of the risk prediction model, so that the accuracy of outputting the first collision risk is improved.
In some embodiments of the present disclosure, the risk prediction model is used to process data such as the real-time three-dimensional model and the first planned path, so that a rule can be found from a large amount of data by using the self-learning capability of the machine learning model, and the association relationship between the data such as the real-time three-dimensional model and the first planned path and the first collision risk is obtained, thereby improving the accuracy and efficiency of determining the first collision risk of the user.
Fig. 4 is an exemplary flow diagram of deriving a second planned path, shown in accordance with some embodiments of the present description. In some embodiments, the process 400 is performed by a radiation therapy collision avoidance system (hereinafter referred to as a collision avoidance system) based on three-dimensional scanning and real-time simulation.
At step 410, at least one secure buffer is determined.
The safety buffer refers to a space reserved for preventing a patient from colliding with the therapeutic device. For example, the safety buffer is a space obtained by expanding a three-dimensional model on a local outline by a certain distance, and thus the size of the safety buffer can be represented by the expansion distance of the safety buffer.
In some embodiments, the collision avoidance system determines the at least one safety buffer in a variety of ways. For example, the anti-collision system sets a safety buffer based on the model boundary and a preset space size. The model boundary refers to the boundary point of the initial three-dimensional model or the real-time three-dimensional model, and the edge formed by connecting the boundary points. The preset space size is preset based on historical experience, such as a 5cm space with model boundaries expanding outwards.
In some embodiments, the collision avoidance system may also determine at least one safety buffer based on the region characteristics of the data blind region. See fig. 3 and its associated description for an explanation of the data dead zone.
The region characteristics refer to information related to the data blind region. In some embodiments, the region features include patient body parts corresponding to the data blind spots, and the like. The regional characteristics are synchronously determined by the anti-collision system in the process of determining the dead zone of the data.
Illustratively, the anti-collision system queries the same reference region characteristics as the region characteristics in the preset characteristic table based on the region characteristics of the data blind region, and determines the reference buffer region corresponding to the reference region characteristics as the safety buffer region corresponding to the data blind region. Wherein, each data blind zone corresponds to a safe buffer zone.
The preset feature table is preset based on historical data and comprises a plurality of reference region features and reference buffer areas corresponding to the reference region features. In some embodiments, the anti-collision system counts the historical region features of the historical data shadow as reference region features during the historical radiation therapy and counts the historical safety buffer (e.g., 30cm of space with the data shadow outward) together into a preset feature table.
In some embodiments, the region features further comprise a region area of the data dead zone. The area is the area occupied by the part corresponding to the data blind area. The collision avoidance system may also determine at least one safety buffer based on the area of the region. Illustratively, the larger the area of the region, the larger the expansion distance of the safety buffer corresponding to the data blind zone.
In some embodiments of the present disclosure, the larger the data blind area, the higher the potential collision risk, and the size of the safety buffer area corresponding to the data blind area with a larger area is increased, which helps to further reduce the potential collision risk.
In some embodiments of the present disclosure, different safety buffers are set according to different data blind areas, so that flexibility of guaranteeing radiation therapy safety is improved.
In some embodiments, the anti-collision system may further determine a region activity corresponding to the data blind region based on the region characteristics, and determine at least one safety buffer based on the region characteristics and the region activity. The safety buffer area corresponding to each data blind area is determined based on the area characteristics and the part liveness of the data blind area.
In some embodiments, the anti-collision system determines an activity frequency and an activity amplitude of the patient body part before the data blind zone based on real-time scan data of the patient body part corresponding to the region feature, calculates a weighted sum of the activity frequency and the activity amplitude, and determines the weighted sum as the part liveness of the data blind zone. The activity frequency and the activity amplitude can be determined by any feasible mode such as image comparison, three-dimensional space point cloud change and the like.
In some embodiments, the anti-collision system queries the region preset table to determine at least one safety buffer based on the region characteristics and the region liveness. The region preset table is preset based on historical experience and comprises a plurality of groups of region characteristics and region liveness and the expansion distance of a safety buffer region corresponding to each group of region characteristics and region liveness. The expansion distance of the safety buffer area is positively related to the area of the data blind area and the position activity corresponding to the data blind area.
In some embodiments, the anti-collision system determines the expansion distance of the safety buffer area corresponding to the data blind area through a buffer prediction model based on the region characteristics and the part liveness.
The buffered predictive model refers to a model for determining the expansion distance of the safe buffer, and in some embodiments, may be a machine learning model. For example, the buffer prediction model may include any one or combination of a convolutional neural network (Convolutional Neural Networks, CNN) model, a neural network (Neural Networks, NN) model, or other custom model structure, etc.
In some embodiments, the collision avoidance system trains the buffer predictive model by a gradient descent method or the like based on a plurality of second training samples with second tags. The second training sample includes a sample region feature and a sample portion liveness, and the second label of the second training sample may be an actual expansion distance of the safety buffer corresponding to the second training sample.
In some embodiments, the second training samples and the second labels are determined based on historical data. For example, the anti-collision system counts the historical radiation treatment process with higher radiation treatment efficiency and no collision in the historical data, takes the regional characteristics and the part liveness corresponding to the historical data blind areas in the historical radiation treatment process as a second training sample, and takes the expansion distance of the safety buffer area actually used in the historical radiation treatment process as a second label. Wherein the radiation treatment efficiency can be represented by radiation treatment time, etc., the radiation quality efficiency is higher as the radiation treatment time is shorter.
In some embodiments, the training process of the buffer predictive model is similar to the training process of the risk predictive model, and the method of implementation is referred to as the training process of the risk predictive model.
In some embodiments of the present disclosure, the higher the site liveness is, the higher the potential collision risk is, and based on the site liveness, the size of the safety buffer region corresponding to the data blind region is determined in a plurality of ways, which is helpful to improve the efficiency of determining the safety buffer region and further reduce the potential collision risk.
In some embodiments, the anti-collision system performs step 420 to obtain a second planned path 430 based on the real-time three-dimensional model 310 and the at least one safety buffer acquired in step 410.
Step 420, the first planned path is modified.
In some embodiments, the anti-collision system fuses the safety buffer area and the real-time three-dimensional model through any feasible method such as point cloud union, so as to obtain a fused three-dimensional model, and corrects the first planning path based on the fused three-dimensional model, so as to obtain a second planning path. The method for correcting the first planned path based on the fused three-dimensional model is similar to the method for correcting the first planned path based on the real-time three-dimensional model, and the implementation method can be seen in fig. 2 and the related description.
In some embodiments of the present disclosure, by reserving a safety buffer, it is ensured that even if a patient experiences a slight unintended movement during radiation therapy, no collision will occur, improving the safety of the radiation therapy process.
Fig. 5 is an exemplary flow chart for determining a first planned path, according to some embodiments of the present description. In some embodiments, the process 500 is performed by a radiation therapy collision avoidance system (hereinafter referred to as a collision avoidance system) based on three-dimensional scanning and real-time simulation.
Step 510, determining a target path library.
The target landmark database refers to a database containing a plurality of planned paths. In some embodiments, the anti-collision system stores a plurality of target path libraries, each target path library corresponding to a class of treatment needs, and the anti-collision system determines the target path library corresponding to the class based on the current patient's treatment needs. See fig. 2 and the associated description for an illustration of the need for treatment.
In some embodiments, the construction process of the target path library comprises counting, for each treatment requirement, the same or similar historical treatment requirements as the treatment requirement in a large amount of historical data by the anti-collision system, determining a historical radiation treatment process corresponding to the historical treatment requirement as target historical data, determining target historical data with better radiation treatment effect in the target historical data as preferable historical data, taking a planning path used in the preferable historical data as a candidate planning path, and adding the planning path into the target path library. Wherein the radiation therapy effect is manually noted by the user based on historical experience.
The anti-collision system determines whether the treatment needs are similar to the historical treatment needs by converting the treatment needs to feature vectors, calculating vector similarity and the like based on the treatment needs and the historical treatment needs. Vector similarity is inversely related to vector distance. Vector distances include euclidean distances, and the like.
At step 520, at least one simulation is performed based on the initial three-dimensional model and the target path library.
In some embodiments, the process of single-pass simulation includes:
S1, the anti-collision system selects a candidate planning path from a target path library based on a preset selection sequence to serve as a target planning path for simulation. The preset selection sequence comprises selecting a candidate planning path with the highest historical use times from unselected candidate planning paths. The unselected candidate planning paths refer to candidate planning paths which are not selected in the current at least one simulation. The historical number of uses is determined by the collision avoidance system statistics history.
For example, the current at least one simulation has been performed 1 time, denoted as 1 st simulation, the current simulation is 2 nd simulation, if the target path library includes 5 paths (denoted as paths 1-5, respectively, and the historical usage times of the paths 1-5 are ordered from high to low), the target planned path selected by the 1 st simulation is path 1, the target planned path selected by the 2 nd simulation is path 2, and the unselected candidate planned paths include paths 3-5.
And S2, determining the test collision risk corresponding to the target planning path based on the target planning path and the initial three-dimensional model. The test collision risk is used to characterize the likelihood of collision with the patient and/or the support structure as the treatment apparatus moves along the target planned path. The method of determining the risk of a test collision is similar to the method of determining the risk of a first collision, and the method of implementation is described with reference to fig. 2 and its associated description.
And S3, responding to the fact that the test collision risk does not meet the ending condition, and performing next simulation. The end condition includes the test collision risk not being less than a risk threshold. For a description of risk thresholds, see fig. 2 and its associated description.
And S4, responding to the collision risk corresponding to the target planning path to meet the ending condition, and ending at least one simulation.
Step 530, determining a first planned path based on a simulation result of at least one simulation.
For more description of the first planned path see fig. 2 and its associated description.
In some embodiments, the anti-collision system determines a candidate planned path with the smallest test collision risk in the simulation result as the first planned path based on the simulation result of the at least one simulation. The simulation results include the test collision risk determined by each simulation.
In some embodiments of the present disclosure, in combination with a target path library required for treatment of a patient, an optimal first planned path is comprehensively evaluated and determined, improving the efficiency of determining the first planned path.
It should be noted that the above description of the flow 400 and the flow 500 is for illustration and description only, and is not intended to limit the scope of applicability of the present description. Various modifications and changes to the flow may be made by those skilled in the art under the guidance of this specification. However, such modifications and variations are still within the scope of the present description.
Some embodiments of the present description also provide a computer-readable storage medium storing computer instructions that, when read by a computer, perform the method of any one of the above embodiments.
Furthermore, certain features, structures, or characteristics of one or more embodiments of the present description may be combined as suitable.
In some embodiments, numbers describing the components, number of attributes are used, it being understood that such numbers being used in the description of embodiments are modified in some examples by the modifier "about," approximately, "or" substantially. Unless otherwise indicated, "about," "approximately," or "substantially" indicate that the number allows for a 20% variation. Accordingly, in some embodiments, numerical parameters set forth in the specification and claims are approximations that may vary depending upon the desired properties sought to be obtained by the individual embodiments. In some embodiments, the numerical parameters should take into account the specified significant digits and employ a method for preserving the general number of digits. Although the numerical ranges and parameters set forth herein are approximations that may be employed in some embodiments to confirm the breadth of the range, in particular embodiments, the setting of such numerical values is as precise as possible.
If the description, definition, and/or use of a term in this specification makes reference to a material that is inconsistent or conflicting with the disclosure provided herein, the description, definition, and/or use of the term in this specification controls.