High-energy light beam path optimization method and system for tumor radiotherapy equipment

Abstract

The invention relates to the technical field of tumor treatment, and discloses a high-energy light beam path optimization method and system for tumor radiotherapy equipment. The method comprises the following steps: acquiring CT data of a patient; inputting the CT data into a pre-trained deep convolutional neural network to obtain a tumor point position coordinate; according to the tumor point position coordinates, target region planning is carried out, and treatment target region coordinates are obtained; performing coordinate transformation according to the coordinates of the treatment target area, and calculating high-energy ray parameters to obtain ray directions, ray paths and ray doses; and the ray direction, the ray path and the ray dose are guided into medical equipment, so that the medical equipment works according to the ray direction, the ray path and the ray dose. According to the method, under the assistance of a deep learning algorithm, the contour of a target region and an organ can be automatically sketched, the coordinates of a tumor point are determined, the beam path, the beam direction and the ray dose are optimized, and more accurate treatment is given.

Description

Technical Field

[0001] The present invention relates to the technical field of tumor treatment, and in particular to a high-energy light beam path optimization method and system for tumor radiotherapy equipment. Background Art

[0002] High-energy beam therapy is a radiotherapy technique used to treat cancer. It uses high-energy radiation (such as X-rays or particle beams) to kill cancer cells and inhibit or eliminate tumors. The principle of this treatment method is that radiation can damage the DNA of cells, prevent the proliferation and growth of cancer cells, and thus achieve the purpose of treating tumors.

[0003] The current high-energy beam therapy process is generally the same. The radiation oncologist first manually outlines the target area and organs, then determines the coordinates of the tumor point, and then selects the appropriate beam path and beam direction, performs calculations, and finally performs treatment.

[0004] It is not difficult to see from the above steps that the target area and organs manually outlined by the doctor will produce errors. The selection of the beam path is also very critical to the entire treatment process, and the calculation of the beam path is the difficulty of the entire treatment. Therefore, the treatment process of the existing technology is cumbersome, and the accuracy of the treatment mainly depends on the doctor's experience and subjective judgment, which is prone to errors. At the same time, the manual selection of the beam path will also affect the treatment effect to a certain extent. Summary of the invention

[0005] The present invention provides a high-energy beam path optimization method and system for tumor radiotherapy equipment, which can automatically outline the target area and organ contours, determine the coordinates of the tumor point, optimize the beam path, beam direction and radiation dose, and provide more accurate treatment with the assistance of a deep learning algorithm.

[0006] In a first aspect, in order to solve the above technical problems, the present invention provides a high-energy beam path optimization method for a tumor radiotherapy device, comprising: Obtain CT data of the patient; Inputting the CT data into a pre-trained deep convolutional neural network to obtain the position coordinates of the tumor point; According to the coordinates of the tumor point location, target area planning is performed to obtain the coordinates of the treatment target area; According to the treatment target coordinates, coordinate transformation is performed to obtain the device target coordinates in the medical device coordinate system; According to the target area coordinates of the device, the incident direction of the ray is determined to obtain the ray direction; Determining a ray path of the ray passing through the human body according to the ray direction; Perform attenuation modeling according to the ray path to determine the ray dose; Inputting the ray direction, the ray path and the ray dose into a pre-trained beam path scoring model to obtain a beam path score value; When the beam path score value is greater than or equal to a preset score threshold, a control instruction is generated and sent to a medical device, so that the medical device works according to the ray direction, the ray path and the ray dose; When the beam path score value is less than a preset score threshold, the steps of determining the incident direction of the ray, determining the ray path of the ray passing through the human body, and determining the ray dose are re-executed.

[0007] Preferably, the training method of the pre-trained deep convolutional neural network comprises: Obtain medical imaging data of patients; Performing data preprocessing on the medical image data to obtain 3D data in a deep convolutional neural network format; Segmenting the 3D data based on a deep convolutional neural network to obtain 3D data labels; voxelize the 3D data according to the 3D data label to obtain voxel data; wherein the voxel data and the 3D data label have a mapping relationship; The voxel data is input into a deep convolutional neural network for iterative training, a loss value of a loss function of the deep convolutional neural network is set, and the training is determined to be completed when the loss value meets a condition, thereby obtaining a pre-trained deep convolutional neural network.

[0008] Preferably, the preprocessing of the medical image data to obtain 3D data in a deep convolutional neural network format includes: According to the medical image data, resolution transformation is performed based on an interpolation method to obtain first deep convolutional neural network format data; According to the medical image data, grayscale values ​​are uniformly processed based on grayscale normalization to obtain data in a second deep convolutional neural network format; Data fusion is performed according to the first deep convolutional neural network format data and the second deep convolutional neural network format data to obtain 3D data in the deep convolutional neural network format.

[0009] Preferably, segmenting the 3D data based on a deep convolutional neural network comprises: Inputting the 3D data into a convolutional layer of a deep convolutional neural network to extract image features to obtain a first feature image; Input the first feature image into the pooling layer and perform downsampling to obtain a second feature image; Inputting the second feature image into the residual layer and performing gradient transfer to obtain a third feature image; The third feature image is input into the deconvolution layer to perform size restoration to obtain the 3D data label.

[0010] Preferably, voxelizing the 3D data according to the 3D data label to obtain voxel data includes: Binarizing the 3D data labels to obtain a target area; voxelize the target area to obtain a voxel area; Convolution mapping and deconvolution mapping are performed on the voxel data to obtain the voxel data.

[0011] Preferably, the loss function is a binary cross entropy loss function; wherein the calculation formula of the binary cross entropy loss function is: ; In the formula, represents the loss value; represents the number of voxels; Indicates The actual label of the pixel, which is 0 or 1; Represents the probability predicted by the model, that is, The probability that a pixel belongs to the tumor area ranges from 0 to 1.

[0012] Preferably, the device target area coordinates are calculated by the following formula: ; In the formula, Indicates the target area coordinates of the device; represents the transformation matrix; represents the coordinates of the treatment target; represents the translation vector; The ray direction is calculated using the following formula: ; In the formula, Represents the coordinates of the tumor point in the target coordinate system of the device; Represents the coordinates of the ray source in the target coordinate system of the device; Indicates the direction of the ray; The radiation dose is determined by the following formula: ; In the formula, Indicates at time The radiation dose at the time Indicates the initial dose; Indicates at time Tissue attenuation coefficient at ; Represents the ray at time The length of the path traversed.

[0013] Preferably, the training process of the beam path scoring model includes: Acquire historical data sets including historical ray directions, historical ray paths, and historical ray doses; Performing data preprocessing on the historical data set to obtain initial ray data; The initial ray data is used as input data, an initial beam path scoring model is constructed based on a particle swarm algorithm, and the initial beam path scoring model is trained according to a membership matrix and a possibility partitioning matrix to obtain a beam path scoring model.

[0014] In a second aspect, the present invention provides a high-energy beam path optimization system for a tumor radiotherapy device, which is used to implement any one of the above-mentioned high-energy beam path optimization methods for a tumor radiotherapy device, including: A data acquisition module, used for acquiring CT data of patients; A coordinate acquisition module, used to input the CT data into a pre-trained deep convolutional neural network to obtain the coordinates of the tumor point position; The target area planning module is used to plan the target area according to the location coordinates of the tumor point and obtain the treatment target area coordinates; A coordinate calculation module, used to perform coordinate transformation according to the treatment target coordinates to obtain the device target coordinates in the medical device coordinate system; A ray direction calculation module is used to determine the ray incident direction according to the target area coordinates of the device to obtain the ray direction; A ray path determination module, used to determine the ray path of the ray passing through the human body according to the ray direction; A radiation dose determination module, used for performing attenuation modeling according to the radiation path to determine the radiation dose; A ray scoring module, inputting the ray direction, the ray path and the ray dose into a pre-trained beam path scoring model to obtain a beam path score value; The path optimization control module generates a control instruction and sends it to the medical device when the beam path score value is greater than or equal to a preset score threshold, so that the medical device works according to the ray direction, the ray path and the ray dose; when the beam path score value is less than the preset score threshold, the steps of determining the ray incident direction, determining the ray path of the ray through the human body and determining the ray dose are re-executed.

[0015] In a third aspect, the present invention also provides an electronic device comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, the high-energy beam path optimization method for tumor radiotherapy equipment described in any one of the above is implemented.

[0016] In a fourth aspect, the present invention further provides a computer-readable storage medium, the computer-readable storage medium comprising a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute any one of the above-mentioned high-energy beam path optimization methods for tumor radiotherapy equipment.

[0017] Compared with the prior art, the present invention has the following beneficial effects: the embodiment of the present invention provides a high-energy beam path optimization method and system for tumor radiotherapy equipment. The method includes: obtaining CT data of the patient; inputting the CT data into a pre-trained deep convolutional neural network to obtain the position coordinates of the tumor point; performing target area planning according to the position coordinates of the tumor point to obtain the treatment target area coordinates; performing coordinate transformation according to the treatment target area coordinates to obtain the device target area coordinates in the medical device coordinate system; determining the incident direction of the ray according to the device target area coordinates to obtain the ray direction; determining the ray path of the ray passing through the human body according to the ray direction; performing attenuation modeling according to the ray path to determine the ray dose; introducing the ray direction, the ray path and the ray dose into the medical device so that the medical device works according to the ray direction, the ray path and the ray dose.

[0018] In the present invention, the method can use a 3D convolutional neural network to segment the patient's clinical data, segment the tumor point and the range of the target area, determine the tumor point; then calculate the beam direction, path and path time distribution information of the high-energy beam; and finally use it for clinical radiotherapy. With the assistance of a deep learning algorithm, the method can automatically outline the target area and organ contours, determine the coordinates of the tumor point, optimize the beam path, beam direction and radiation dose, and provide more accurate treatment. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a flow chart of a high-energy beam path optimization method for tumor radiotherapy equipment provided by the first embodiment of the present invention; Figure 2 It is a schematic diagram of the structure of a high-energy beam path optimization system for tumor radiotherapy equipment provided by the second embodiment of the present invention. DETAILED DESCRIPTION

[0020] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0021] Reference Figure 1 The first embodiment of the present invention provides a high-energy beam path optimization method for a tumor radiotherapy device, comprising the following steps: S11, acquiring CT data of the patient; S12, inputting the CT data into a pre-trained deep convolutional neural network to obtain the position coordinates of the tumor point; S13, performing target area planning according to the position coordinates of the tumor point to obtain the treatment target area coordinates; S14, performing coordinate transformation according to the treatment target coordinates to obtain the device target coordinates in the medical device coordinate system; S15, determining the incident direction of the ray according to the target area coordinates of the device to obtain the ray direction; S16, determining a ray path of the ray passing through the human body according to the ray direction; S17, performing attenuation modeling according to the ray path to determine the ray dose; S18, inputting the ray direction, the ray path and the ray dose into a pre-trained beam path scoring model to obtain a beam path score value; S19, when the beam path score value is greater than or equal to a preset score threshold, a control instruction is generated and sent to the medical device, so that the medical device works according to the ray direction, the ray path and the ray dose; when the beam path score value is less than the preset score threshold, the steps of determining the ray incident direction, determining the ray path of the ray through the human body and determining the ray dose are re-executed.

[0022] It is worth noting that high-energy beam therapy for tumors is a radiotherapy technology used to treat cancer. It uses high-energy radiation (such as X-rays or particle beams) to kill cancer cells and inhibit or eliminate tumors. The principle of this treatment method is that radiation can damage the DNA of cells and prevent the proliferation and growth of cancer cells, thereby achieving the purpose of treating tumors.

[0023] In order to facilitate the understanding of the present invention, some preferred embodiments of the present invention are further described below.

[0024] In step S11, the method for obtaining the patient's CT data may be: scanning with a conventional computed tomography (CT) device to obtain the patient's three-dimensional tomographic image data. This is one of the most common medical imaging methods, which can provide a detailed three-dimensional structure of the patient's internal tissue. When acquiring CT data, in order to ensure the integrity of the image, CT scans at different angles can be used to shoot the patient from multiple directions to obtain a full range of image data. This method helps to build a three-dimensional model more accurately. During CT scanning, appropriate resolution and layer thickness (such as 0.5 mm layer thickness) are usually set to ensure that the acquired image is clear and fine. Images with higher resolution can be better used for subsequent segmentation and target area planning. For different types of treatments, the range of CT data to be obtained is different. For example, for head treatment, it is usually necessary to obtain CT images from the top of the head to the neck, while for systemic treatment, CT data from head to toe is required.

[0025] It is worth mentioning that after the CT scan, the acquired data needs to be standardized, such as unifying images at different levels into a standard format and optimizing image quality, including denoising and grayscale normalization, so that it can be used for subsequent deep learning models.

[0026] In step S12, the CT data is input into a pre-trained deep convolutional neural network to obtain the position coordinates of the tumor point. Preferably, the training method of the pre-trained deep convolutional neural network includes: obtaining the patient's medical imaging data; performing data preprocessing on the medical imaging data to obtain 3D data in a deep convolutional neural network format; segmenting the 3D data based on the deep convolutional neural network to obtain 3D data labels; voxelizing the 3D data according to the 3D data labels to obtain voxel data; wherein the voxel data and the 3D data labels have a mapping relationship; inputting the voxel data into the deep convolutional neural network for iterative training, setting the loss value of the deep convolutional neural network loss function, and determining that the training is completed when the loss value meets the conditions to obtain a pre-trained deep convolutional neural network.

[0027] Specifically, the medical imaging data may include: CT (computed tomography) image data. The three-dimensional images provided by CT scans can accurately display the patient's internal anatomical structure, including bones, soft tissues, blood vessels, etc., which are the most commonly used imaging data in tumor radiotherapy; MRI (magnetic resonance imaging) data. MRI images are more advantageous than CT in displaying soft tissue details (such as the brain, spinal cord, etc.), and are often used to accurately locate tumors, especially tumors in the brain and spine; PET (positron emission tomography) image data. PET scans can display the metabolic activity of tumors. By using radioactive tracers, abnormal changes in cell activity can be detected, such as high metabolic activity of cancer cells; X-ray imaging data. Although X-ray imaging is usually used to obtain two-dimensional planar images, in some cases it can assist in determining the position of bones and other hard tissue structures, which is helpful for target location in radiotherapy; ultrasound imaging data. Ultrasound imaging can be used to obtain real-time soft tissue dynamic information, especially in cases where dynamic monitoring of tumor or organ movement is required (such as displacement caused by breathing).

[0028] It is worth noting that different medical imaging devices (such as CT, MRI, PET, etc.) may generate images in different formats, and the resolution, size, and pixel depth may also be inconsistent. The preprocessing step can standardize these images, such as adjusting all images to a uniform resolution (such as 512×512) or a standardized voxel format (Voxel), so that these data are suitable for the input format of deep convolutional neural networks. In other words, medical imaging data often comes from different sources, and images from different hospitals or equipment may differ in grayscale, contrast, etc. Through data preprocessing, operations such as grayscale normalization can be performed to normalize data from different sources so that they meet the learning requirements of the network.

[0029] Preferably, the data preprocessing of the medical image data to obtain 3D data in a deep convolutional neural network format includes: performing resolution transformation based on an interpolation method according to the medical image data to obtain data in a first deep convolutional neural network format; performing grayscale value uniform processing based on grayscale normalization according to the medical image data to obtain data in a second deep convolutional neural network format; and performing data fusion according to the first deep convolutional neural network format data and the second deep convolutional neural network format data to obtain 3D data in the deep convolutional neural network format.

[0030] Specifically, interpolation is a method of estimating unknown data points between known data points. Its basic principle is to infer the function value between these points through given known data points. In many fields such as image processing and data analysis, interpolation is used to enlarge, reduce, resample or complete data. In the preprocessing of medical image processing and deep learning models, interpolation is often used to adjust the resolution of the image or convert the data format to match the input requirements of the network. Common interpolation methods are: linear interpolation, which is suitable for relatively smooth changes between data points; bilinear interpolation, which is an extended linear interpolation method in two-dimensional space, often used for image enlargement or reduction; cubic interpolation, which uses cubic polynomial functions for interpolation, can provide a smoother transition effect than linear interpolation and bilinear interpolation. In the present invention, interpolation is widely used in operations such as image scaling, rotation, and resampling to ensure that the image is consistent with the input requirements of the deep learning model. Specifically, medical images (such as CT, MRI) usually have different resolutions, and convolutional neural networks usually require fixed-size input images. For example, the resolution of the image is converted from 300×300 to 512×512 through interpolation to ensure that the image can adapt to the input requirements of the model.

[0031] Specifically, Grayscale Normalization is an image processing technique used to standardize the grayscale values ​​(pixel intensity values) in an image to a specific range, usually to enhance the contrast of the image or to adapt to subsequent processing tasks, such as the input requirements of a deep learning model. By standardizing the pixel grayscale values, it can be ensured that each image has a similar grayscale distribution, thereby reducing the deviation caused by grayscale differences, and helping to improve the stability and training effect of subsequent models. In the present invention, the grayscale values ​​of CT or MRI images may represent different densities or tissue types. For example, the grayscale value range of a CT image may be very wide, and it is usually normalized according to different needs so that the model can better distinguish between tumors, normal tissues, and other structures.

[0032] Based on the interpolation method, the resolution is transformed to obtain the first deep convolutional neural network format data, and the grayscale value is uniformly processed based on grayscale normalization to obtain the second deep convolutional neural network format data. According to the first deep convolutional neural network format data and the second deep convolutional neural network format data, data fusion is performed to obtain the 3D data in the deep convolutional neural network format. It is worth noting that the 3D data is three-dimensional medical imaging data that meets the input requirements of the deep convolutional neural network (DCNN) after preprocessing. This type of data usually comes from three-dimensional imaging devices such as CT and MRI, and after processing steps such as format conversion, it can be effectively understood and processed by the convolutional neural network.

[0033] Preferably, the segmentation of the 3D data based on the deep convolutional neural network includes: inputting the 3D data into the convolution layer of the deep convolutional neural network, performing image feature extraction, and obtaining a first feature image; inputting the first feature image into the pooling layer, performing downsampling, and obtaining a second feature image; inputting the second feature image into the residual layer, performing gradient transfer, and obtaining a third feature image; inputting the third feature image into the deconvolution layer, performing size restoration, and obtaining the 3D data label.

[0034] Among them, the convolution layer is the core part of the convolutional neural network. It uses a set of convolution kernels (filters) to slide on the 3D image and extract the spatial features of the image by performing convolution operations on the local area of ​​the image; the pooling layer, also known as the downsampling layer, usually uses maximum pooling or average pooling to reduce the spatial size of the feature map; the residual layer is a special layer used to solve the gradient disappearance problem in deep neural networks. It passes the input directly to the subsequent layers through skip connections to ensure that the network can effectively pass the gradient during training; the deconvolution layer (also called the transposed convolution layer) is used to restore the downsampled feature map to the original resolution. Deconvolution is the inverse process of convolution. It increases the size of the image by interpolation or padding to restore it to the same size as the input image.

[0035] Specifically, the main purpose of the convolution layer is to capture different features in the input 3D image, such as edges, textures, shapes, etc. In 3D medical images, these features help identify the boundaries, locations, and volumes of tumors. The result of the convolution operation is the first feature image, which is a feature map containing the spatial feature information of the image. Its size is slightly smaller than the original image because the convolution process introduces boundary effects. The main function of pooling is to reduce the dimension of the data and reduce the amount of calculation while retaining important features. After downsampling, the resolution of the image will be reduced, but the key features are retained, which is helpful for subsequent feature extraction. The pooling operation generates a second feature image, which is smaller in size but retains the main feature information in the first feature image. The residual layer uses jump connections to enable the model to retain the original information in the deep structure, avoiding information loss or gradient reduction in the deep network after multiple convolution operations. This can improve the training efficiency of the model and enhance the feature extraction ability of the model. The output of the residual layer is the third feature image, which still retains the key features after multiple convolutions, and due to the role of the residual structure, the performance and training effect of the network are improved. The main purpose of the deconvolution layer is to resample the low-resolution feature map after downsampling and feature extraction to the size of the original image in order to generate a segmentation result consistent with the original image. In medical image segmentation, deconvolution is used for pixel-by-pixel classification to classify each pixel (or voxel) as a target area (such as a tumor) or a non-target area. The output of the deconvolution layer is the restored 3D data label, that is, the category label corresponding to each pixel. These labels indicate the classification result of each pixel belonging to a tumor, normal tissue, or other category. In general, the process of the segmentation operation can be summarized as follows: first, the input 3D medical image data is subjected to feature extraction through the convolution layer to extract the main spatial information; the feature map is downsampled through the pooling layer to reduce the dimension of the data and retain the key feature information; the residual layer helps the network maintain feature information and effective gradient transfer through jump connections, so that the deep network can be trained more stably; finally, the low-resolution feature map is restored to its original size through the deconvolution layer, and the classification result of each voxel is output to achieve accurate segmentation of tumors and normal tissues.

[0036] Preferably, voxelizing the 3D data according to the 3D data label to obtain voxel data includes: binarizing the 3D data label to obtain a target area; voxelizing the target area to obtain a voxel area; and performing convolution mapping and deconvolution mapping on the voxel data to obtain the voxel data.

[0037] Specifically, binarization is the process of converting each pixel (or voxel) in an image to 0 or 1 according to a certain threshold. In medical images, it is usually used to distinguish target areas (such as tumors) from background areas (such as normal tissue). In 3D data labels, each voxel contains classification information, indicating whether the voxel belongs to the target area (such as a tumor) or the non-target area (such as the background). Through the binarization operation, all voxels belonging to the target area are assigned a value of 1, and voxels in the non-target area are assigned a value of 0. The binarized data only contains two values ​​0 and 1, forming a clear target area, that is, the tumor area or other areas of interest in the image are marked as 1, and the rest are 0.

[0038] Specifically, voxelization refers to the standardization and unification of each voxel in the three-dimensional data to ensure that each voxel has a consistent size and position in the three-dimensional space. The purpose of voxelization is to convert three-dimensional data into a standard format suitable for deep learning network processing. During the voxelization process, the binarized target area is remapped to a three-dimensional voxel grid, and each voxel represents a specific spatial volume (such as 1 cubic millimeter). The voxelization process may involve interpolation operations to ensure the resolution of the original image and the consistency of the data. After voxelization, a three-dimensional voxel area is generated, which consists of a series of voxels of the same size, and each voxel represents a part of the target area (for example, a tumor area).

[0039] Specifically, convolution mapping is the process of extracting features from voxel data using a convolutional neural network. The convolution layer extracts spatial features by applying convolution kernels to the voxel data. The convolution operation can capture key features in the tumor area, such as boundaries, shapes, and structures. The purpose of convolution mapping is to pass the input voxel data through multiple layers of convolution kernels to generate feature maps, which represent different levels of features of the data, such as local spatial information and global spatial relationships of voxels. Deconvolution mapping is the inverse process of convolution, which is mainly used to restore the low-resolution feature map after convolution to the original resolution. Through deconvolution, the feature map generated by convolution mapping can be remapped to the size of the original voxel. Deconvolution mapping restores the convolution mapped data to the same resolution as the voxelized data through interpolation and upsampling operations, thereby retaining the key spatial information in the data. The combined result of convolution mapping and deconvolution mapping is voxel data with high resolution and accurate spatial information. These data not only retain the spatial features of the target area, but also extract more deep features through convolution operations, which can be used for subsequent tasks such as segmentation, classification or path planning. In general, the above steps can be summarized as follows: Through the binarization operation, the scope of the target area (tumor) is determined by thresholding the 3D data labels, forming a simple and clear binary image. The voxelization process standardizes the target area into a consistent three-dimensional voxel grid, so that these data can effectively represent the volume and position of the target area in three-dimensional space. Convolution mapping and deconvolution mapping further process the voxel data. Convolution mapping extracts the features of the voxel data, while deconvolution mapping restores the resolution of the voxel data to ensure that spatial consistency is maintained while retaining the feature information. Finally, the obtained voxel data contains both the spatial structure information of the target area and the deep feature information processed by the neural network. These data can be used for further medical analysis or treatment planning.

[0040] Preferably, the loss function is a binary cross entropy loss function; wherein the calculation formula of the binary cross entropy loss function is: ; In the formula, represents the loss value; represents the number of voxels; Indicates The actual label of the pixel, which is 0 or 1; Represents the probability predicted by the model, that is, The probability that a pixel belongs to the tumor area ranges from 0 to 1.

[0041] Specifically, the binary cross-entropy loss function (BCE) is a commonly used loss function for binary classification problems, and is particularly suitable for processing binary classification tasks, such as determining whether each pixel or voxel belongs to the target area (such as a tumor) or the background area (such as normal tissue). In deep learning, the binary cross-entropy loss function helps the model gradually optimize its parameters by calculating the difference between the output predicted by the model and the true label, so that it can be classified more accurately during the training process. In the present invention, the optimization process of binary cross entropy is: during the forward propagation process of the model, the input data passes through the neural network to generate a predicted probability value , according to the true label and predicted values , calculate the loss value of each sample. The overall loss of the model is the average loss of all samples. The loss value is used for back propagation, calculates the gradient of each parameter, updates the network weights, and gradually reduces the loss value, so that the model's prediction is more accurate. The present invention uses a binary cross entropy loss function to evaluate the difference between the model's prediction and the true label. Specifically, the model's prediction result for each voxel or pixel is a probability value, which represents the possibility that the voxel belongs to the target area (such as a tumor). By comparing the probability value with the true label (tumor or normal tissue), the loss value is calculated. The calculation result of the loss function can be used to update the weight of the model through the gradient descent method (or other optimization algorithms). As the training progresses, the loss value gradually decreases, and the model's ability to classify voxels gradually improves. Ultimately, this optimization process helps the model to more accurately segment tumor areas and normal tissues. After multiple iterations of training, the model can better learn the difference between tumors and normal tissues, and ultimately achieve high-precision segmentation results.

[0042] In step S13, target area planning is performed according to the coordinates of the tumor point position to obtain the coordinates of the treatment target area. Specifically, the steps of target area planning may be: determining the area that needs to be irradiated based on the tumor range that can be clearly seen in the medical image; determining the clinical target volume, which refers to the area where tumor cells may spread, usually slightly larger than the tumor volume; planning the target volume, which is the area that needs to be considered during actual radiotherapy. Due to factors such as breathing and movement of the patient during treatment, PTV will add a safety margin on the basis of CTV to ensure that the radiation can accurately cover the entire tumor area even when the patient moves.

[0043] In step S14, coordinate transformation is performed according to the treatment target coordinates to obtain the device target coordinates in the medical device coordinate system.

[0044] Preferably, the device target area coordinates are calculated by the following formula: ; In the formula, Indicates the target area coordinates of the device; represents the transformation matrix; represents the coordinates of the treatment target; represents the translation vector;

[0045] Specifically, in step S13, the treatment target coordinates refer to the spatial positioning of the tumor or lesion area in the patient's body. Since the coordinate system of the medical device and the coordinate system of the treatment target are not completely consistent, a transformation matrix H is required to transform the coordinate system. In the coordinate transformation process, the coordinates also need to be translated, that is, the origin position of the coordinates is adjusted. This translation amount is represented by a vector T, which ensures that the converted device target coordinates are consistent with the actual working area of ​​the device.

[0046] Specifically, the acquisition of transformation matrix and translation vector usually depends on the relationship between two coordinate systems: treatment target coordinate system and device target coordinate system. These parameters are obtained through geometric transformation, calibration or calculation. The specific methods include the following: First, it can be done through geometric transformation. During the treatment process, the relationship between the coordinate system of the treatment target and the coordinate system of the device includes geometric transformations such as rotation, scaling, and misalignment. The transformation matrix can be determined by describing the rotation, scaling, and translation between the coordinate systems. The rotation matrix is ​​usually used to represent the angular difference between two coordinate systems. The calculation formula is as follows: ; Where Rij represents the element in the rotation matrix, which is used to represent the rotation angle of the coordinate axis.

[0047] The second method is to use an optimization algorithm to obtain it. For example, a least squares numerical optimization method is used to estimate the matrix and offset vector through multiple calculations and adjustment of parameters.

[0048] In addition, it can also be obtained through automatic learning based on deep learning. The model can be trained with a large amount of medical imaging data to automatically learn the relationship between the target coordinates and the device coordinates. During the training process, the model will automatically learn the transformation matrix and translation vector, and optimize the rotation matrix and translation vector by minimizing the error between the predicted coordinates and the actual device coordinates.

[0049] In step S15, the incident direction of the ray is determined according to the coordinates of the target area of ​​the device to obtain the ray direction.

[0050] Preferably, the ray direction is calculated by the following formula: ; In the formula, Represents the coordinates of the tumor point in the target coordinate system of the device; Represents the coordinates of the ray source in the target coordinate system of the device; Indicates the direction of the ray.

[0051] Specifically, is the ray direction, which is a unit vector indicating the direction from the ray source to the tumor point. A unit vector means that the magnitude of the vector is 1, indicating a pure direction without distance information. The coordinates of the tumor point in the device target coordinate system represent the spatial position of the target area (tumor or lesion). These coordinates are obtained through medical imaging technology (such as CT scanning) and have been transformed and optimized. It is the coordinate of the ray source in the device target coordinate system, indicating the position of the ray source in the device coordinate system. It refers to the radiation source of the radiotherapy device, or the beam source used to irradiate the tumor.

[0052] It is worth mentioning that Representation vector The modulus of represents the magnitude of the vector, which represents the actual distance from the ray source to the tumor point. The purpose of the formula is to calculate the direction of the ray from the source point (ray source) to the tumor point. In order to make the direction calculation unaffected by the actual distance, the result is a unit vector (i.e., direction vector) with a length of 1. Through this unit vector, the medical device can determine the direction of the ray.

[0053] In step S16, a ray path of the ray passing through the human body is determined according to the ray direction.

[0054] In one achievable manner, the use of ray tracing and algorithms in computer graphics can be used to determine the ray path of the ray through the human body based on the ray direction. Specifically, ray tracing is a technique that simulates light (or rays) propagating from a source point in a specified direction until it interacts with an object (such as human tissue). In radiotherapy, this technique can help determine the propagation path of the ray and calculate its interaction with various tissues in the human body. In order to determine the path of the ray, it is first necessary to build a three-dimensional model of the human body in a computer, usually constructed through medical imaging data (such as CT scans, MRI, etc.). The model includes all anatomical structures of the patient, such as skin, muscle, bone, tumor and other tissues, each of which has different physical properties (such as density and attenuation coefficient). Among them, the CT image or MRI scan creates a three-dimensional volume model through slice data. Each pixel (or voxel) in these volume data has a specific density value, representing different types of tissues (such as bone, soft tissue, etc.).

[0055] During the propagation of rays, the rays will interact with the tissues of the human body. The key to ray tracing is to determine whether the rays intersect with certain tissues (that is, whether the rays pass through certain tissues). This can be achieved by checking whether the rays enter a new voxel or tissue area. In one feasible way, the human body model is divided into small voxel units, each of which contains certain tissue characteristics (such as attenuation coefficient). Each time the ray moves forward, it is calculated whether it intersects with the boundary of the current voxel. If the ray intersects with the boundary of the voxel, then the ray passes through the voxel. The intersection test between the ray and the voxel is used to determine whether the ray passes through the current tissue. Use the ray-voxel intersection detection algorithm in computer graphics, such as the Bresenham algorithm or the DDA (digital differential analysis) algorithm. When the ray intersects with the voxel, the ray position is updated and the calculation continues. Every time a voxel is passed, the path of the ray is recalculated until the ray passes through all target tissues or hits the tumor target. At this time, the path of the ray can be determined.

[0056] In step S17, attenuation modeling is performed according to the ray path to determine the ray dose.

[0057] Preferably, the radiation dose is determined by the following formula: ; In the formula, Indicates at time The radiation dose at the time Indicates the initial dose; Indicates at time Tissue attenuation coefficient at ; Represents the ray at time The length of the path traversed.

[0058] Specifically, For in time When the radiation passes through the human body, the energy of the radiation will gradually decrease with the change of propagation distance and tissue type. is the initial dose, which indicates the initial dose emitted by the radiation source, and indicates the energy intensity of the radiation when it is emitted from the source point. For in time The tissue attenuation coefficient at the time of radiation. Each tissue has a different ability to absorb radiation. The attenuation coefficient indicates the degree of energy attenuation of the radiation per unit length. This attenuation coefficient varies with time (or the path of radiation propagation) and tissue type. Different tissues (such as skin, muscle, bone, tumor, etc.) have different responses to the attenuation of radiation, so the attenuation coefficient will vary depending on the type of tissue the path passes through. For the ray at time The distance that a ray travels is an important factor in determining the attenuation of a ray. As time goes by, the ray continues to move forward, so the path length increases with time. When the distance that a ray travels in a tissue increases, the energy of the ray will gradually decrease according to the attenuation characteristics of the tissue.

[0059] It is worth noting that this formula can calculate in real time the dose attenuation of radiation when it passes through different tissues during treatment, and can provide accurate dose data for medical equipment, thereby optimizing the treatment effect and minimizing damage to healthy tissues.

[0060] In step S18, the training process of the beam path scoring model includes: Acquire historical data sets including historical ray directions, historical ray paths, and historical ray doses; Performing data preprocessing on the historical data set to obtain initial ray data; The initial ray data is used as input data, an initial beam path scoring model is constructed based on a particle swarm algorithm, and the initial beam path scoring model is trained according to a membership matrix and a possibility partitioning matrix to obtain a beam path scoring model.

[0061] Specifically, the data set is preprocessed, such as removing noise points, filling missing points, normalizing, etc., and then the covariance matrix is ​​calculated to obtain a correlation covariance matrix; the correlation covariance matrix is ​​eigenvalue decomposition is performed to obtain eigenvalues ​​and eigenvectors, and a preset number of eigenvectors are selected in descending order of eigenvalues ​​to obtain an eigenvector matrix; the data set that has undergone data preprocessing is projected onto the eigenvector matrix to obtain reduced-dimensional data.

[0062] Furthermore, this embodiment combines the Possibility Fuzzy C-Means (PFCM) algorithm with the Particle Swarm Optimization (PSO) algorithm, and uses the improved PSO algorithm to optimize the initial center and number of the PFMC algorithm. Initialize parameters such as the number of particles, the number of iterations, the learning coefficient, the number of clusters, the fuzzy weighted index, etc.; perform full gradient inertial particle swarm optimization according to the number of clusters and the number of particles to obtain the optimal initial cluster center and the optimal initial number of clusters; initialize the parameters, membership matrix and possibility partition matrix of the initial beam path scoring model according to the optimal initial cluster center and the optimal initial number of clusters; update and train the cluster center according to the current membership matrix and the possibility partition matrix; when the number of training times is greater than or equal to the preset maximum number of training times, determine that the training is completed, and obtain the beam path scoring model. Among them, the initial clustering parameters of the PFCM algorithm are highly random, and the initial value of the number of clusters c determines the performance of the algorithm; the PSO algorithm is a global evolutionary algorithm based on particles for solving multi-objective, nonlinear, and multivariable problems; the method of the present invention introduces the PFCM algorithm into the membership rectangular matrix and the possibility partition matrix to describe the clustering results, and solves the problem that the initialization parameters of the PFCM algorithm are difficult to select and are prone to fall into local optimality by combining the PSO algorithm, and uses the improved PSO algorithm to optimize the initial center and number of the PFCM algorithm. The preset maximum number of training times is a key parameter, which determines when the training process of the possibility fuzzy C clustering is terminated. In this embodiment, the preset maximum number of training times is set to 200 times, which is a result verified by experiments.

[0063] In the embodiment of the present invention, the calculation formula of the covariance matrix is: ; in, A normalized matrix representing the data; represents the transpose of a normalized matrix; represents the degrees of freedom of the sample covariance; Represents the covariance matrix. The covariance matrix is ​​a square matrix that reflects the correlation between various parameters. Eigenvalue decomposition is the process of decomposing the covariance matrix into eigenvalues ​​and eigenvectors. The eigenvalue represents the variance of each principal component, and the eigenvector represents the direction of the principal component. The purpose of selecting eigenvectors is to retain the most important principal components, and a preset number of eigenvectors are selected in order from large to small eigenvalues. Usually, eigenvectors that can explain most of the variation are selected. In this embodiment, the preset number of eigenvectors is 4. Depending on the application scenario and user needs, the preset number of eigenvectors can also be set to 5, 10, etc., and the present invention is not limited to this.

[0064] It is worth noting that the training process based on the membership matrix and the possibility partition matrix includes: performing distance calculation based on the membership matrix and the possibility partition matrix to obtain the distance from each data point to all cluster centers; for each cluster center, performing contribution value calculation based on the membership matrix to obtain the position contribution value of each data point to each cluster center; according to the position contribution value, weighted summing the position contribution values ​​of all data points to obtain the second cluster center; according to the second cluster center, recalculating the membership of all data points to obtain the second membership matrix and the second possibility partition matrix; replacing the cluster center, the membership matrix and the possibility partition matrix with the second cluster center, the second membership matrix and the second possibility partition matrix respectively. The membership matrix is ​​a two-dimensional matrix, in which each row corresponds to a data point and each column corresponds to a cluster center. Each element in the matrix represents the membership or probability of the data point belonging to a certain cluster center. The membership value is usually between 0 and 1, indicating the degree to which the data point belongs to the cluster center. Like the membership matrix, the possibility partition matrix is ​​also a two-dimensional matrix, but it is usually used to represent the uncertainty or possibility that a data point belongs to a certain cluster center. The elements in the possibility partition matrix represent the possibility or confidence that a data point belongs to a certain cluster center. It reflects the possibility that a data point belongs to each cluster center given the membership matrix.

[0065] The membership is updated by calculating the following formula: ; in, Indicates The data points belong to The class membership, Indicates The data points belong to The class membership, Indicates The data points belong to The likelihood of the class splitting value, Indicates The data points belong to The likelihood of the class splitting value, represents the number of clusters; The possibility partition matrix is ​​updated by calculating through the following formula: ; in, Indicates The data points belong to The likelihood of the class split, Represents the data point and The Euclidean distance between cluster centers, Indicates The scale parameter of the class, represents the likelihood weighted index.

[0066] Preferably, after obtaining the beam path scoring model, the ray direction, the ray path and the ray dose are input into the beam path scoring model, and the beam path scoring value is output and compared with the score threshold.

[0067] In step S19, when the beam path score value is greater than or equal to the preset score threshold, a control instruction is generated and sent to the medical device, so that the medical device works according to the ray direction, the ray path and the ray dose. When the beam path score value is less than the preset score threshold, the steps of determining the ray incident direction, determining the ray path of the ray passing through the human body and determining the ray dose are re-executed.

[0068] It is worth noting that the score threshold is a key parameter that defines the minimum acceptable level of the beam path score value; in a specific implementation, the score ranges from 0 to 100, where 100 indicates the best beam path and 0 indicates the worst beam path. A score of 90 or more indicates a good beam path, 70 to 89 indicates an average beam path, 60 to 69 indicates a poor beam path, and a score of 60 or less indicates an unqualified beam path. The preset score threshold is set to 60. If the beam path score value exceeds 60, a control instruction is generated and sent to the medical device so that the medical device works according to the ray direction, the ray path, and the ray dose.

[0069] It is worth noting that if the beam path score value is less than the preset score threshold, it means that the beam path is unqualified, and the production control instruction re-determines the incident direction of the ray, determines the ray path of the ray through the human body, and determines the ray dose. Exemplarily, the path can be optimized by changing the incident angle of the ray or using different incident points to reduce damage to normal tissue while ensuring that the tumor area receives sufficient dose; or, the path length can be reduced by adjusting the nodes on the ray path or changing the layout of the path, thereby reducing the patient's treatment time. For example, a "Z-shaped" scan or other more efficient scanning mode can be used; or, the dose distribution can be optimized by changing the dwell time of each scanning point or adjusting the setting of the energy layer to ensure that the tumor target area receives sufficient dose while minimizing the dose to the surrounding normal tissue.

[0070] The following describes the working process of the present invention by taking a relatively common scenario as an example. The working process is as follows:

[0071] First, the doctor takes a CT scan of the patient to obtain three-dimensional image data of the body, which contains information about the density and morphology of the tissue. Then, the CT data is input into a pre-trained deep convolutional neural network, which automatically identifies the location of the tumor and determines the coordinates of the tumor point.

[0072] Based on the three-dimensional coordinates of the tumor, the system will perform target area planning and calculate the coordinates of the treatment target area. This target area not only includes the area where the tumor is located, but also needs to consider the need to protect the surrounding healthy tissue. Then, the system converts the coordinates of the treatment target area into the coordinate system of the medical device through coordinate transformation.

[0073] After completing the coordinate transformation, the system calculates the incident direction of the ray based on the coordinates of the device's target area. The direction of the ray points from the ray source to the location of the tumor, and this direction tells the device how to adjust the position and angle of the ray source. Next, the system determines the path of the ray, taking into account the situation when the ray passes through different tissues, such as skin, muscle, bone and other different types of tissue, and calculates the length of each path segment and the attenuation of the ray on each path segment.

[0074] In order to ensure the accuracy of treatment, the system also performs radiation attenuation modeling. Radiation will lose some energy when passing through different tissues of the human body. The system calculates the dose of radiation after passing through different tissues by modeling these attenuations.

[0075] Finally, all the radiation direction, path and dose data will be imported into the treatment device, and the medical device will accurately adjust the angle, path and intensity of the radiation source based on this information to achieve accurate radiotherapy.

[0076] In summary, the present invention provides a high-energy beam path optimization method for a tumor radiotherapy device, comprising:

[0077] Acquire the patient's CT data; input the CT data into a pre-trained deep convolutional neural network to obtain the coordinates of the tumor point position; perform target area planning based on the tumor point position coordinates to obtain the treatment target area coordinates; perform coordinate transformation based on the treatment target area coordinates to obtain the device target area coordinates in the medical device coordinate system; determine the incident direction of the ray based on the device target area coordinates to obtain the ray direction; determine the ray path of the ray through the human body based on the ray direction; perform attenuation modeling based on the ray path to determine the ray dose; input the ray direction, the ray path and the ray dose into a pre-trained beam path scoring model to obtain a beam path score value, compare the beam path score value with a preset score threshold, and issue a corresponding control instruction to enable the medical device to work according to the ray direction, the ray path and the ray dose.

[0078] In the present invention, the method can use a 3D convolutional neural network to segment the patient's clinical data, segment the tumor point and the range of the target area, determine the tumor point; then calculate the beam direction, path and path time distribution information of the high-energy beam; and finally use it for clinical radiotherapy. With the assistance of a deep learning algorithm, the method can automatically outline the target area and organ contours, determine the coordinates of the tumor point, optimize the beam path, beam direction and radiation dose, and provide more accurate treatment.

[0079] Reference Figure 2 The second embodiment of the present invention provides a high-energy beam path optimization system for a tumor radiotherapy device, comprising: A data acquisition module, used for acquiring CT data of patients; A coordinate acquisition module, used to input the CT data into a pre-trained deep convolutional neural network to obtain the coordinates of the tumor point position; The target area planning module is used to plan the target area according to the location coordinates of the tumor point and obtain the treatment target area coordinates; A coordinate calculation module, used to perform coordinate transformation according to the treatment target coordinates to obtain the device target coordinates in the medical device coordinate system; A ray direction calculation module is used to determine the ray incident direction according to the target area coordinates of the device to obtain the ray direction; A ray path determination module, used to determine the ray path of the ray passing through the human body according to the ray direction; A radiation dose determination module, used for performing attenuation modeling according to the radiation path to determine the radiation dose; A ray scoring module, inputting the ray direction, the ray path and the ray dose into a pre-trained beam path scoring model to obtain a beam path score value; The path optimization control module generates a control instruction and sends it to the medical device when the beam path score value is greater than or equal to a preset score threshold, so that the medical device works according to the ray direction, the ray path and the ray dose; when the beam path score value is less than the preset score threshold, the steps of determining the ray incident direction, determining the ray path of the ray through the human body and determining the ray dose are re-executed.

[0080] It should be noted that the high-energy beam path optimization system for tumor radiotherapy equipment provided in an embodiment of the present invention is used to execute all the process steps of the high-energy beam path optimization method for tumor radiotherapy equipment in the above-mentioned embodiment. The working principles and beneficial effects of the two correspond one to one, and thus will not be repeated here.

[0081] An embodiment of the present invention further provides an electronic device. The electronic device includes: a processor, a memory, and a computer program stored in the memory and executable on the processor, such as a high-energy beam path optimization method program for tumor radiotherapy equipment. When the processor executes the computer program, the steps in the above-mentioned high-energy beam path optimization method for tumor radiotherapy equipment are implemented, such as Figure 1 Alternatively, when the processor executes the computer program, the functions of the modules / units in the above-mentioned device embodiments are realized, such as the ray calculation module.

[0082] Exemplarily, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to implement the present invention. The one or more modules / units may be a series of computer program instruction segments capable of implementing specific functions, which are used to describe the execution process of the computer program in the electronic device.

[0083] The electronic device may be a computing device such as a desktop computer, a notebook, a PDA, and a smart tablet. The electronic device may include, but is not limited to, a processor and a memory. Those skilled in the art will appreciate that the above components are merely examples of electronic devices and do not constitute a limitation on the electronic device. The electronic device may include more or fewer components than the above components, or may combine certain components, or different components. For example, the electronic device may also include input and output devices, network access devices, buses, etc.

[0084] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the electronic device, and uses various interfaces and lines to connect various parts of the entire electronic device.

[0085] The memory can be used to store the computer program and / or module, and the processor realizes various functions of the electronic device by running or executing the computer program and / or module stored in the memory, and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; the data storage area can store data created according to the use of the mobile phone (such as audio data, a phone book, etc.), etc. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0086] Wherein, if the module / unit integrated in the electronic device is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Wherein, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electric carrier signals and telecommunication signals.

[0087] It should be noted that the device embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. In addition, in the accompanying drawings of the device embodiments provided by the present invention, the connection relationship between the modules indicates that there is a communication connection between them, which may be specifically implemented as one or more communication buses or signal lines. A person of ordinary skill in the art may understand and implement it without paying any creative effort.

[0088] The specific embodiments described above further illustrate the purpose, technical solutions and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. It is particularly pointed out that for those skilled in the art, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the scope of protection of the present invention.

Claims

1. A high-energy beam path optimization method for tumor radiotherapy equipment, characterized in that: include: Obtain CT data of the patient; Inputting the CT data into a pre-trained deep convolutional neural network to obtain the position coordinates of the tumor point; According to the coordinates of the tumor point location, target area planning is performed to obtain the coordinates of the treatment target area; According to the treatment target coordinates, coordinate transformation is performed to obtain the device target coordinates in the medical device coordinate system; According to the target area coordinates of the device, the incident direction of the ray is determined to obtain the ray direction; Determining a ray path of the ray passing through the human body according to the ray direction; Perform attenuation modeling according to the ray path to determine the ray dose; Inputting the ray direction, the ray path and the ray dose into a pre-trained beam path scoring model to obtain a beam path score value; When the beam path score value is greater than or equal to a preset score threshold, a control instruction is generated and sent to a medical device, so that the medical device works according to the ray direction, the ray path and the ray dose; When the beam path score value is less than a preset score threshold, the steps of determining the incident direction of the ray, determining the ray path of the ray passing through the human body, and determining the ray dose are re-executed.

2. The high-energy beam path optimization method for tumor radiotherapy equipment according to claim 1, characterized in that: The training method of the pre-trained deep convolutional neural network comprises: Obtain medical imaging data of patients; Performing data preprocessing on the medical image data to obtain 3D data in a deep convolutional neural network format; Segmenting the 3D data based on a deep convolutional neural network to obtain 3D data labels; voxelize the 3D data according to the 3D data label to obtain voxel data; wherein the voxel data and the 3D data label have a mapping relationship; The voxel data is input into a deep convolutional neural network for iterative training, a loss value of a loss function of the deep convolutional neural network is set, and the training is determined to be completed when the loss value meets a condition, thereby obtaining a pre-trained deep convolutional neural network.

3. The high-energy beam path optimization method for tumor radiotherapy equipment according to claim 2, characterized in that: The preprocessing of the medical image data to obtain 3D data in a deep convolutional neural network format includes: According to the medical image data, performing resolution transformation based on an interpolation method to obtain first deep convolutional neural network format data; According to the medical image data, grayscale values ​​are uniformly processed based on grayscale normalization to obtain data in a second deep convolutional neural network format; Data fusion is performed according to the first deep convolutional neural network format data and the second deep convolutional neural network format data to obtain 3D data in the deep convolutional neural network format.

4. The high-energy beam path optimization method for tumor radiotherapy equipment according to claim 2, characterized in that: The segmenting of the 3D data based on a deep convolutional neural network includes: Inputting the 3D data into a convolutional layer of a deep convolutional neural network to extract image features to obtain a first feature image; Input the first feature image into the pooling layer and perform downsampling to obtain a second feature image; Inputting the second feature image into the residual layer and performing gradient transfer to obtain a third feature image; The third feature image is input into the deconvolution layer to perform size restoration to obtain the 3D data label.

5. The high-energy beam path optimization method for tumor radiotherapy equipment according to claim 2, characterized in that: The step of voxelizing the 3D data according to the 3D data label to obtain voxel data includes: Binarizing the 3D data labels to obtain a target area; voxelize the target area to obtain a voxel area; Convolution mapping and deconvolution mapping are performed on the voxel data to obtain the voxel data.

6. The high-energy beam path optimization method for tumor radiotherapy equipment according to claim 2, characterized in that: The loss function is a binary cross entropy loss function; wherein the calculation formula of the binary cross entropy loss function is:

7. In the formula, represents the loss value; represents the number of voxels; Indicates The actual label of the pixel, which is 0 or 1; Represents the probability predicted by the model, that is, The probability that a pixel belongs to the tumor area ranges from 0 to 1.

8. The high-energy beam path optimization method for tumor radiotherapy equipment according to claim 1, characterized in that: The target coordinates of the device are calculated using the following formula:

9. In the formula, Indicates the target area coordinates of the device; represents the transformation matrix; represents the coordinates of the treatment target; represents the translation vector; The ray direction is calculated using the following formula:

10. In the formula, Represents the coordinates of the tumor point in the target coordinate system of the device; Represents the coordinates of the ray source in the target coordinate system of the device; Indicates the direction of the ray; in, The radiation dose is determined by the following formula:

11. In the formula, Indicates at time The radiation dose at the time Indicates the initial dose; Indicates at time Tissue attenuation coefficient at ; Represents the ray at time The length of the path traversed.

12. The high-energy beam path optimization method for tumor radiotherapy equipment according to claim 1, characterized in that: The training process of the beam path scoring model includes: Acquire historical data sets including historical ray directions, historical ray paths, and historical ray doses; Performing data preprocessing on the historical data set to obtain initial ray data; The initial ray data is used as input data, an initial beam path scoring model is constructed based on a particle swarm algorithm, and the initial beam path scoring model is trained according to a membership matrix and a possibility partitioning matrix to obtain a beam path scoring model.

13. A high-energy beam path optimization system for tumor radiotherapy equipment, characterized in that: A method for optimizing the path of a high-energy beam for a tumor radiotherapy device according to any one of claims 1 to 8, comprising: A data acquisition module, used for acquiring CT data of patients; A coordinate acquisition module, used to input the CT data into a pre-trained deep convolutional neural network to obtain the coordinates of the tumor point position; The target area planning module is used to plan the target area according to the location coordinates of the tumor point and obtain the treatment target area coordinates; A coordinate calculation module, used to perform coordinate transformation according to the treatment target coordinates to obtain the device target coordinates in the medical device coordinate system; A ray direction calculation module is used to determine the ray incident direction according to the target area coordinates of the device to obtain the ray direction; A ray path determination module, used to determine the ray path of the ray passing through the human body according to the ray direction; A radiation dose determination module, used for performing attenuation modeling according to the radiation path to determine the radiation dose; A ray scoring module, inputting the ray direction, the ray path and the ray dose into a pre-trained beam path scoring model to obtain a beam path score value; The path optimization control module generates a control instruction and sends it to the medical device when the beam path score value is greater than or equal to a preset score threshold, so that the medical device works according to the ray direction, the ray path and the ray dose; when the beam path score value is less than the preset score threshold, the steps of determining the ray incident direction, determining the ray path of the ray through the human body and determining the ray dose are re-executed.

14. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored computer program, wherein when the computer program is executed, the device where the computer-readable storage medium is located is controlled to execute the high-energy beam path optimization method for tumor radiotherapy equipment as described in any one of claims 1 to 8.

Images