CN-122025009-A - Heterogeneity correction model applied to radiation dose calculation

CN122025009ACN 122025009 ACN122025009 ACN 122025009ACN-122025009-A

Abstract

The specification discloses a heterogeneity correction model applied to radiation dose calculation, and relates to the technical field of radiation therapy. According to the above, the model includes an input layer, a block embedding layer, an encoder, a bottleneck layer, a decoder, and an output head. The bottleneck layer includes two feature modeling modules. The input layer is for receiving four-dimensional tensor data. The block embedding layer is used for dividing four-dimensional tensor data through the three-dimensional convolution layer to determine a plurality of small blocks to be processed. The encoder includes means for sampling a plurality of tiles to be processed to determine a high-level feature. The feature modeling module includes a continuous window multi-head attention sub-block and a shift window multi-head attention sub-block for extracting local and global features based on input features to determine high quality features. The output head is used for determining the heterogeneity correction coefficient based on the target feature. Therefore, the heterogeneity correction is carried out through the heterogeneity correction model, so that the calculation accuracy is improved, and the accuracy of radiation dose calculation is improved.

Inventors

LIANG RUNCHENG
LIU ZHAOXING
LIU XIN
SHI ZHONGYAN
JIN HAIJING
ZHAO RI
LIU LIYE
LI HUA
CHEN FAGUO
YANG BIAO
GUO RONG
ZHANG JING

Assignees

中国辐射防护研究院

Dates

Publication Date: 20260512
Application Date: 20251231

Claims (10)

1. The heterogeneity correction model applied to radiation dose calculation is characterized by comprising an input layer, a block embedding layer, an encoder, a bottleneck layer, a decoder and an output head, wherein the bottleneck layer comprises two feature modeling modules; the input layer is used for receiving four-dimensional tensor data, wherein the four-dimensional tensor data comprises a ray scanning matrix, a total energy release matrix, an anatomical structure sketching matrix and a treatment plan information matrix; The block embedding layer is used for dividing the four-dimensional tensor data through a three-dimensional convolution layer to determine a plurality of small blocks to be processed; the encoder comprises a plurality of continuous stages for respectively sampling the plurality of small blocks to be processed and determining high-level characteristics; The feature modeling module comprises a continuous window multi-head attention sub-block and a shift window multi-head attention sub-block, wherein the feature modeling module is used for extracting local and global features based on the input features and determining high-quality features; the decoder is connected with the encoder in a jumping way and is used for upsampling the high-quality characteristic to determine a target characteristic; the output head is used for determining a heterogeneity correction coefficient based on the target feature.
2. The heterogeneity correction model for radiation dose calculation according to claim 1, wherein the window multi-head attention sub-block comprises a normalization layer, an attention module, a normalization layer and a spatial multi-layer perceptron module which are sequentially arranged, wherein residual connection is adopted between the modules for information fusion; The shift window multi-head attention sub-block comprises a normalization layer, a shift window multi-head attention module, a normalization layer and a space multi-layer perceptron module which are sequentially arranged, wherein residual error connection is adopted between the modules for information fusion.
3. The heterogeneity correction model for use in radiation dose calculation of claim 2, wherein the spatial multi-layer perceptron module is provided with an axial depth convolution structure, the spatial multi-layer perceptron module comprising a first branch, a second branch, a third branch, and a fourth branch; The first branch comprises a vertical axis feature extractor, wherein the vertical axis feature extractor is used for rearranging input tensors, extracting height dimension data, combining width dimension data, depth dimension data and channel dimension data, applying global linear transformation or large-kernel convolution to the height dimension data, enabling any two positions on the dimension to be directly connected, and simulating scattering propagation characteristics of rays along a longitudinal axis of a human body; The second branch comprises a horizontal axis feature extractor, wherein the horizontal axis feature extractor is used for rearranging input tensors, extracting width dimension data, combining height dimension data, depth dimension data and channel dimension data, applying global linear transformation or large-kernel convolution to the width dimension data, enabling any two positions on the dimension to be directly connected, and simulating scattering propagation characteristics of rays along a transverse axis of a human body; The third branch comprises a depth axis feature extractor, wherein the depth axis feature extractor is used for rearranging input tensors, extracting depth dimension data, combining height dimension data, width dimension data and channel dimension data, applying global linear transformation or large-kernel convolution to the depth dimension data, enabling any two positions on the dimension to establish direct connection, and simulating scattering propagation characteristics of rays along a human depth axis; the fourth branch comprises a residual connection structure, wherein the residual connection structure is used for reserving high-frequency detail information of an input tensor.
4. A heterogeneity correction model for use in radiation dose calculation according to claim 3, wherein the first, second, third, and fourth branches are parallel architectures; The space multi-layer perceptron module is used for splicing the outputs of the first branch, the second branch, the third branch and the fourth branch in the channel dimension, reducing the dimension by using a three-dimensional convolution check of 1 multiplied by 1 to compress the number of channels.
5. The heterogeneity correction model for use in radiation dose calculation according to any one of claims 1-4, wherein the residual connection in the feature modeling module and/or the encoder is provided with a composite loss function of: Wherein, the In order to achieve a loss value, the value of the loss, For the voxel level loss, Is a gradient loss; 、 Are weight coefficients.
6. A heterogeneity correction model for use in radiation dose calculation according to claim 5, wherein the voxel level loss is determined by means of weighted mean square error, using the formula: Wherein, the In order to predict the dose matrix of the patient, The real dose matrix is obtained, and N is the total number of voxels; the gradient loss is determined by the following formula: Wherein, the An operator is detected for the image edge.
7. A radiation dose calculation method, characterized by using the heterogeneity correction model according to any one of claims 1-6 applied to radiation dose calculation, said method comprising: Acquiring CT images of a patient, an anatomical structure sketching file and treatment plan information; Determining a total energy release matrix according to the CT image of the patient, the anatomical structure sketching file and the treatment plan information; Inputting the CT image of the patient, the anatomical structure sketching file, the treatment plan information and the total energy release matrix into the heterogeneity correction model applied to radiation dose calculation, and determining a heterogeneity correction coefficient; and determining a final energy matrix according to the heterogeneity correction coefficient and the total energy release matrix.
8. A computer readable storage medium, characterized in that the storage medium stores a computer program which, when executed by a processor, implements the method of claim 7.
9. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the method of claim 7 when executing the program.
10. A computer program product, characterized in that instructions in the computer program product, when executed by a processor of an electronic device, cause the electronic device to perform the method of claim 7.

Description

Heterogeneity correction model applied to radiation dose calculation Technical Field The present disclosure relates to the field of radiation therapy, and in particular, to a heterogeneity correction model applied to radiation dose calculation. Background In radiotherapy, accurate evaluation of three-dimensional energy deposition distribution (i.e., absorbed dose) of high-energy X-rays (photons) generated by a medical Linac (Linac) in a patient is a key element in planning treatment. The accuracy of the dose calculation algorithm directly determines the balance of tumor control probability and normal tissue complications probability. With the evolution of radiotherapy technology, particularly the widespread use of Intensity modulated radiation therapy (IMRT-Modulated Radiation Therapy) and volume modulated rotational radiotherapy (Volumetric Modulated ARC THERAPY, VMAT), the fields become highly complex and are not only subject to dynamic modulation by Multi-leaf gratings (Multi-Leaf Collimator, MLC) but also require transmission in the complex anatomy of the human body. Currently, the mainstream methods of pencil beam, barrel string convolution, anisotropy and the like generally divide radiation dose calculation into two steps, namely firstly, considering each tissue and organ of a human body as soft tissue (density is similar to water), and performing dose calculation in a uniform medium. And then, carrying out tissue density heterogeneity correction on a dose calculation result according to the actual density of the actual tissue organ to obtain accurate radiotherapy radiation dose distribution. The prior art mainly utilizes the weighted result 'equivalent path' of the radiation incidence path and the tissue density along the way to carry out the tissue density heterogeneity correction, has lower calculation accuracy at the extreme density interface (such as the trachea-tumor interface and the periphery of the metal implant), and limits the accuracy of the calculation result of the radiation dose of radiotherapy. Disclosure of Invention The present specification provides a heterogeneity correction model for use in radiation dose calculation that at least partially addresses the above-identified problems with the prior art. The technical scheme adopted in the specification is as follows: the specification provides a heterogeneity correction model applied to radiation dose calculation, which comprises an input layer, a block embedding layer, an encoder, a bottleneck layer, a decoder and an output head, wherein the bottleneck layer comprises two feature modeling modules; the input layer is used for receiving four-dimensional tensor data, wherein the four-dimensional tensor data comprises a ray scanning matrix, a total energy release matrix, an anatomical structure sketching matrix and a treatment plan information matrix; The block embedding layer is used for dividing the four-dimensional tensor data through a three-dimensional convolution layer to determine a plurality of small blocks to be processed; the encoder comprises a plurality of continuous stages for respectively sampling the plurality of small blocks to be processed and determining high-level characteristics; The feature modeling module comprises a continuous window multi-head attention sub-block and a shift window multi-head attention sub-block, wherein the feature modeling module is used for extracting local and global features based on input features and determining high-quality features; the decoder is connected with the encoder in a jumping way and is used for upsampling the high-quality characteristic to determine a target characteristic; the output head is used for determining a heterogeneity correction coefficient based on the target feature. Preferably, the window multi-head attention sub-block comprises a normalization layer, an attention module, a normalization layer and a space multi-layer perceptron module which are sequentially arranged, wherein residual error connection is adopted between the modules for information fusion; The shift window multi-head attention sub-block comprises a normalization layer, a shift window multi-head attention module, a normalization layer and a space multi-layer perceptron module which are sequentially arranged, wherein residual error connection is adopted between the modules for information fusion. Preferably, the spatial multi-layer perceptron module is provided with an axial depth convolution structure, and comprises a first branch, a second branch, a third branch and a fourth branch; The first branch comprises a vertical axis feature extractor, wherein the vertical axis feature extractor is used for rearranging input tensors, extracting height dimension data, combining width dimension data, depth dimension data and channel dimension data, applying global linear transformation or large-kernel convolution to the height dimension data, enabling any two positions on the dimension to be directly connecte