Heterogeneity correction model applied to radiation dose calculation
By using a heterogeneity correction model based on deep learning, the problem of insufficient accuracy in dose calculation at the density interface in radiotherapy was solved, thereby improving the accuracy of dose calculation and treatment planning.
Patent Information
- Application Number
- CN202512040829.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-31
- Publication Date
- 2026-05-12
AI Technical Summary
In current radiotherapy, the accuracy of dose calculation at density interfaces is insufficient, especially at the trachea-tumor interface and around metal implants, which affects the accuracy of radiotherapy radiation dose calculation.
A heterogeneity correction model is adopted, including an input layer, a block embedding layer, an encoder, a bottleneck layer, a decoder, and an output head. Heterogeneous scattering correction is performed through a deep learning model. The Swin-UNETR model is used to learn the nonlinear mapping between the uncorrected total energy released per unit mass and the Monte Carlo standard dose, extracting local and global features to determine the heterogeneity correction coefficients.
It improves the precision and accuracy of radiation dose calculation, especially the dose distribution calculation of radiation fields in complex anatomical structures, and enhances the accuracy of treatment planning.
Smart Images
Figure CN122025009A_ABST
Abstract
Description
Technical Field
[0001] This specification relates to the field of radiotherapy technology, and in particular to a heterogeneity correction model applied to radiation dose calculation. Background Technology
[0002] In radiotherapy, accurately assessing the three-dimensional energy deposition distribution (i.e., absorbed dose) of high-energy X-rays (photons) generated by a medical linear accelerator (Linac) within the patient's body is a core aspect of treatment planning. The accuracy of the dose calculation algorithm directly determines the balance between the probability of tumor control and the probability of complications in normal tissues.
[0003] With the evolution of radiotherapy technology, especially the widespread application of intensity-modulated radiation therapy (IMRT) and volumetric modulated arc therapy (VMAT), the radiation field has become highly complex and is not only subject to dynamic modulation by multi-leaf collimators (MLCs), but also needs to be transmitted through the complex anatomical structures of the human body.
[0004] Current mainstream methods, such as pencil beam, tube-string convolution, and anisotropic methods, typically divide radiation dose calculation into two steps: First, all human tissues and organs are treated as soft tissues (with a density approximately similar to water), and dose calculations are performed in a homogeneous medium. Then, based on the actual density of the tissues and organs, the dose calculation results are corrected for tissue density heterogeneity to obtain an accurate radiation dose distribution for radiotherapy.
[0005] Existing technologies mainly use the weighted result of the incident radiation path and the tissue density along the path, known as the "equivalent path," to correct for tissue density heterogeneity. However, the accuracy of calculations is low at extreme density interfaces (such as the trachea-tumor interface or around metal implants), which limits the accuracy of radiotherapy radiation dose calculations. Summary of the Invention
[0006] This specification provides a heterogeneity correction model for radiation dose calculation, which at least partially solves the aforementioned problems existing in the prior art.
[0007] The following technical solution is adopted in this specification: This specification provides a heterogeneity correction model for radiation dose calculation, including 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 used to receive four-dimensional tensor data; the four-dimensional tensor data includes a ray scanning matrix, a total energy release matrix, an anatomical structure delineation matrix, and a treatment plan information matrix; The block embedding layer is used to divide the four-dimensional tensor data through a three-dimensional convolutional layer to determine multiple small blocks to be processed; The encoder includes multiple consecutive stages for sampling the multiple small blocks to be processed to determine high-level features; The feature modeling module includes continuous window multi-head attention sub-blocks and shifted window multi-head attention sub-blocks; the feature modeling module is used to extract local and global features based on the input features and determine high-quality features; The decoder is hoppedly connected to the encoder and is used to upsample the high-quality features to determine the target features; The output head is used to determine the heterogeneity correction coefficient based on the target features.
[0008] Preferably, the multi-head attention sub-block of the window includes a normalization layer, an attention module, a normalization layer, and a spatial multilayer perceptron module arranged in sequence, and the aforementioned modules are fused by residual connections; The shift window multi-head attention sub-block includes a normalization layer, a shift window multi-head attention module, a normalization layer, and a spatial multilayer perceptron module arranged in sequence. The aforementioned modules are fused using residual connections.
[0009] Preferably, the spatial multilayer perceptron module is provided with an axial depth convolution structure, and the spatial multilayer perceptron module includes a first branch, a second branch, a third branch and a fourth branch; The first branch includes a vertical axis feature extractor; the vertical axis feature extractor is used to rearrange the input tensor, extract the height dimension data, and merge the width dimension data, depth dimension data, and channel dimension data; apply a global linear transformation or large kernel convolution to the height dimension data so that any two positions in this dimension can establish a direct connection, simulating the scattering and propagation characteristics of rays along the longitudinal axis of the human body; The second branch includes a horizontal axis feature extractor; the horizontal axis feature extractor is used to rearrange the input tensor, extract the width dimension data, and merge the height dimension data, depth dimension data, and channel dimension data; apply a global linear transformation or large kernel convolution to the width dimension data so that any two positions in this dimension can establish a direct connection, simulating the scattering and propagation characteristics of rays along the horizontal axis of the human body; The third branch includes a depth axis feature extractor; the horizontal axis feature extractor is used to rearrange the input tensor, extract the depth dimension data, and merge the height dimension data, width dimension data, and channel dimension data; apply a global linear transformation or large kernel convolution to the depth dimension data so that any two positions in this dimension can establish a direct connection, simulating the scattering and propagation characteristics of rays along the human body's depth axis; The fourth branch includes a residual connection structure; the residual connection structure is used to preserve high-frequency detail information of the input tensor.
[0010] Preferably, the first branch, the second branch, the third branch, and the fourth branch are in a parallel architecture; The spatial multilayer perceptron module is used to concatenate the outputs of the first branch, the second branch, the third branch, and the fourth branch in the channel dimension, and use a 1×1×1 three-dimensional convolution kernel to reduce the dimensionality of the concatenated features and compress the number of channels.
[0011] Preferably, the feature modeling module and / or the residual connections in the encoder are configured with the following composite loss function: in, The loss value. For voxel-level loss, For gradient loss; , All are weighting coefficients.
[0012] Preferably, the voxel-level loss is determined using a weighted mean square error method, and is calculated using the following formula: in, For the predicted dose matrix, This is the actual dose matrix, where N is the total number of voxels; The gradient loss is determined by the following formula: in, This is an image edge detection operator.
[0013] On the other hand, this specification provides a method for calculating radiation dose, utilizing the heterogeneity correction model for radiation dose calculation provided in the above-mentioned aspect, the method comprising: Acquire patient CT images, anatomical structure delineation files, and treatment plan information; The total energy release matrix is determined based on the patient's CT images, the anatomical structure delineation file, and the treatment plan information. The patient's CT images, the anatomical structure delineation file, the treatment plan information, and the total energy release matrix are input into the heterogeneity correction model applied to radiation dose calculation to determine the heterogeneity correction coefficients. The final energy matrix is determined based on the heterogeneity correction coefficient and the total energy release matrix.
[0014] On the other hand, the computer-readable storage medium provided in this specification stores a computer program that, when executed by a processor, implements the heterogeneity correction model for radiation dose calculation provided in the above aspect.
[0015] On the other hand, this specification provides an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the heterogeneity correction model for radiation dose calculation provided in one aspect above.
[0016] On the other hand, this specification provides a computer program product in which instructions, when executed by a processor of an electronic device, cause the electronic device to implement the heterogeneity correction model for radiation dose calculation provided in the above-mentioned aspect.
[0017] The above-mentioned technical solutions adopted in this specification can achieve the following beneficial effects: Based on the above, the heterogeneity correction model applied to radiation dose calculation 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 receives four-dimensional tensor data, including a radiation scan matrix, a total energy release matrix, an anatomical structure delineation matrix, and a treatment plan information matrix. The block embedding layer partitions the four-dimensional tensor data using a three-dimensional convolutional layer to determine multiple small blocks to be processed. The encoder includes multiple consecutive stages for sampling each of these small blocks to determine high-level features. The feature modeling modules include consecutive windowed multi-head attention sub-blocks and shifted windowed multi-head attention sub-blocks. These modules extract local and global features based on the input features to determine high-quality features. The decoder is skipped to the encoder to upsample these high-quality features and determine the target features. The output head determines the heterogeneity correction coefficients based on the target features.
[0018] It is evident that heterogeneity correction through a heterogeneity correction model improves computational accuracy and the precision of radiation dose calculation. Attached Figure Description
[0019] The accompanying drawings, which are included to provide a further understanding of this specification and form part of this specification, illustrate exemplary embodiments and are used to explain this specification, but do not constitute an undue limitation thereof. In the drawings: Figure 1 A schematic diagram of the architecture of a heterogeneity correction model for radiation dose calculation provided as an embodiment of this specification; Figure 2A schematic diagram of the architecture of a heterogeneity correction model for radiation dose calculation provided for one embodiment of this specification; Figure 3 A schematic flowchart illustrating a radiation dose calculation method provided as an embodiment of this specification; Figure 4 The embodiment provided in this specification corresponds to Figure 1 A schematic diagram of the structure of an electronic device. Detailed Implementation
[0020] To make the objectives, technical solutions, and advantages of this specification clearer, the technical solutions of this specification will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this specification, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments in this specification without creative effort are within the scope of protection of this application.
[0021] In the description of this invention, it should be noted that the term "or" is generally used to include the meaning of "and / or" unless otherwise expressly stated in the content.
[0022] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. Furthermore, in the description of this application, the terms "first," "second," etc., are used only for distinguishing descriptions and should not be construed as indicating or implying relative importance.
[0023] Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.
[0024] The technical solutions provided in the various embodiments of this specification are described in detail below with reference to the accompanying drawings.
[0025] Figure 1 This is a schematic diagram of the architecture of a heterogeneity correction model for radiation dose calculation provided as an embodiment of this specification. Figure 2 A schematic diagram of the architecture of a heterogeneity correction model for radiation dose calculation, provided as an embodiment of this specification, is shown below. Figure 1 as well as Figure 2 As shown, the heterogeneity correction model applied to radiation dose calculation 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.
[0026] Preferably, the heterogeneity correction model for radiation dose calculation provided in this specification can decouple the dose calculation task into two sub-tasks.
[0027] First, analytical methods are used to perform geometric and basic physical calculations, employing a mature and extremely fast ray tracing algorithm to calculate the "uncorrected total energy released per unit mass (TERMA)" or "water equivalent dose." This step explicitly encodes all treatment planning parameters: beam angle, multi-leaf collimator (MLC) shape, and monitor unit (MU) hop count. This solves the problem that end-to-end deep learning (DL) models cannot effectively extract beam features.
[0028] Second, heterogeneous scattering correction is performed using a deep learning model. Leveraging the powerful global context modeling capabilities of Swin-UNETR, it learns a nonlinear mapping from "Uncorrected Total Energy Released per unit mass (TERMA) + Computed Tomography (CT) density" to "Monte Carlo Gold standard dose". The network no longer needs to learn "how rays propagate and how energy is deposited in matter", but only needs to focus on learning "how different densities of matter change the dose distribution".
[0029] Preferably, the heterogeneity correction model used for radiation dose calculation is the Swin-UNETR model.
[0030] Preferably, the input layer is used to receive four-dimensional tensor data.
[0031] Preferably, the four-dimensional tensor data includes a ray scanning matrix, a total energy release matrix, an anatomical structure delineation matrix, and a treatment plan information matrix.
[0032] Preferably, the X-ray scanning matrix is determined from the patient's CT image, the anatomical structure delineation matrix is determined from the anatomical structure delineation file, the treatment plan information matrix is determined from the treatment plan information, and the total energy release matrix is determined from the treatment plan information and CT image using algorithms such as pencil beam, tube-string convolution, and anisotropy.
[0033] Further preferably, the treatment plan information includes radiation parameters and treatment machine parameters. Radiation parameters include beam direction, beam energy, dose rate, multi-leaf grating motion trajectory, target dose, etc. Treatment machine parameters include treatment machine geometry parameters, beam collimator parameters, beam physics parameters (incident photon energy spectrum, primary photon flux normalization method, etc.), etc.
[0034] Preferably, the block embedding layer is used to divide the four-dimensional tensor data through a three-dimensional convolutional layer to determine multiple small blocks to be processed.
[0035] More preferably, the kernel of the three-dimensional convolutional layer is 2 or 4, and the stride is 2 or 4. More preferably, the block embedding layer is used to divide the four-dimensional vector into non-overlapping 2×2×2 blocks through the three-dimensional convolutional layer, and flatten each block and linearly map it to an embedding vector of dimension C (such as 48 or 96).
[0036] Preferably, the encoder includes multiple consecutive stages for sampling the multiple blocks to be processed separately to determine high-level features.
[0037] More preferably, the encoder comprises four consecutive stages, and each stage contains multiple sliding window transformer blocks. This encoder is used to perform block merging operations between every two stages, downsampling, halving the resolution of the feature map, doubling the number of channels, and achieving spatial compression and channel expansion.
[0038] Preferably, the feature modeling module includes continuous window multi-head attention sub-blocks and shifted window multi-head attention sub-blocks.
[0039] Preferably, this feature modeling module is used to extract local and global features based on the input features, and determine high-quality features.
[0040] Preferably, the bottleneck layer is used to integrate, compress, and perform global semantic modeling on the high-level features through two feature modeling modules to determine high-quality features.
[0041] More preferably, the bottleneck layer is set to two layers to capture the global dose distribution trend.
[0042] Preferably, the decoder is hopped with the encoder to upsample the bottleneck feature and determine the target feature.
[0043] More preferably, the decoder is used to upsample using a sliding window transform block and recover spatial resolution using transposed convolution.
[0044] Preferably, the output head is used to determine the heterogeneity correction coefficient based on the target feature.
[0045] Further preferably, the output head is used to compress the number of feature channels to 1 using a 1×1×1 convolutional layer, and ReLU is used as the activation function. This approach takes into account that dose prediction is a regression problem and that the dose value is non-negative, ensuring that the output is always greater than or equal to 0.
[0046] Based on the above, the heterogeneity correction model applied to radiation dose calculation 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 receives four-dimensional tensor data, including a radiation scan matrix, a total energy release matrix, an anatomical structure delineation matrix, and a treatment plan information matrix. The block embedding layer partitions the four-dimensional tensor data using a three-dimensional convolutional layer to determine multiple small blocks to be processed. The encoder includes multiple consecutive stages for sampling each of these small blocks to determine high-level features. The feature modeling modules include consecutive windowed multi-head attention sub-blocks and shifted windowed multi-head attention sub-blocks. These modules extract local and global features based on the input features to determine high-quality features. The decoder is skipped to the encoder to upsample these high-quality features and determine the target features. The output head determines the heterogeneity correction coefficients based on the target features.
[0047] It is evident that heterogeneity correction through a heterogeneity correction model improves computational accuracy and the precision of radiation dose calculation.
[0048] Preferably, the multi-head attention sub-block of the window includes a normalization layer, an attention module, a normalization layer, and a spatial multilayer perceptron module arranged in sequence, and the aforementioned modules are fused by residual connections.
[0049] Preferably, the shift window multi-head attention sub-block includes a normalization layer, a shift window multi-head attention module, a normalization layer, and a spatial multilayer perceptron module arranged in sequence, and the aforementioned modules are fused using residual connections.
[0050] More preferably, the spatial multilayer perceptron module is provided with an axial depth convolution structure, and the spatial multilayer perceptron module includes a first branch, a second branch, a third branch and a fourth branch.
[0051] In a further preferred embodiment, the spatial multilayer perceptron module uses axial depth convolution instead of the traditional fully connected layer MLP module, and adopts a four-branch parallel architecture, specifically designed to extract anisotropic features in three-dimensional medical images.
[0052] Further preferably, the first branch includes a vertical axis feature extractor. This vertical axis feature extractor is used to rearrange the input tensor, extract the height dimension data, and merge the width, depth, and channel dimension data. A global linear transformation or large-kernel convolution is applied to the height dimension data so that a direct connection can be established between any two locations in this dimension, simulating the scattering and propagation characteristics of rays along the longitudinal axis of the human body.
[0053] Further preferably, after the first branch is processed, the tensor is restored to its original five-dimensional shape.
[0054] Further preferably, the second branch includes a horizontal axis feature extractor. This horizontal axis feature extractor is used to rearrange the input tensor, extract the width dimension data, and merge the height dimension data, depth dimension data, and channel dimension data. A global linear transformation or large kernel convolution is applied to the width dimension data so that a direct connection can be established between any two positions in this dimension, simulating the scattering and propagation characteristics of rays along the horizontal axis of the human body.
[0055] Further preferably, the horizontal axis feature extractor is used to isolate the width dimension through dimension rearrangement and apply a global transformation to capture long-range dependencies in the horizontal direction. Further preferably, the above processing method takes into account the lateral scattering phenomenon in radiotherapy physics.
[0056] Further preferably, after the second branch is processed, the tensor is restored to its original five-dimensional shape.
[0057] Further preferably, the third branch includes a depth axis feature extractor. This depth axis feature extractor is used to rearrange the input tensor, extract the depth dimension data, and merge the height dimension data, width dimension data, and channel dimension data. A global linear transformation or large kernel convolution is applied to the depth dimension data so that a direct connection can be established between any two locations in this dimension, simulating the scattering and propagation characteristics of rays along the depth axis of the human body.
[0058] Further preferably, this depth axis feature extractor, through global connectivity, can calculate the dose distribution of deep tissues based on information (such as CT values) at all locations along the ray path, directly simulating the cumulative attenuation effect when rays penetrate tissues of different densities.
[0059] Further preferably, the fourth branch includes a residual connection structure. This residual connection structure is used to preserve high-frequency detail information of the input tensor.
[0060] Furthermore, the residual connection structure does not perform any transformation and directly transmits the original input. This residual connection structure is used to preserve high-frequency detail information in the input, preventing the loss of important local structural information during long-range feature extraction. This high-frequency detail information includes the sharp contours of bone edges, strong grayscale changes at tissue interfaces, and detailed textures of organ boundaries and lesion edges.
[0061] Preferably, the spatial multilayer perceptron module is used to stitch together the data output from the four branches along the channel dimension after the aforementioned four branches have been processed.
[0062] In a further optimized configuration, the number of channels is quadrupled during stitching. The multilayer perceptron module is used to reduce the dimensionality of the stitched features using a 3D 1×1×1 convolutional kernel, compressing the number of channels back to its original size.
[0063] Further preferably, the first branch, the second branch, the third branch, and the fourth branch are in a parallel architecture.
[0064] In a further preferred embodiment, the spatial multilayer perceptron module is used to concatenate the outputs of the first branch, the second branch, the third branch, and the fourth branch in the channel dimension, and to use a 1×1×1 three-dimensional convolution kernel to reduce the dimensionality of the concatenated features and compress the number of channels.
[0065] More preferably, the feature modeling module and / or the residual connections in the encoder are configured with the following composite loss function: in, The loss value. For voxel-level loss, For gradient loss; , All are weighting coefficients.
[0066] More preferably, the voxel-level loss is determined using a weighted mean square error method, through the following formula: in, For the predicted dose matrix, This is the actual dose matrix, where N is the total number of voxels; The gradient loss is determined by the following formula: in, This is an image edge detection operator.
[0067] More preferably, the treatment plan information includes beam geometry parameters, irradiation parameters, and beam physical parameters corresponding to the treatment machine.
[0068] It should be noted that all actions involving the acquisition of signals, information, or data in this application are carried out in compliance with the relevant data protection laws and policies of the country where the application is located, and with the authorization granted by the owner of the relevant device.
[0069] The above are one or more embodiments of the heterogeneity correction model for radiation dose calculation provided in this specification. Based on the same idea, this specification also provides a corresponding radiation dose calculation method. Figure 3 A schematic flowchart of a radiation dose calculation method provided for one embodiment of this specification is shown below. Figure 3 As shown, the method includes the following steps.
[0070] S900: Acquires patient CT images, anatomical structure delineation files, and treatment plan information.
[0071] Preferably, the method is performed by an electronic device that stores and is capable of running the heterogeneity correction model for radiation dose calculation provided in one or more of the foregoing embodiments. The electronic device may be a computer, server, etc., and this specification does not limit it.
[0072] S902: Determine the total energy release matrix based on the patient's CT images, the anatomical structure delineation file, and the treatment plan information.
[0073] Preferably, the relevant technologies for this step are already quite mature, and will not be elaborated upon in this specification.
[0074] S904: Input the patient's CT image, the anatomical structure delineation file, the treatment plan information, and the total energy release matrix into the heterogeneity correction model applied to radiation dose calculation, and determine the heterogeneity correction coefficient.
[0075] S908: Determine the final energy matrix based on the heterogeneity correction coefficient and the total energy release matrix.
[0076] The final energy matrix L can be determined using the following formula: Where R is the heterogeneity correction coefficient. For the total energy release matrix, This is the final energy matrix.
[0077] The above are one or more embodiments of the radiation dose calculation method provided in this specification. Based on the same idea, this specification also provides a corresponding radiation dose calculation device.
[0078] The radiation dose calculation device includes: The acquisition module is used to acquire patient CT images, anatomical structure delineation files, and treatment plan information.
[0079] The determination module is used to determine the total energy release matrix based on the patient's CT images, the anatomical structure delineation file, and the treatment plan information.
[0080] The generation module is used to input the patient's CT image, the anatomical structure delineation file, the treatment plan information, and the total energy release matrix into the heterogeneity correction model applied to radiation dose calculation, and determine the heterogeneity correction coefficient.
[0081] The output module is used to determine the final energy matrix based on the heterogeneity correction coefficient and the total energy release matrix.
[0082] This specification also provides a computer-readable storage medium storing a computer program that can be used to execute the above-described... Figure 2 The provided method for calculating radiation dose.
[0083] This specification also provides a computer-readable storage medium storing a computer program that can be used to execute the above-described... Figure 2 The provided method for calculating radiation dose.
[0084] This specification also provides a computer program product in which instructions, when executed by the processor of an electronic device, cause the electronic device to perform the above-described functions. Figure 3 The provided method for calculating radiation dose.
[0085] This instruction manual also provides Figure 4 The diagram shows the structure of the electronic device. Figure 4 At the hardware level, the electronic device includes a processor, internal bus, network interface, memory, and non-volatile memory, and may also include other hardware required for the business operations. The processor reads the corresponding computer program from the non-volatile memory into memory and then runs it to achieve the above-mentioned functions. Figure 3 The radiation dose calculation method described above. Of course, in addition to software implementation, this specification does not exclude other implementation methods, such as logic devices or a combination of hardware and software, etc. That is to say, the execution subject of the following processing flow is not limited to individual logic units, but can also be hardware or logic devices.
[0086] In the 1990s, improvements to a technology could be clearly distinguished as either hardware improvements (e.g., improvements to the circuit structure of diodes, transistors, switches, etc.) or software improvements (improvements to the methodology). However, with technological advancements, many methodological improvements today can be considered direct improvements to the hardware circuit structure. Designers almost always obtain the corresponding hardware circuit structure by programming the improved methodology into the hardware circuit. Therefore, it cannot be said that a methodological improvement cannot be implemented using hardware physical modules. For example, a Programmable Logic Device (PLD) (such as a Field Programmable Gate Array (FPGA)) is such an integrated circuit whose logic function is determined by the user programming the device. Designers can program and "integrate" a digital system onto a PLD themselves, without needing chip manufacturers to design and manufacture dedicated integrated circuit chips. Furthermore, nowadays, instead of manually manufacturing integrated circuit chips, this programming is mostly implemented using "logic compiler" software. Similar to the software compiler used in program development, the original code before compilation must also be written in a specific programming language, called a Hardware Description Language (HDL). There are many HDLs, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, and RHDL (Ruby Hardware Description Language). Currently, the most commonly used are VHDL (Very-High-Speed Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art should also understand that by simply performing some logic programming on the method flow using one of these hardware description languages and programming it into an integrated circuit, the hardware circuit implementing the logical method flow can be easily obtained.
[0087] The controller can be implemented in any suitable manner. For example, it can take the form of a microprocessor or processor and a computer-readable medium storing computer-readable program code (e.g., software or firmware) executable by the (micro)processor, logic gates, switches, application-specific integrated circuits (ASICs), programmable logic controllers, and embedded microcontrollers. Examples of controllers include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicon Labs C8051F320. A memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art will also recognize that, in addition to implementing the controller in purely computer-readable program code form, the same functionality can be achieved by logically programming the method steps to make the controller take the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers. Therefore, such a controller can be considered a hardware component, and the means included therein for implementing various functions can also be considered as structures within the hardware component. Alternatively, the means for implementing various functions can be considered as both software modules implementing the method and structures within the hardware component.
[0088] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, a computer can be, for example, a personal computer, laptop computer, cellular phone, camera phone, smartphone, personal digital assistant, media player, navigation device, email device, game console, tablet computer, wearable device, or any combination of these devices.
[0089] For ease of description, the above devices are described in terms of function, divided into various units. Of course, in implementing this specification, the functions of each unit can be implemented in one or more software and / or hardware components.
[0090] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0091] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0092] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0093] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0094] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.
[0095] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.
[0096] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.
[0097] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.
[0098] Those skilled in the art will understand that the embodiments of this specification can be provided as methods, systems, or computer program products. Therefore, this specification may take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this specification may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0099] This specification can be described in the general context of computer-executable instructions that are executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform a specific task or implement a specific abstract data type. This specification can also be practiced in distributed computing environments, where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.
[0100] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.
[0101] The above description is merely an embodiment of this specification and is not intended to limit this specification. Various modifications and variations can be made to this specification by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this specification should be included within the scope of the claims of this application.
Claims
1. A heterogeneity correction model applied to radiation dose calculation, characterized in that, It 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 used to receive four-dimensional tensor data; the four-dimensional tensor data includes a ray scanning matrix, a total energy release matrix, an anatomical structure delineation matrix, and a treatment plan information matrix; The block embedding layer is used to divide the four-dimensional tensor data through a three-dimensional convolutional layer to determine multiple small blocks to be processed; The encoder includes multiple consecutive stages for sampling the multiple small blocks to be processed to determine high-level features; The feature modeling module includes continuous window multi-head attention sub-blocks and shifted window multi-head attention sub-blocks; The feature modeling module is used to extract local and global features based on the input features and determine high-quality features; The decoder is connected to the encoder in a skip connection and is used to upsample the high-quality features to determine the target features; The output head is used to determine the heterogeneity correction coefficient based on the target features.
2. The heterogeneity correction model for radiation dose calculation according to claim 1, characterized in that, The multi-head attention sub-block of the window includes a normalization layer, an attention module, a normalization layer, and a spatial multilayer perceptron module arranged in sequence. The aforementioned modules are fused by residual connections. The shift window multi-head attention sub-block includes a normalization layer, a shift window multi-head attention module, a normalization layer, and a spatial multilayer perceptron module arranged in sequence. The aforementioned modules are fused using residual connections.
3. The heterogeneity correction model for radiation dose calculation according to claim 2, characterized in that, The spatial multilayer perceptron module is provided with an axial depth convolution structure, and the spatial multilayer perceptron module includes a first branch, a second branch, a third branch and a fourth branch; The first branch includes a vertical axis feature extractor; the vertical axis feature extractor is used to rearrange the input tensor, extract the height dimension data, and merge the width dimension data, depth dimension data, and channel dimension data; apply a global linear transformation or large kernel convolution to the height dimension data so that any two positions in this dimension can establish a direct connection, simulating the scattering and propagation characteristics of rays along the longitudinal axis of the human body; The second branch includes a horizontal axis feature extractor; the horizontal axis feature extractor is used to rearrange the input tensor, extract the width dimension data, and merge the height dimension data, depth dimension data, and channel dimension data; apply a global linear transformation or large kernel convolution to the width dimension data so that any two positions in this dimension can establish a direct connection, simulating the scattering and propagation characteristics of rays along the horizontal axis of the human body; The third branch includes a depth axis feature extractor; the depth axis feature extractor is used to rearrange the input tensor, extract the depth dimension data, and merge the height dimension data, width dimension data, and channel dimension data; apply a global linear transformation or large kernel convolution to the depth dimension data so that any two positions in this dimension can establish a direct connection, simulating the scattering and propagation characteristics of rays along the depth axis of the human body; The fourth branch includes a residual connection structure; the residual connection structure is used to preserve high-frequency detail information of the input tensor.
4. The heterogeneity correction model for radiation dose calculation according to claim 3, characterized in that, The first branch, the second branch, the third branch, and the fourth branch are in parallel architecture; The spatial multilayer perceptron module is used to concatenate the outputs of the first branch, the second branch, the third branch, and the fourth branch in the channel dimension, and use a 1×1×1 three-dimensional convolution kernel to reduce the dimensionality of the concatenated features and compress the number of channels.
5. The heterogeneity correction model for radiation dose calculation according to any one of claims 1-4, characterized in that, The feature modeling module and / or the residual connections in the encoder are configured with the following composite loss function: in, The loss value. For voxel-level loss, For gradient loss; , All are weighting coefficients.
6. The heterogeneity correction model for radiation dose calculation according to claim 5, characterized in that, The voxel-level loss is determined using a weighted mean square error method and the following formula: in, For the predicted dose matrix, This is the actual dose matrix, where N is the total number of voxels; The gradient loss is determined by the following formula: in, This is an image edge detection operator.
7. A method for calculating radiation dose, characterized in that, The method, using the heterogeneity correction model for radiation dose calculation according to any one of claims 1-6, comprises: Acquire patient CT images, anatomical structure delineation files, and treatment plan information; The total energy release matrix is determined based on the patient's CT images, the anatomical structure delineation file, and the treatment plan information. The patient's CT images, the anatomical structure delineation file, the treatment plan information, and the total energy release matrix are input into the heterogeneity correction model applied to radiation dose calculation to determine the heterogeneity correction coefficients. The final energy matrix is determined based on 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 described in claim 7.
9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method described in claim 7.
10. A computer program product, characterized in that, When the instructions in the computer program product are executed by the processor of the electronic device, the electronic device causes the electronic device to perform the method as described in claim 7.