Method, device, terminal and medium for generating dose distribution map in radiotherapy plan based on deep learning

Abstract

The present application provides a method, apparatus, terminal, and medium for generating a dose distribution map in a radiotherapy plan based on deep learning. In the present invention, a fitted beam segmentation mask is added on the basis of cascading, and predicting a dose distribution map is disassembled into predicting multiple dose distribution maps along the beam direction, and then the dose distribution maps in multiple directions are fused into a dose distribution map through a multi-beam voting mechanism. This method enables the network to effectively learn and predict the dose distribution in the beam direction. A novel loss function is further developed, and the dose evaluation indexes in the target area and the critical organ area in clinical practice are used as the loss function, which can make the dose in the target area and the critical organ area of the predicted dose distribution map more in line with the clinical treatment standard.

Description

Technical Field

[0001] The present application relates to the technical field of deep learning and dose prediction, and in particular to a method, device, terminal and medium for generating a dose distribution map in a radiotherapy plan based on deep learning. Background Art

[0002] At present, radiotherapy has become an important way to treat cancer. The accuracy of radiotherapy planning determines whether the target area can reach the prescribed dose and whether the critical organ exceeds the maximum tolerated dose. However, in order to formulate a radiotherapy plan that meets the requirements of clinical implementation, physicists or dosimetrists need to repeatedly adjust the treatment plan parameters based on their own experience. This process takes about a week, which prolongs the patient's waiting time for treatment. Therefore, the development of automatic and accurate dose distribution map prediction methods has become a task of great clinical significance. Widely used dose distribution map prediction algorithms such as DoseNet lack constraints on dose distribution prediction in the field direction and dose limit requirements for ROI, so the performance is still unsatisfactory.

[0003] Methods based on deep learning have been widely used in the task of predicting dose distribution maps for radiotherapy plans, and have shown excellent regression performance. At present, methods based on deep learning to predict dose distribution maps can be roughly divided into three categories: the first method is to design a variant network architecture based on U-Net to predict dose distribution maps. Kearney V et al. proposed the DoseNet network. The difference from 3D U-Net is that the residual block is added to help propagate the shallow feature information of the network to the deep part of the network. Liu et al. proposed to use a cascaded three-dimensional U-Net model and integrate local and global anatomical structure features to predict radiation dose distribution maps. The second method is to introduce a new loss function for special fields. Ngyuen et al. proposed to introduce a loss function based on dose volume histogram as a constraint to predict a radiation therapy dose distribution map that better meets clinical treatment needs. The third method is to explore the prior information input model. Zhang et al. proposed to add a distance map as the input of the network, thereby providing the network with prior information on the relationship between voxel dose and distance from the tumor.

[0004] However, these existing methods do not add constraints on the dose prediction task in the direction of the radiation beam. Therefore, the doses of the target area and critical organs in the predicted dose distribution diagram may not necessarily meet the requirements of clinical implementation. Summary of the invention

[0005] In view of the shortcomings of the prior art mentioned above, the purpose of the present application is to provide a method, device, terminal and medium for generating dose distribution maps in radiotherapy plans based on deep learning, so as to solve the problem that the existing dose prediction cannot meet the requirements of clinical implementation.

[0006] To achieve the above-mentioned purpose and other related purposes, the first aspect of the present application provides a method for generating a dose distribution map in a radiotherapy plan based on deep learning, comprising: obtaining a data set including a CT image, a planned target mask, and an organ at risk mask, and inputting the data set into a first neural network model based on an encoder and decoder structure for prediction to obtain a corresponding rough dose distribution map; extracting the planned target mask in the data set to obtain beam segmentation masks for multiple beam directions; inputting the CT image, the planned target mask, the organ at risk mask, the rough dose distribution map and multiple beam segmentation masks into a second neural network model based on an encoder and decoder structure to obtain multiple fine dose distribution maps corresponding to the multiple beam segmentation mask areas; and fusing the multiple fine dose distribution maps according to a voting mechanism to generate a fused dose distribution map for predicting the dose of the target area and the organ at risk area.

[0007] In some embodiments of the first aspect of the present application, the planned target area mask in the data set is extracted to obtain beam segmentation masks for multiple beam directions, and the process includes: extracting the boundary of the planned target area from the planned target area mask in the form of a two-dimensional slice to obtain a two-dimensional boundary slice; according to multiple preset angles, outlining two planned target area mask tangents on the two-dimensional slice at each preset angle, and the area between the two planned target area mask tangents is the beam segmentation mask of the preset angle; and stacking all two-dimensional beam segmentation masks to obtain a three-dimensional beam segmentation mask.

[0008] In some embodiments of the first aspect of the present application, the voting mechanism includes: the dose value of each voxel is determined by the dose value predicted by the beam segmentation mask area passing through the voxel; if a voxel is passed through a single beam segmentation mask, its dose value is only determined by the voxel dose value corresponding to this beam segmentation mask; if a voxel is passed through multiple beam segmentation masks, its dose value is determined by the average vote of the voxel dose values ​​corresponding to these multiple beam segmentation masks.

[0009] In some embodiments of the first aspect of the present application, an evaluation index used in clinical practice for the quality of radiotherapy plans is used as a loss function for supervised prediction of the fused dose distribution map and the true dose distribution map.

[0010] In some embodiments of the first aspect of the present application, in the process of inputting a data set including a CT image, a planned target area mask, and an organ at risk mask into the first neural network model for prediction, a regression loss function is selected for constraint to obtain the rough dose distribution map.

[0011] In some embodiments of the first aspect of the present application, the process of inputting the CT image, the planned target area mask, the organ at risk mask, the rough dose distribution map and multiple beam segmentation masks into the second neural network model for prediction is summarized, and the mean absolute error loss function of multiple beam segmentation mask areas is selected for constraint to obtain the fine dose distribution map.

[0012] To achieve the above-mentioned purpose and other related purposes, the second aspect of the present application provides a device for generating a dose distribution map in a radiotherapy plan based on deep learning, including: a rough dose distribution map generation module, which is used to obtain a data set including a CT image, a planned target area mask, and an organ at risk mask, and input it into a first neural network model based on an encoder and decoder structure for prediction to obtain a corresponding rough dose distribution map; a fine dose distribution map generation module, which is used to extract the planned target area mask in the data set to obtain a beam segmentation mask for multiple beam directions; the CT image, the planned target area mask, the organ at risk mask, the rough dose distribution map and the multiple beam segmentation masks are input into a second neural network model based on an encoder and decoder structure to obtain multiple fine dose distribution maps corresponding to the multiple beam segmentation mask areas; a fused dose distribution map generation module, which is used to fuse the multiple fine dose distribution maps according to a voting mechanism to generate a fused dose distribution map for predicting the dose of the target area and the organ at risk area.

[0013] In some embodiments of the second aspect of the present application, the fine dose distribution map generation module extracts the planned target area mask in the data set to obtain beam segmentation masks for multiple beam directions, and the process includes: extracting the boundary of the planned target area from the planned target area mask in the form of a two-dimensional slice to obtain a two-dimensional boundary slice; according to multiple preset angles, outlining two planned target area mask tangents on the two-dimensional slice at each preset angle, and the area between the two planned target area mask tangents is the beam segmentation mask of the preset angle; and stacking all two-dimensional beam segmentation masks to obtain a three-dimensional beam segmentation mask.

[0014] To achieve the above-mentioned purpose and other related purposes, the third aspect of the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for generating a dose distribution map in a radiotherapy plan based on deep learning.

[0015] To achieve the above-mentioned purpose and other related purposes, the fourth aspect of the present application provides an electronic terminal, including: a processor and a memory; the memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory, so that the terminal executes the method for generating a dose distribution map in a radiotherapy plan based on deep learning.

[0016] As described above, the method, device, terminal and medium for generating dose distribution maps in radiotherapy plans based on deep learning of the present application have the following beneficial effects: the present invention adds a fitted beam segmentation mask on the basis of cascade, and dismembers the predicted dose distribution map into multiple dose distribution maps along the beam direction, and then fuses the dose distribution maps in multiple directions into one dose distribution map through a multi-beam voting mechanism. This method enables the network to effectively learn to predict the dose distribution in the direction of the beam. A novel loss function is further developed, which uses the dose evaluation index of the target area and the critical organ area in the clinic as the loss function, so that the dose of the predicted dose distribution map in the target area and the critical organ area can be more in line with the clinical treatment standards. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 Shown is a flow chart of a method for generating a dose distribution map in a radiotherapy plan based on deep learning in one embodiment of the present application.

[0018] Figure 2 Shown is a schematic diagram of the structure of a dose distribution map generating terminal in a radiotherapy plan in an embodiment of the present application.

[0019] Figure 3 Shown is a structural schematic diagram of a device for generating a dose distribution map in a radiotherapy plan based on deep learning in one embodiment of the present application.

[0020] Figure 4 Shown is a structural schematic diagram of a device for generating a dose distribution map in a radiotherapy plan based on deep learning in one embodiment of the present application. DETAILED DESCRIPTION

[0021] The following describes the embodiments of the present application through specific examples, and those skilled in the art can easily understand other advantages and effects of the present application from the contents disclosed in this specification. The present application can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present application. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict.

[0022] As used herein, the singular forms "a", "an", and "the" are intended to include the plural forms as well, unless the context indicates otherwise. It should be further understood that the terms "comprise", "include" indicate the presence of the described features, operations, elements, components, items, kinds, and / or groups, but do not exclude the presence, occurrence, or addition of one or more other features, operations, elements, components, items, kinds, and / or groups. The terms "or" and "and / or" used herein are interpreted as inclusive, or mean any one or any combination. Therefore, "A, B or C" or "A, B and / or C" means "any of the following: A; B; C; A and B; A and C; B and C; A, B and C". Exceptions to this definition will only occur when the combination of elements, functions, or operations is inherently mutually exclusive in some way.

[0023] In order to solve the problems in the above-mentioned background technology, the present invention provides a method, device, terminal and medium for generating a dose distribution map in a radiotherapy plan based on deep learning, which aims to dismember the dose distribution map along the beam direction through the fitted beam mask to perform dose prediction, and constrain the dose limit of the target area and the critical organ area as a loss function. At the same time, in order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution in the embodiment of the present invention is further described in detail through the following examples and in combination with the accompanying drawings. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the invention.

[0024] Before further describing the present invention in detail, the nouns and terms involved in the embodiments of the present invention are explained. The nouns and terms involved in the embodiments of the present invention are applicable to the following interpretations:

[0025] <1> OAR (Organ At Risk): Organ at risk refers to normal organs that may be damaged by radiation, and their radiation sensitivity may have a direct impact on the treatment plan or prescribed dose.

[0026] <2> PTV (Planning Target Volume): Planning target volume refers to the irradiation range that is expanded during radiotherapy planning due to consideration of organ movement during irradiation and changes in target position and target volume during positioning.

[0027] The embodiments of the present invention provide a method for generating a dose distribution map in a radiotherapy plan based on deep learning, a system for the method for generating a dose distribution map in a radiotherapy plan based on deep learning, and a storage medium storing an executable program for implementing the method for generating a dose distribution map in a radiotherapy plan based on deep learning. With regard to the implementation of the method for generating a dose distribution map in a radiotherapy plan based on deep learning, the embodiments of the present invention will illustrate an exemplary implementation scenario of generating a dose distribution map in a radiotherapy plan based on deep learning.

[0028] like Figure 1 As shown, a flow chart of a method for generating a dose distribution map in a radiotherapy plan based on deep learning in an embodiment of the present invention is shown. The method for generating a dose distribution map in a radiotherapy plan based on deep learning in this embodiment mainly includes the following steps:

[0029] Step S11: Obtain a data set including a CT image, a planned target area mask, and an organ at risk mask, and input the data set into a first neural network model based on an encoder and decoder structure for prediction to obtain a corresponding rough dose distribution map.

[0030] It should be understood that the CT image is an electronic computed tomography image, including but not limited to X-ray CT and gamma-ray CT images; the planned target volume mask can be abbreviated as PTV mask, so the planned target volume mask and PTV mask appearing below have the same meaning; the organ at risk mask can be abbreviated as OAR mask, so the organ at risk mask and OAR mask appearing below have the same meaning.

[0031] In this embodiment, the CT image, planned target area mask, and risk organ mask in the data set are input into a first neural network model based on an encoder and decoder architecture, and are respectively encoded and decoded, and are constrained by a regression loss function to output a rough dose distribution map.

[0032] Furthermore, the first neural network model based on the encoder and decoder architecture includes but is not limited to convolutional neural network CNN or recurrent neural network RNN. In the convolutional neural network CNN, the image first passes through the convolution layer and then the linear layer, and finally outputs the classification result; the convolution layer is used for feature extraction, and the linear layer is used for result prediction; feature extraction can be regarded as an encoder, which encodes the original image into an intermediate expression that is conducive to its learning, and the decoder converts the intermediate expression into another expression. In the recurrent neural network RNN, RNN is also an encoder-decoder structure, the encoder encodes the text into a vector, and the decoder decodes the vector into the final output.

[0033] In this embodiment, the regression loss function includes but is not limited to the mean absolute error loss function (MAE), the mean square error loss function (MSE), the smoothed mean absolute error loss function (Huber loss function), the Log-Cosh loss function, the quantile loss function, etc.

[0034] This embodiment preferably uses the mean absolute error loss function (MAE). Since the absolute error is selected as the median, the median is more robust to outliers. The mean absolute error is the sum of the absolute values ​​of the differences between the target value and the predicted value, which represents the average error amplitude of the predicted value without considering the direction of the error.

[0035] Furthermore, the formula of the mean absolute error loss function (MAE) is as follows:

[0036]

[0037] Where n represents the number of voxels, represents the predicted dose value of the ith voxel, Y i represents the target dose value of the i-th voxel.

[0038] Step S12: extracting the planned target area mask in the data set to obtain beam segmentation masks for multiple beam directions.

[0039] In this embodiment, the planned target area mask in the data set is extracted to obtain beam segmentation masks for multiple beam directions, and the process includes: extracting the boundary of the planned target area from the planned target area mask to obtain a two-dimensional boundary slice in the form of a two-dimensional slice; according to multiple preset angles, outlining two planned target area mask tangents on the two-dimensional slice at each preset angle, and the area between the two planned target area mask tangents is the beam segmentation mask of the preset angle; all two-dimensional beam segmentation masks are stacked to obtain a three-dimensional beam segmentation mask.

[0040] Step S13: Input the CT image, the planned target area mask, the organ at risk mask, the rough dose distribution map and multiple beam segmentation masks into a second neural network model based on an encoder and decoder structure to obtain multiple fine dose distribution maps corresponding to the multiple beam segmentation mask areas.

[0041] Specifically, the CT image, PTV mask, OAR mask, rough dose distribution map and multiple beam segmentation masks are input into the second neural network model for encoding and decoding, and then multiple fine dose distribution maps corresponding to the beam segmentation mask area are predicted and output.

[0042] In this embodiment, the second neural network model includes but is not limited to a convolutional neural network CNN or a recurrent neural network RNN. In a convolutional neural network CNN, the image first passes through a convolutional layer and then a linear layer, and finally outputs a classification result; the convolutional layer is used for feature extraction, and the linear layer is used for result prediction; feature extraction can be regarded as an encoder that encodes the original image into an intermediate expression that is conducive to its learning, and the decoder converts the intermediate expression into another expression. In the recurrent neural network RNN, RNN is also an encoder-decoder structure, the encoder encodes the text into a vector, and the decoder decodes the vector into the final output.

[0043] Furthermore, the process of inputting the CT image, the planned target area mask, the organ at risk mask, the rough dose distribution map and multiple beam segmentation masks into the second neural network model for prediction is summarized, and the mean absolute error loss function of multiple beam segmentation mask areas is selected for constraint to obtain the fine dose distribution map.

[0044] Furthermore, the formula of the mean absolute error loss function of the multiple ray beam segmentation mask area is as follows:

[0045]

[0046] Where M represents the number of beam segmentation masks, Nm represents the number of voxels in the mth beam segmentation mask, represents the predicted dose value of the nth voxel in the mth beam segmentation mask, Y mn Represents the target dose value of the nth voxel within the mth beam segmentation mask.

[0047] Step S14: The multiple fine dose distribution maps are fused according to a voting mechanism to generate a fused dose distribution map for predicting the dose of the target area and the organ at risk area.

[0048] In this embodiment, the voting mechanism includes: the dose value of each voxel is determined by the dose value predicted by the beam segmentation mask area passing through the voxel; if a voxel is passed through a single beam segmentation mask, its dose value is only determined by the voxel dose value corresponding to this beam segmentation mask; if a voxel is passed through multiple beam segmentation masks, its dose value is determined by the average vote of the voxel dose values ​​corresponding to these multiple beam segmentation masks.

[0049] It should be understood that voxel is the abbreviation of volume pixel, including

[0050] The voxel can be expressed by stereo rendering or extracting polygonal isosurfaces with given threshold contours. Voxel is the smallest unit in three-dimensional space segmentation. In other words, voxel is a data structure that uses a fixed-size cubic block as the smallest unit to represent a three-dimensional object.

[0051] Preferably, this embodiment uses the evaluation index used in clinical practice for radiotherapy plan quality as a loss function to supervise the prediction of the fused dose distribution map and the true dose distribution map, thereby predicting a more accurate dose distribution map. The evaluation index used for radiotherapy plan quality includes but is not limited to the gradient index (GI), the homogeneity index (HI), the suitability index (CI), the minimum dose received by 1% of the PTV volume, and the Minimum dose received by 95% of the PTV volume Minimum dose received by 99% of the PTV volume Maximum dose received by the first 0.1cc of the OAR volume The average dose within the OAR region wait.

[0052] For example, the gradient index, also known as the conformal gradient index, measures the dose falloff outside the target volume and relies on the effective radius of the target volume being easily calculated by the treatment planning system. The homogeneity index is affected by many factors, including target volume, target location, and prescription dose; head and neck treatment plans, especially simultaneous integrated boost (SIB) plans, have the highest degree of heterogeneity, or have poor homogeneity index (HI) values ​​if calculated individually for different target volumes; the HI index is also affected by the proximity of organs at risk (OARs), the degree of overlap with the PTV, and the respective tolerance doses. Identifying cold spots and hot spots in the PTV that are underdose and overdose is a key step in plan evaluation.

[0053] Furthermore, the formula for using the evaluation index used in clinical radiotherapy planning quality as the loss function is as follows:

[0054]

[0055]

[0056] L DVH =α 1 L cDVH +α 2 L cDVH ; Formula 5)

[0057] Among them, L cDVH represents the value-based dose-volume histogram loss function, L cDVH represents the dose-volume histogram loss function based on the evaluation index, α 1 Indicates L cDVH The weight, α 2 Indicates L vDVH The weight of , H represents the maximum number of ROI masks (planned target mask and risk organ mask), Ns represents the number of voxels in the sth ROI mask, Represents the predicted dose distribution map, W s represents the sth region of interest mask, represents the predicted dose value of the nth voxel after the predicted dose values ​​in the sth region of interest mask are sorted from large to small, Y represents the target dose distribution map, R(Y*W s )n It represents the target dose value of the nth voxel after sorting the target dose values ​​in the mask of the sth region of interest from large to small; It represents the minimum dose received by 1% of the PTV volume. represents the minimum dose received by 95% of the PTV volume. represents the minimum dose received by 99% of the PTV volume, It represents the maximum dose received by the first 0.1cc of the OAR volume. represents the average dose within the OAR region.

[0058] The present invention adds a fitted beam segmentation mask on the basis of cascade, dismembers the predicted dose distribution map into multiple dose distribution maps along the beam direction, and then fuses the dose distribution maps in multiple directions into one dose distribution map through a multi-beam voting mechanism. This method enables the network to effectively learn to predict the dose distribution in the beam direction. A novel loss function is further developed, which uses the dose evaluation index of the target area and the organ at risk area in the clinic as the loss function, so that the dose of the predicted dose distribution map in the target area and the organ at risk area can be more in line with the clinical treatment standards. In comparison, the existing dose prediction method does not consider the beam direction, and simply uses a convolutional neural network to predict the dose distribution map, which will result in poor prediction accuracy of the predicted dose distribution map on the radiation path and the target area and the organ at risk area.

[0059] The method for generating a dose distribution map in a radiotherapy plan based on deep learning provided in an embodiment of the present invention can be implemented on the terminal side or the server side. As for the hardware structure of the terminal for generating a dose distribution map in a radiotherapy plan based on deep learning, please refer to Figure 2 , is an optional hardware structure diagram of a terminal 200 for generating a dose distribution map in a radiotherapy plan based on deep learning provided in an embodiment of the present invention. The terminal 200 may be a mobile phone, a computer device, a tablet device, a personal digital processing device, a factory background processing device, etc. The terminal 200 for generating a dose distribution map in a radiotherapy plan based on deep learning includes: at least one processor 201, a memory 202, at least one network interface 204 and a user interface 206. The various components in the device are coupled together through a bus system 205. It will be understood that the bus system 205 is used to realize connection and communication between these components. In addition to the data bus, the bus system 205 also includes a power bus, a control bus and a status signal bus. However, for the sake of clarity, in Figure 2 In the specification, various buses are labeled as bus systems.

[0060] The user interface 206 may include a display, a keyboard, a mouse, a trackball, a click gun, keys, buttons, a touch pad or a touch screen.

[0061] It is understood that the memory 202 can be a volatile memory or a non-volatile memory, and can also include both volatile and non-volatile memories. Among them, the non-volatile memory can be a read-only memory (ROM), a programmable read-only memory (PROM), which is used as an external cache. By way of example but not limitation, many forms of RAM are available, such as static random access memory (SRAM), synchronous static random access memory (SSRAM). The memory described in the embodiments of the present invention is intended to include but is not limited to these and any other suitable categories of memory.

[0062] The memory 202 in the embodiment of the present invention is used to store various categories of data to support the operation of the dose distribution map generation terminal 200 in the radiotherapy plan based on deep learning. Examples of these data include: any executable program used to operate on the dose distribution map generation terminal 200 in the radiotherapy plan based on deep learning, such as an operating system 2021 and an application 2022; the operating system 2021 includes various system programs, such as a framework layer, a core library layer, a driver layer, etc., for implementing various basic services and processing hardware-based tasks. The application 2022 may include various applications, such as a media player (MediaPlayer), a browser (Browser), etc., for implementing various application services. The method for generating a dose distribution map in a radiotherapy plan based on deep learning provided in the embodiment of the present invention may be included in the application 2022.

[0063] The method disclosed in the above embodiment of the present invention can be applied to the processor 201, or implemented by the processor 201. The processor 201 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the hardware integrated logic circuit in the processor 201 or the instruction in the form of software. The above processor 201 may be a general processor, a digital signal processor (DSP, Digital Signal Processor), or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. The processor 201 can implement or execute the various methods, steps and logic block diagrams disclosed in the embodiment of the present invention. The general processor 201 can be a microprocessor or any conventional processor, etc. In combination with the steps of the accessory optimization method provided in the embodiment of the present invention, it can be directly embodied as a hardware decoding processor to execute, or it can be executed by a combination of hardware and software modules in the decoding processor. The software module can be located in a storage medium, which is located in a memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware.

[0064] In an exemplary embodiment, the dose distribution map generation terminal 200 in the deep learning-based radiotherapy plan can be used by one or more application-specific integrated circuits (ASIC), DSP, programmable logic device (PLD), complex programmable logic device (CPLD) to execute the aforementioned method.

[0065] like Figure 3 As shown, a schematic diagram of the structure of a device for generating a dose distribution map in a radiotherapy plan based on deep learning in an embodiment of the present invention is shown. In this embodiment, the device 300 for generating a dose distribution map in a radiotherapy plan based on deep learning includes a rough dose distribution map generating module 301, a fine dose distribution map generating module 302, and a fused dose distribution map generating module 303.

[0066] The rough dose distribution map generation module 301 is used to obtain a data set including a CT image, a planned target area mask, and an organ at risk mask, and input it into a first neural network model based on an encoder and decoder structure for prediction to obtain a corresponding rough dose distribution map.

[0067] The fine dose distribution map generation module 302 is used to extract the planned target area mask in the data set to obtain beam segmentation masks for multiple beam directions; the CT image, the planned target area mask, the organ at risk mask, the rough dose distribution map and the multiple beam segmentation masks are input into a second neural network model based on an encoder and decoder structure to obtain multiple fine dose distribution maps corresponding to the multiple beam segmentation mask areas.

[0068] In some examples, the fine dose distribution map generation module 302 extracts the planned target area mask in the data set to obtain beam segmentation masks for multiple beam directions, and the process includes: extracting the boundary of the planned target area from the planned target area mask to obtain a two-dimensional boundary slice in the form of a two-dimensional slice; according to multiple preset angles, outlining two planned target area mask tangents on the two-dimensional slice at each preset angle, and the area between the two planned target area mask tangents is the beam segmentation mask of the preset angle; and stacking all two-dimensional beam segmentation masks to obtain a three-dimensional beam segmentation mask.

[0069] The fused dose distribution map generating module 303 is used to fuse the multiple fine dose distribution maps according to a voting mechanism to generate a fused dose distribution map for predicting the dose of the target area and the organ at risk area.

[0070] In some examples, the voting mechanism includes: the dose value of each voxel is determined by the dose value predicted by the beam segmentation mask area passing through the voxel; if a voxel is passed through a single beam segmentation mask, its dose value is determined only by the voxel dose value corresponding to this beam segmentation mask; if a voxel is passed through multiple beam segmentation masks, its dose value is determined by the average vote of the voxel dose values ​​corresponding to these multiple beam segmentation masks.

[0071] In some examples, the fused dose distribution map generation module 303 uses the evaluation index used in the clinic for the quality of radiotherapy plans as a loss function for the supervised prediction of the fused dose distribution map and the true dose distribution map.

[0072] In some examples, the rough dose distribution map generation module 301 selects a regression loss function for constraint to obtain the rough dose distribution map when inputting the data set including the CT image, the planned target area mask, and the risk organ mask into the first neural network model for prediction.

[0073] In some examples, the fine dose distribution map generation module 302 summarizes the process of inputting the CT image, the planned target area mask, the organ at risk mask, the rough dose distribution map and multiple beam segmentation masks into the second neural network model for prediction, and selects the mean absolute error loss function of multiple beam segmentation mask areas for constraint to obtain the fine dose distribution map.

[0074] It should be noted that: the device for generating a dose distribution map in a radiotherapy plan based on deep learning provided in the above-mentioned embodiment only uses the division of the above-mentioned program modules as an example when generating a dose distribution map in a radiotherapy plan based on deep learning. In actual applications, the above-mentioned processing can be assigned to different program modules as needed, that is, the internal structure of the device is divided into different program modules to complete all or part of the processing described above. In addition, the device for generating a dose distribution map in a radiotherapy plan based on deep learning provided in the above-mentioned embodiment and the method for generating a dose distribution map in a radiotherapy plan based on deep learning belong to the same concept. The specific implementation process is detailed in the method embodiment, which will not be repeated here.

[0075] For the convenience of understanding by those skilled in the art, Figure 4 The displayed model structure diagram further explains the execution process of the dose distribution map generating device in the radiotherapy plan based on deep learning provided in an embodiment of the present invention.

[0076] In the rough dose distribution map generation module: first obtain the CT image, the planned target area mask, and the mask of the organ at risk, and input these data into the first neural network model based on the encoder and decoder structure for prediction to obtain the corresponding rough dose distribution map.

[0077] In the fine dose distribution map generation module: the beam generator first extracts the planned target area mask in the data set to obtain beam segmentation masks for multiple beam directions; then the CT image, the planned target area mask, the mask of the organ at risk, the rough dose distribution map and the multiple beam segmentation masks are input into the second neural network model based on the encoder and decoder structure to obtain multiple fine dose distribution maps corresponding to the multiple beam segmentation mask areas. In addition, the fine dose distribution map generation module selects the mean absolute error loss function of the multiple beam segmentation mask areas for constraint to obtain the fine dose distribution map; the mean absolute error loss function is intended to minimize the distance between the true dose distribution map (Ground-Truth Beam Voters) and the predicted fine dose distribution map (predicted Beam Voters).

[0078] In the fused dose distribution map generation module: the beam voter fuses the multiple fine dose distribution maps according to the voting mechanism to generate a fused dose distribution map for predicting the dose of the target area and the organ at risk area. In addition, the fused dose distribution map generation module uses the evaluation index used in the clinic for the quality of radiotherapy plans as a loss function for the supervised prediction of the fused dose distribution map and the true dose distribution map; the loss function is intended to minimize the distance between the true dose distribution map (Ground Truth) and the predicted fused dose distribution map.

[0079] Those skilled in the art can understand that all or part of the steps of implementing the above-mentioned method embodiments can be completed by hardware related to the computer program. The aforementioned computer program can be stored in a computer-readable storage medium. When the program is executed, the steps of the above-mentioned method embodiments are executed; and the aforementioned storage medium includes: ROM, RAM, magnetic disk or optical disk and other media that can store program codes.

[0080] In the embodiments provided in the present application, the computer readable and writable storage medium may include a read-only memory, a random access memory, an EEPROM, a CD-ROM or other optical disk storage device, a disk storage device or other magnetic storage device, a flash memory, a USB flash drive, a mobile hard disk, or any other medium that can be used to store a desired program code in the form of an instruction or data structure and can be accessed by a computer. In addition, any connection can be appropriately referred to as a computer-readable medium. For example, if the instruction is sent from a website, a server or other remote source using a coaxial cable, an optical fiber cable, a twisted pair, a digital subscriber line (DSL), or wireless technologies such as infrared, radio, and microwaves, the coaxial cable, optical fiber cable, twisted pair, DSL, or wireless technologies such as infrared, radio, and microwaves are included in the definition of the medium. However, it should be understood that computer readable and writable storage media and data storage media do not include connections, carriers, signals, or other temporary media, but are intended to be non-temporary, tangible storage media. Disk and disc, as used in this application, includes compact disc (CD), laser disc, optical disc, digital versatile disc (DVD), floppy disk and Blu-ray disc where disks usually reproduce data magnetically, while discs reproduce data optically with lasers.

[0081] In summary, the present application provides a method, device, terminal and medium for generating a dose distribution map in a radiotherapy plan based on deep learning. The present invention adds a fitted beam segmentation mask on the basis of cascade, and dismembers the predicted dose distribution map into multiple dose distribution maps along the beam direction, and then fuses the dose distribution maps in multiple directions into one dose distribution map through a multi-beam voting mechanism. This method enables the network to effectively learn to predict the dose distribution in the direction of the beam. A novel loss function is further developed, which uses the dose evaluation index of the target area and the critical organ area in the clinic as the loss function, so that the dose of the predicted dose distribution map in the target area and the critical organ area can be more in line with the clinical treatment standards. Therefore, the present application effectively overcomes the various shortcomings in the prior art and has a high industrial utilization value.

[0082] The above embodiments are merely illustrative of the principles and effects of the present application and are not intended to limit the present application. Anyone familiar with the technology may modify or change the above embodiments without violating the spirit and scope of the present application. Therefore, all equivalent modifications or changes made by a person of ordinary skill in the art without departing from the spirit and technical ideas disclosed in the present application shall still be covered by the claims of the present application.

Claims

1. A method for generating dose distribution maps in radiotherapy plans based on deep learning. It is characterized in that include: A data set including a CT image, a planned target mask, and an organ at risk mask is obtained, and input into a first neural network model based on an encoder and decoder structure for prediction to obtain a corresponding rough dose distribution map; Extracting the planned target area mask in the data set to obtain beam segmentation masks for multiple beam directions, the process includes: extracting the boundary of the planned target area from the planned target area mask in the form of a two-dimensional slice to obtain a two-dimensional boundary slice; outlining two planned target area mask tangents on the two-dimensional slice according to each preset angle according to multiple preset angles, and the area between the two planned target area mask tangents is the beam segmentation mask of the preset angle; stacking all the two-dimensional beam segmentation masks to obtain a three-dimensional beam segmentation mask; Inputting the CT image, the planned target area mask, the organ at risk mask, the rough dose distribution map and the multiple beam segmentation masks into a second neural network model based on an encoder and decoder structure to obtain multiple fine dose distribution maps corresponding to the multiple beam segmentation mask areas; The multiple fine dose distribution maps are fused according to a voting mechanism to generate a fused dose distribution map for predicting the dose of the target area and the organ at risk area, and the voting mechanism includes: the dose value of each voxel is determined by the dose value predicted by the beam segmentation mask area passing through the voxel; if a voxel is passed through a single beam segmentation mask, its dose value is only determined by the voxel dose value corresponding to the beam segmentation mask; if a voxel is passed through multiple beam segmentation masks, its dose value is determined by an average vote of the voxel dose values ​​corresponding to the multiple beam segmentation masks.

2. The method for generating a dose distribution map in a radiotherapy plan based on deep learning according to claim 1, It is characterized in that It also includes using the evaluation indicators used in clinical practice for the quality of radiotherapy plans as loss functions for supervised prediction of the fused dose distribution map and the true dose distribution map.

3. According to the method for generating dose distribution map in radiotherapy plan based on deep learning in claim 1, It is characterized in that In the process of inputting the data set including the CT image, the planned target area mask and the risk organ mask into the first neural network model for prediction, a regression loss function is selected for constraint to obtain the rough dose distribution map.

4. The method for generating a dose distribution map in a radiotherapy plan based on deep learning according to claim 1, It is characterized in that The process of inputting the CT image, the planned target area mask, the organ at risk mask, the rough dose distribution map and multiple beam segmentation masks into the second neural network model for prediction is summarized, and the mean absolute error loss function of multiple beam segmentation mask areas is selected for constraint to obtain the fine dose distribution map.

5. A device for generating dose distribution maps in radiotherapy plans based on deep learning, It is characterized in that include: A rough dose distribution map generation module is used to obtain a data set including a CT image, a planned target area mask, and an organ at risk mask, and input the data set into a first neural network model based on an encoder and decoder structure for prediction to obtain a corresponding rough dose distribution map; A fine dose distribution map generation module is used to extract the planned target area mask in the data set to obtain beam segmentation masks for multiple beam directions, and the process includes: extracting the boundary of the planned target area from the planned target area mask in the form of a two-dimensional slice to obtain a two-dimensional boundary slice; according to multiple preset angles, outlining two planned target area mask tangents on the two-dimensional slice according to each preset angle, and the area between the two planned target area mask tangents is the beam segmentation mask of the preset angle; stacking all the two-dimensional beam segmentation masks to obtain a three-dimensional beam segmentation mask; inputting the CT image, the planned target area mask, the organ at risk mask, the rough dose distribution map and the multiple beam segmentation masks into a second neural network model based on an encoder and decoder structure to obtain multiple fine dose distribution maps corresponding to the multiple beam segmentation mask areas; A fused dose distribution map generation module is used to fuse the multiple fine dose distribution maps according to a voting mechanism to generate a fused dose distribution map for predicting the dose of the target area and the organ at risk area. The voting mechanism includes: the dose value of each voxel is determined by the dose value predicted by the beam segmentation mask area passing through the voxel; if a voxel is passed through a single beam segmentation mask, its dose value is only determined by the voxel dose value corresponding to the beam segmentation mask; if a voxel is passed through multiple beam segmentation masks, its dose value is determined by an average vote of the voxel dose values ​​corresponding to the multiple beam segmentation masks.

6. The device for generating dose distribution map in radiotherapy plan based on deep learning according to claim 5, It is characterized in that The fine dose distribution map generation module extracts the planned target area mask in the data set to obtain beam segmentation masks for multiple beam directions, and the process includes: extracting the boundary of the planned target area from the planned target area mask to obtain a two-dimensional boundary slice in the form of a two-dimensional slice; according to multiple preset angles, outlining two tangent lines of the planned target area mask on the two-dimensional slice at each preset angle, and the area between the two tangent lines of the planned target area mask is the beam segmentation mask of the preset angle; All two-dimensional ray beam segmentation masks are stacked to obtain a three-dimensional ray beam segmentation mask.

7. A computer-readable storage medium having a computer program stored thereon, It is characterized in that When the computer program is executed by a processor, it implements the method for generating a dose distribution map in a radiotherapy plan based on deep learning as described in any one of claims 1 to 4.

8. An electronic terminal, It is characterized in that include: Processor and memory; The memory is used to store computer programs; The processor is used to execute the computer program stored in the memory so that the terminal performs the method for generating a dose distribution map in a radiotherapy plan based on deep learning as described in any one of claims 1 to 4.

Images