CN114913261A - Three-dimensional in-vivo dose reconstruction method and device based on deep neural network - Google Patents
Three-dimensional in-vivo dose reconstruction method and device based on deep neural network Download PDFInfo
- Publication number
- CN114913261A CN114913261A CN202210572747.9A CN202210572747A CN114913261A CN 114913261 A CN114913261 A CN 114913261A CN 202210572747 A CN202210572747 A CN 202210572747A CN 114913261 A CN114913261 A CN 114913261A
- Authority
- CN
- China
- Prior art keywords
- dimensional
- dose
- map
- frame
- field
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T11/00—2D [Two Dimensional] image generation
- G06T11/003—Reconstruction from projections, e.g. tomography
- G06T11/006—Inverse problem, transformation from projection-space into object-space, e.g. transform methods, back-projection, algebraic methods
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61N—ELECTROTHERAPY; MAGNETOTHERAPY; RADIATION THERAPY; ULTRASOUND THERAPY
- A61N5/00—Radiation therapy
- A61N5/10—X-ray therapy; Gamma-ray therapy; Particle-irradiation therapy
- A61N5/1048—Monitoring, verifying, controlling systems and methods
- A61N5/1071—Monitoring, verifying, controlling systems and methods for verifying the dose delivered by the treatment plan
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T15/00—3D [Three Dimensional] image rendering
- G06T15/06—Ray-tracing
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Biomedical Technology (AREA)
- General Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Evolutionary Computation (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Biophysics (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Radiology & Medical Imaging (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Pathology (AREA)
- Animal Behavior & Ethology (AREA)
- Public Health (AREA)
- Veterinary Medicine (AREA)
- Algebra (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Pure & Applied Mathematics (AREA)
- Computer Graphics (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a three-dimensional in-vivo dose reconstruction method and a three-dimensional in-vivo dose reconstruction device based on a deep neural network, wherein the method comprises the following steps: acquiring a single-frame or single-field portal gray-scale image of a patient based on an Electronic Portal Imaging Device (EPID), and performing matrix correction on the portal gray-scale image to obtain a two-dimensional single-frame or single-field transmission dose map; converting the two-dimensional single-frame or single-field transmission dose map into a three-dimensional single-frame or single-field initial dose map based on a field angle and a back projection algorithm; and acquiring a three-dimensional structure diagram of the patient, inputting the three-dimensional single-frame or single-field initial dose diagram and the three-dimensional structure diagram into a deep neural network trained by a model for prediction, and acquiring a three-dimensional single-frame or single-field dose distribution diagram. The invention is based on three modules of matrix correction, back projection conversion and network prediction, and can quickly and accurately reconstruct the three-dimensional in-vivo dosage of a patient according to an actually measured portal gray scale image.
Description
Technical Field
The invention relates to the technical field of auxiliary medical treatment, in particular to a three-dimensional in-vivo dose reconstruction method and device based on a deep neural network.
Background
With the gradual progress of medical technology, more and more difficult miscellaneous diseases or cancer diseases can be treated, and the service life of people is prolonged. One of the treatments for cancer is Radiation Therapy (RT), which is aimed at providing a spatially conformal radiation dose to a target region of a tumor.
One of the radiotherapy application devices is a magnetic resonance accelerator, and the treatment plan can be re-optimized according to the patient treatment by fractions and even the anatomical change in the treatment process, so that online Adaptive Radiotherapy (ART) becomes possible. In order to ensure the treatment quality of the patient, whether the actual irradiated dose of the patient is equal to the planned dose or not needs to be verified in the online treatment process so as to find out the serious error of the actual planned execution. The currently common dose verification method is to record the actual transmission dose of a patient based on an Electronic Portal Imaging Device (EPID), and can realize forward dose verification by comparing the actually measured two-dimensional transmission dose of the EPID with a planned transmission dose, but the forward method cannot quantitatively analyze the three-dimensional dose in the body of the patient; inverse dose verification can reconstruct a patient three-dimensional in-vivo dose based on the EPID transmitted dose and compare the planned three-dimensional dose. For EPID in a magnetic resonance accelerator, efficient three-dimensional in-vivo dose reconstruction has been realized based on a dose reconstruction technique of transmission dose back projection and scattering correction, but the technique has the disadvantages that the Electron Return Effect (ERE) under a magnetic field cannot be considered, and the reconstructed dose has large errors at a heterogeneous tissue interface; another dose reconstruction technique, which may take into account ERE but the reconstruction speed is slow and fails to find errors in planning execution in time, may iteratively back-derive the actual incident flux based on the EPID transmitted dose, and then calculate the dose a second time by the MC engine.
From the above background, an accurate and efficient dose reconstruction method for dose verification is urgently needed in the planning execution process of the online adaptive radiotherapy.
Disclosure of Invention
The invention provides a three-dimensional in-vivo dose reconstruction method and a three-dimensional in-vivo dose reconstruction device based on a deep neural network.
The first aspect of the embodiments of the present invention provides a three-dimensional in-vivo dose reconstruction method based on a deep neural network, where the method includes:
acquiring a single-frame or single-field portal gray-scale image of a patient based on the EPID, and performing matrix correction on the portal gray-scale image to obtain a two-dimensional single-frame or single-field transmission dose map;
converting the two-dimensional single-frame or single-field transmission dose map into a three-dimensional single-frame or single-field initial dose map based on a field angle and a back projection algorithm;
acquiring a three-dimensional structure diagram of a patient, inputting the three-dimensional single-frame or single-field initial dose diagram and the three-dimensional structure diagram into a deep neural network trained by a model for prediction to obtain a three-dimensional single-frame or single-field dose distribution diagram, wherein the three-dimensional structure diagram comprises: electron density map or mass density map.
In a possible implementation manner of the first aspect, the back-projecting the two-dimensional single-frame or single-field transmission dose map into a three-dimensional single-frame or single-field initial dose map based on a field angle and a back-projection algorithm includes:
according to the geometric effect of inverse square ratio, back-projecting the two-dimensional single-frame or single-field transmission dose map into a three-dimensional basic dose map by using a back-projection algorithm;
obtaining a back projection matrix, and calculating by using the back projection matrix and the three-dimensional basic dose map to obtain a three-dimensional single-frame or single-field initial dose map;
the conversion calculation of the three-dimensional single-frame or single-field initial dose map is shown as the following formula:
wherein d is v Is a three-dimensional single-frame or single-field initial dose map,
for two-dimensional single-frame or single-field transmission dose maps, A for inverse projection matrix rotation, a ij For back-projecting the elements of the matrix rotation, r i Is the distance, r, from the source to the dose voxel i j，EPID Is the distance from the source to pixel j of the portal gray scale image.
In a possible implementation manner of the first aspect, the obtaining a back projection matrix includes:
and carrying out ray tracing on the radiation rays in the field gray image in a preset BEV coordinate to obtain a back projection matrix.
In one possible implementation manner of the first aspect, the matrix correction includes: EPID sensitivity matrix correction, EPID dose response correction, and EPID internal scatter correction.
In one possible implementation manner of the first aspect, the deep neural network is a two-dimensional neural network;
inputting the three-dimensional single-frame or single-field initial dose map and the three-dimensional structure map into a deep neural network trained by a model for prediction to obtain a three-dimensional single-frame or single-field dose distribution map, wherein the three-dimensional single-frame or single-field initial dose map comprises the following steps:
repeatedly inputting the three-dimensional single-frame or single-field initial dose map and the three-dimensional structure map slices into the two-dimensional neural network to obtain a plurality of cross-section two-dimensional dose distribution maps;
and splicing the plurality of cross section two-dimensional dose distribution maps layer by layer to obtain a three-dimensional single-frame or single-field dose distribution map.
In one possible implementation manner of the first aspect, the deep neural network is a three-dimensional neural network;
inputting the three-dimensional single-frame or single-field initial dose map and the three-dimensional structure map into a deep neural network trained by a model for prediction to obtain a three-dimensional single-frame or single-field dose distribution map, wherein the three-dimensional single-frame or single-field initial dose map comprises the following steps:
repeatedly inputting the three-dimensional single-frame or single-field initial dose map and the three-dimensional structure map into the three-dimensional neural network to obtain a plurality of output dose blocks;
and splicing and combining a plurality of output dose blocks to obtain a three-dimensional single-frame or single-field dose distribution map.
In a possible implementation manner of the first aspect, the model training specifically includes:
acquiring radiotherapy planning data of a plurality of different organ parts and a three-dimensional structure diagram of a patient, the radiotherapy planning data comprising: CT images and planning data, the three-dimensional structure map comprising: converting the CT image into a three-dimensional electron density map or a three-dimensional mass density map;
based on the three-dimensional structure chart and the field angle and single-frame or single-field flux extracted from the planning data, respectively calculating an MC three-dimensional dose map of each field and a two-dimensional transmission dose map corresponding to the position of the EPID by adopting a preset Monte Carlo dose engine;
and performing model training on the neural network by using the three-dimensional structure diagram, the MC three-dimensional dose map and the three-dimensional initial dose map obtained by back projection of the two-dimensional transmission dose map.
A second aspect of an embodiment of the present invention provides a deep neural network-based three-dimensional in-vivo dose reconstruction apparatus, including:
the acquisition and correction module is used for acquiring a single-frame or single-field portal gray-scale image of a patient based on the EPID and carrying out matrix correction on the portal gray-scale image to obtain a two-dimensional single-frame or single-field transmission dose map;
the back projection module is used for converting the two-dimensional single-frame or single-field transmission dose map into a three-dimensional single-frame or single-field initial dose map based on a field angle and a back projection algorithm;
the network prediction module is used for acquiring a three-dimensional structure diagram of a patient, inputting the three-dimensional single-frame or single-field initial dose map and the three-dimensional structure diagram into a deep neural network trained by a model for prediction, and obtaining a three-dimensional single-frame or single-field dose distribution diagram, wherein the three-dimensional structure diagram comprises: electron density map or mass density map.
Compared with the prior art, the three-dimensional in-vivo dose reconstruction method and device based on the deep neural network have the advantages that: the invention converts the actually measured EPID (extended peripheral identification) portal gray-scale map into a three-dimensional initial dose map through a matrix correction and back projection algorithm, and then quickly predicts the three-dimensional in-vivo dose of a patient through a deep neural network by combining an actual three-dimensional structure map of the body of the patient and the three-dimensional initial dose map so as to realize quick and accurate three-dimensional in-vivo dose reconstruction.
Drawings
Fig. 1 is a schematic flowchart of a three-dimensional in-vivo dose reconstruction method based on a deep neural network according to an embodiment of the present invention;
FIG. 2 is a flowchart illustrating the operation of model training according to an embodiment of the present invention;
FIG. 3 is a flowchart illustrating an operation of a deep neural network-based three-dimensional in-vivo dose reconstruction method according to an embodiment of the present invention;
fig. 4 is a schematic structural diagram of a three-dimensional in-vivo dose reconstruction device based on a deep neural network according to an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The currently used dose reconstitution methods have the following technical problems: although the existing dose reconstruction method combining EPID transmission dose back projection with scattering correction is high-efficiency, the magnetic field dose effect cannot be considered, so that the reconstructed dose is not accurate; the existing method for reconstructing the incident flux based on the EPID transmission dose back-estimation and then using the MC dose engine for secondary calculation has the advantages of accurate dose reconstruction and long reconstruction time.
In order to solve the above problems, a three-dimensional in-vivo dose reconstruction method based on a deep neural network provided by the embodiments of the present application will be described and explained in detail by the following specific embodiments.
Referring to fig. 1, a schematic flow chart of a three-dimensional in-vivo dose reconstruction method based on a deep neural network according to an embodiment of the present invention is shown.
By way of example, the deep neural network-based three-dimensional in-vivo dose reconstruction method may include:
s11, acquiring a single-frame or single-field portal gray-scale image of the patient based on the EPID, and performing matrix correction on the portal gray-scale image to obtain a two-dimensional single-frame or single-field transmission dose map.
In one embodiment, the portal grayscale image can be a portal grayscale image acquired by an Electronic Portal Imaging Device (EPID) about the patient. And then, converting the field gray level image into an image matrix, and then carrying out a series of matrix correction on the image matrix to obtain a two-dimensional transmission dose map.
In an optional embodiment, the matrix correction includes: EPID sensitivity matrix correction, EPID dose response correction, and EPID internal scatter correction.
And S12, converting the two-dimensional single-frame or single-field transmission dose map into a three-dimensional single-frame or single-field initial dose map based on the field angle and the back projection algorithm.
The two-dimensional single-frame or single-field transmission dose map is a two-dimensional single-frame transmission dose map or a two-dimensional single-field transmission dose map.
In one embodiment, for convenience of processing, a two-dimensional single-frame or single-field transmission dose map can be made into a two-dimensional flux map, and then the two-dimensional flux map can be back-projected into a three-dimensional single-frame or single-field initial dose map only from a geometric perspective using a back-projection algorithm. Specifically, the three-dimensional single-frame or single-field initial dose map is a three-dimensional single-frame initial dose map or a three-dimensional single-field initial dose map.
In the embodiment, physical effects such as ray attenuation and scattering can be not considered, and only geometric effects of inverse square ratio can be considered, so that the dose actually irradiated to the patient can be determined more stereoscopically.
In order to accurately convert the two-dimensional transmitted dose map into a three-dimensional initial dose map, in an alternative embodiment, step S12 may include the following sub-steps:
and a substep S121 of back-projecting the two-dimensional single-frame or single-field transmission dose map into a three-dimensional basic dose map by using a back-projection algorithm according to a geometric effect of inverse square ratio.
And a substep S122 of obtaining a back projection matrix, and calculating to obtain a three-dimensional single-frame or single-field initial dose map by using the back projection matrix and the three-dimensional basic dose map.
Wherein, the conversion calculation of the three-dimensional single-frame or single-field initial dose map is shown as the following formula:
wherein d is v Is a three-dimensional single-frame or single-field initial dose map,
for two-dimensional single-frame or single-field transmission dose maps, A for inverse projection matrix rotation, a ij For back-projecting the elements of the matrix rotation, r i Is the distance, r, from the source to the dose voxel i j，EPID Is the distance from the source to pixel j of the portal gray scale image.
In an optional embodiment, the obtaining a back projection matrix includes:
and carrying out ray tracing on the radiation rays in the field gray image in a preset BEV coordinate to obtain a back projection matrix.
It should be noted that, since the back projection matrix is a geometric matrix of the reconstruction system and is independent of the patient, in practical applications, the back projection matrix may be acquired in advance and then stored, and calculation for each patient is not required.
S13, obtaining a three-dimensional structure diagram of the patient, inputting the three-dimensional single-frame or single-field initial dose diagram and the three-dimensional structure diagram into a deep neural network trained by a model for prediction, and obtaining a three-dimensional single-frame or single-field dose distribution diagram, wherein the three-dimensional structure diagram comprises: electron density map or mass density map.
The electron density map or the mass density map may be obtained by performing HU-ED/MD correction curve conversion on a CT image of a patient taken by clinical CT.
In an optional embodiment, a user can perform a large amount of network training on the deep neural network in advance, and then input the three-dimensional initial dose map and the three-dimensional structure map into the deep neural network in a double-channel mode, so that the actually required dose is predicted.
Alternatively, the deep neural network may be a two-dimensional deep neural network or a three-dimensional deep neural network, the deep neural network includes two paths for encoding and decoding, and the deep neural network model may be constructed based on modules such as CNN, Transformer, MLP and the like in actual use.
In one embodiment, when the deep neural network is a two-dimensional neural network, the step S13 may include the following sub-steps:
and the substep S131 of repeatedly inputting the three-dimensional single-frame or single-field initial dose map and the three-dimensional structure map slices into the two-dimensional neural network to obtain a plurality of cross section two-dimensional dose distribution maps.
And a substep S132, splicing the plurality of cross section two-dimensional dose distribution maps layer by layer to obtain a three-dimensional single-frame or single-field dose distribution map.
In one embodiment, when the deep neural network is a three-dimensional neural network, the step S13 may include the following sub-steps:
and a substep S133 of repeatedly inputting the three-dimensional single-frame or single-field initial dose map and the three-dimensional structure map into the three-dimensional neural network to obtain a plurality of output dose blocks.
And a substep S134, splicing and combining a plurality of output dose blocks to obtain a three-dimensional single-frame or single-field dose distribution map.
Dose reconstruction is carried out through the deep neural network, the dose reconstruction speed can be improved while the magnetic field dose accuracy is ensured, and the requirements of an online adaptive radiotherapy process on accuracy and efficiency are met.
In one embodiment, after obtaining the patient three-dimensional in-vivo reconstructed dose map, the patient three-dimensional in-vivo reconstructed dose map may be compared to a three-dimensional planned dose map, and if a large dose error is found (e.g., 3 mm/3% gamma pass < 90%), the treatment is terminated, and subsequent fractions of the treatment plan are re-planned based on the measured reconstructed dose.
Referring to fig. 2, a flowchart of the operation of model training provided by an embodiment of the present invention is shown.
In one optional embodiment, the model training specifically includes:
s21, acquiring radiotherapy planning data of a plurality of different organ parts and a three-dimensional structure chart of the patient, wherein the radiotherapy planning data comprises: CT images and planning data, the three-dimensional structure map comprising: and (3) converting the CT image into a three-dimensional electron density map or a three-dimensional mass density map.
And S22, respectively calculating the MC three-dimensional dose map of each field and the two-dimensional transmission dose map of the position of the corresponding EPID by adopting a preset Monte Carlo dose engine based on the three-dimensional structure chart and the field angle and the single-frame or single-field flux extracted from the planning data.
And S23, performing model training on the neural network by using the three-dimensional structure diagram, the MC three-dimensional dose diagram and the three-dimensional initial dose diagram obtained by the back projection of the two-dimensional transmission dose diagram.
In an alternative embodiment, the three-dimensional structure diagram, the MC three-dimensional dose diagram, and the two-dimensional transmission dose diagram may be divided into a training set and a test set, respectively, and then model training is performed using the training set, and then verification is performed using the test set.
In yet another alternative embodiment, to further expand the training data set, techniques for enhancing the original illumination field data may be employed, including rotating the angle of the original illumination field, modifying the isocenter coordinates of the original illumination field, and the like. And then recalculating the enhanced MC three-dimensional dose map and the corresponding two-dimensional transmission dose map by using a Monte Carlo dose engine based on the modified original irradiation field isocenter coordinate, and ensuring that the statistical error of the Monte Carlo dose distribution is less than 1%.
Referring to fig. 3, an operation flowchart of a three-dimensional volume dose reconstruction method based on a deep neural network according to an embodiment of the present invention is shown.
Specifically, a portal gray scale map of a patient can be acquired, then the portal gray scale map is converted into a two-dimensional transmission dose map based on the portal gray scale map, then the two-dimensional transmission dose map is projected through a back projection algorithm to obtain a three-dimensional initial dose map (wherein the three-dimensional initial dose map contains a coarse predicted dose), and then the three-dimensional initial dose map is input into a deep neural network for prediction calculation to obtain a three-dimensional dose distribution map for dose verification performed on a plan.
Compared with the prior art, the existing dose reconstruction method cannot consider the nonuniform tissue interface dose redistribution caused by the magnetic field ERE, and the method is limited to abdominal cases with uniform tissue distribution; the invention can consider the above effect and effectively improve the accuracy of the reconstructed dose.
Referring to the table below, the test results for four sites are shown, wherein the gamma passage rate for the rectal site is also higher than for the existing correction algorithm,
TABLE 1.1. the algorithm 20 patients tested for mean γ passage (mean. + -. standard deviation)
TABLE 1.2 patient mean gamma throughput (mean ± variance) for existing backprojection reconstruction algorithms
Compared with the existing iterative back-push-based MC secondary dose calculation method, the method has higher speed, and the single-frame or single-field dose reconstruction time of the 4 cases is within 1s, so that the requirements of online adaptive radiotherapy are met.
In this embodiment, an embodiment of the present invention provides a method for predicting a three-dimensional in-vivo dose based on a deep neural network, which has the following beneficial effects: the invention converts the actually measured EPID (extended peripheral identification) portal gray-scale map into a three-dimensional initial dose map through a matrix correction and back projection algorithm, and then quickly predicts the three-dimensional in-vivo dose of a patient through a deep neural network by combining an actual three-dimensional structure map and the three-dimensional initial dose map of the patient so as to realize quick and accurate three-dimensional in-vivo dose reconstruction.
The embodiment of the invention also provides a three-dimensional in-vivo dose reconstruction device based on the deep neural network, and referring to fig. 4, a schematic structural diagram of the three-dimensional in-vivo dose reconstruction device based on the deep neural network provided by the embodiment of the invention is shown.
Wherein, as an example, the deep neural network-based three-dimensional in-vivo dose reconstruction apparatus may include:
the acquisition and correction module 401 is used for acquiring a single-frame or single-field portal gray scale map of a patient based on the EPID and performing matrix correction on the portal gray scale map to obtain a two-dimensional single-frame or single-field transmission dose map;
a back projection module 402, configured to convert the two-dimensional single-frame or single-field transmission dose map into a three-dimensional single-frame or single-field initial dose map based on a field angle and a back projection algorithm;
a network prediction module 403, configured to obtain a three-dimensional structure diagram of the patient, and input the three-dimensional single-frame or single-field initial dose map and the three-dimensional structure diagram into a deep neural network trained by a model for prediction to obtain a three-dimensional single-frame or single-field dose distribution diagram, where the three-dimensional structure diagram includes: electron density map or mass density map.
Optionally, the back projection module is further configured to:
according to the geometric effect of inverse square proportion, utilizing a back projection algorithm to back-project the two-dimensional single-frame or single-field transmission dose map into a three-dimensional basic dose map;
obtaining a back projection matrix, and calculating by using the back projection matrix and the three-dimensional basic dose map to obtain a three-dimensional single-frame or single-field initial dose map;
the conversion calculation of the three-dimensional single-frame or single-field initial dose map is shown as the following formula:
wherein, d v Is a three-dimensional single-frame or single-field initial dose map,
is a two-dimensional single-frame or single-field transmission dose map, A is a back projection matrix transformation, a ij For back-projecting the elements of the matrix rotation, r i Is the distance, r, from the source to the dose voxel i j，EPID Is the distance from the source to pixel j of the portal gray scale image.
Optionally, the obtaining a back projection matrix comprises:
and carrying out ray tracing on the radiation rays in the field gray image in a preset BEV coordinate to obtain a back projection matrix.
Optionally, the matrix correction comprises: EPID sensitivity matrix correction, EPID dose response correction, and EPID internal scatter correction.
Optionally, the deep neural network is a two-dimensional neural network;
the network prediction module is further configured to:
repeatedly inputting the three-dimensional single-frame or single-field initial dose map and the three-dimensional structure map slices into the two-dimensional neural network to obtain a plurality of cross-section two-dimensional dose distribution maps;
and splicing the plurality of cross section two-dimensional dose distribution maps layer by layer to obtain a three-dimensional single-frame or single-field dose distribution map.
Optionally, the deep neural network is a three-dimensional neural network;
the network prediction module is further configured to:
repeatedly inputting the three-dimensional single-frame or single-field initial dose map and the three-dimensional structure map into the three-dimensional neural network to obtain a plurality of output dose blocks;
and splicing and combining a plurality of output dose blocks to obtain a three-dimensional single-frame or single-field dose distribution map.
Optionally, the model training specifically includes:
acquiring radiotherapy planning data of a plurality of different organ parts and a three-dimensional structure diagram of a patient, the radiotherapy planning data comprising: CT images and planning data, the three-dimensional structure map comprising: converting the CT image into a three-dimensional electron density map or a three-dimensional mass density map;
based on the three-dimensional structure chart and the field angle and single-frame or single-field flux extracted from the planning data, respectively calculating an MC three-dimensional dose map of each field and a two-dimensional transmission dose map corresponding to the position of the EPID by adopting a preset Monte Carlo dose engine;
and performing model training on the neural network by using the three-dimensional structure diagram, the MC three-dimensional dose map and the three-dimensional initial dose map obtained by back projection of the two-dimensional transmission dose map.
It can be clearly understood by those skilled in the art that, for convenience and brevity, the specific working process of the apparatus described above may refer to the corresponding process in the foregoing method embodiment, and is not described herein again.
Further, an embodiment of the present application further provides an electronic device, including: a memory, a processor and a computer program stored on the memory and executable on the processor, the processor when executing the program implementing the method for three-dimensional volume-in-vivo dose reconstruction based on a deep neural network as described in the above embodiments.
Further, the present application also provides a computer-readable storage medium storing computer-executable instructions for causing a computer to execute the method for reconstructing a three-dimensional volume dose based on a deep neural network according to the above embodiment.
While the foregoing is directed to the preferred embodiment of the present invention, it will be understood by those skilled in the art that various changes and modifications may be made without departing from the spirit and scope of the invention.
Claims (10)
1. A three-dimensional in-vivo dose reconstruction method based on a deep neural network is characterized by comprising the following steps:
acquiring a single-frame or single-field portal gray-scale map of a patient based on the EPID, and performing matrix correction on the portal gray-scale map to obtain a two-dimensional single-frame or single-field transmission dose map;
converting the two-dimensional single-frame or single-field transmission dose map into a three-dimensional single-frame or single-field initial dose map based on a field angle and a back projection algorithm;
acquiring a three-dimensional structure diagram of a patient, inputting the three-dimensional single-frame or single-field initial dose diagram and the three-dimensional structure diagram into a deep neural network trained by a model for prediction to obtain a three-dimensional single-frame or single-field dose distribution diagram, wherein the three-dimensional structure diagram comprises: electron density map or mass density map.
2. The method for reconstructing three-dimensional in-vivo dose based on the deep neural network as claimed in claim 1, wherein the back-projecting the two-dimensional single-frame or single-field transmission dose map into a three-dimensional single-frame or single-field initial dose map based on the field angle and back-projection algorithm comprises:
according to the geometric effect of inverse square proportion, utilizing a back projection algorithm to back-project the two-dimensional single-frame or single-field transmission dose map into a three-dimensional basic dose map;
obtaining a back projection matrix, and calculating by using the back projection matrix and the three-dimensional basic dose map to obtain a three-dimensional single-frame or single-field initial dose map;
the conversion calculation of the three-dimensional single-frame or single-field initial dose map is shown as the following formula:
wherein d is v Is a three-dimensional single-frame or single-field initial dose map,
for two-dimensional single-frame or single-field transmission dose maps, A for inverse projection matrix rotation, a ij For back-projecting the elements of the matrix rotation, r i Is the distance, r, from the source to the dose voxel i j，EPID Is the distance from the source to pixel j of the portal gray scale image.
3. The deep neural network-based three-dimensional in-vivo dose reconstruction method according to claim 2, wherein the obtaining a back projection matrix comprises:
and carrying out ray tracing on the radiation rays in the field gray image in a preset BEV coordinate to obtain a back projection matrix.
4. The deep neural network-based three-dimensional in-vivo dose reconstruction method according to claim 1, wherein the matrix correction comprises: EPID sensitivity matrix correction, EPID dose response correction, and EPID internal scatter correction.
5. The deep neural network-based three-dimensional in-vivo dose reconstruction method according to any one of claims 1 to 4, wherein the deep neural network is a two-dimensional neural network;
inputting the three-dimensional single-frame or single-field initial dose map and the three-dimensional structure map into a deep neural network trained by a model for prediction to obtain a three-dimensional single-frame or single-field dose distribution map, wherein the three-dimensional single-frame or single-field initial dose map comprises the following steps:
repeatedly inputting the three-dimensional single-frame or single-field initial dose map and the three-dimensional structure map slices into the two-dimensional neural network to obtain a plurality of cross-section two-dimensional dose distribution maps;
and splicing the plurality of cross section two-dimensional dose distribution maps layer by layer to obtain a three-dimensional single-frame or single-field dose distribution map.
6. The deep neural network-based three-dimensional in-vivo dose reconstruction method according to any one of claims 1 to 4, wherein the deep neural network is a three-dimensional neural network;
inputting the three-dimensional single-frame or single-field initial dose map and the three-dimensional structure map into a deep neural network trained by a model for prediction to obtain a three-dimensional single-frame or single-field dose distribution map, wherein the three-dimensional single-frame or single-field initial dose map comprises the following steps:
repeatedly inputting the three-dimensional single-frame or single-field initial dose map and the three-dimensional structure map into the three-dimensional neural network to obtain a plurality of output dose blocks;
and splicing and combining a plurality of output dose blocks to obtain a three-dimensional single-frame or single-field dose distribution map.
7. The deep neural network-based three-dimensional in-vivo dose reconstruction method according to claim 1, wherein the model training specifically comprises:
acquiring radiotherapy planning data of a plurality of different organ parts and a three-dimensional structure diagram of a patient, the radiotherapy planning data comprising: CT images and planning data, the three-dimensional structure map comprising: converting the CT image into a three-dimensional electron density map or a three-dimensional mass density map;
based on the three-dimensional structure chart and the field angle and single-frame or single-field flux extracted from the planning data, respectively calculating an MC three-dimensional dose map of each field and a two-dimensional transmission dose map corresponding to the position of the EPID by adopting a preset Monte Carlo dose engine;
and performing model training on the neural network by using the three-dimensional structure diagram, the MC three-dimensional dose map and the three-dimensional initial dose map obtained by back projection of the two-dimensional transmission dose map.
8. A deep neural network-based three-dimensional in-vivo dose reconstruction apparatus, the apparatus comprising:
the acquisition and correction module is used for acquiring a single-frame or single-field portal gray-scale image of a patient based on the EPID and carrying out matrix correction on the portal gray-scale image to obtain a two-dimensional single-frame or single-field transmission dose map;
the back projection module is used for converting the two-dimensional single-frame or single-field transmission dose map into a three-dimensional single-frame or single-field initial dose map based on a field angle and a back projection algorithm;
the network prediction module is used for acquiring a three-dimensional structure diagram of a patient, inputting the three-dimensional single-frame or single-field initial dose map and the three-dimensional structure diagram into a deep neural network trained by a model for prediction, and obtaining a three-dimensional single-frame or single-field dose distribution diagram, wherein the three-dimensional structure diagram comprises: electron density map or mass density map.
9. An electronic device, comprising: memory, processor and computer program stored on the memory and executable on the processor, characterized in that the processor when executing the program implements the method for deep neural network based three-dimensional volumetric dose reconstruction as claimed in any one of claims 1 to 7.
10. A computer-readable storage medium storing computer-executable instructions for causing a computer to perform the method for deep neural network-based three-dimensional volumetric dose reconstruction as claimed in any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210572747.9A CN114913261A (en) | 2022-05-25 | 2022-05-25 | Three-dimensional in-vivo dose reconstruction method and device based on deep neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210572747.9A CN114913261A (en) | 2022-05-25 | 2022-05-25 | Three-dimensional in-vivo dose reconstruction method and device based on deep neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114913261A true CN114913261A (en) | 2022-08-16 |
Family
ID=82768830
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210572747.9A Pending CN114913261A (en) | 2022-05-25 | 2022-05-25 | Three-dimensional in-vivo dose reconstruction method and device based on deep neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114913261A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118471436A (en) * | 2024-07-15 | 2024-08-09 | 福建自贸试验区厦门片区Manteia数据科技有限公司 | Illuminated dose determination device, electronic apparatus, and computer-readable storage medium |
-
2022
- 2022-05-25 CN CN202210572747.9A patent/CN114913261A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118471436A (en) * | 2024-07-15 | 2024-08-09 | 福建自贸试验区厦门片区Manteia数据科技有限公司 | Illuminated dose determination device, electronic apparatus, and computer-readable storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US10765888B2 (en) | System and method for automatic treatment planning | |
| CN109069858B (en) | Radiotherapy system and computer readable storage device | |
| US10188874B2 (en) | Three dimensional localization and tracking for adaptive radiation therapy | |
| US10791958B2 (en) | Magnetic resonance projection imaging | |
| US10668304B2 (en) | Phantom for adaptive radiotherapy | |
| US10152790B2 (en) | Three dimensional localization of a moving target for adaptive radiation therapy | |
| JP2007514499A (en) | System and method for global optimization of treatment plans for external beam irradiation therapy | |
| AU2018266458B2 (en) | Systems and methods of accounting for shape change during radiotherapy | |
| WO2020087257A1 (en) | Image guidance method and device, and medical equipment and computer readable storage medium | |
| Sumida et al. | A convolution neural network for higher resolution dose prediction in prostate volumetric modulated arc therapy | |
| Jiang et al. | 3D in vivo dose verification in prostate proton therapy with deep learning-based proton-acoustic imaging | |
| CN114913261A (en) | Three-dimensional in-vivo dose reconstruction method and device based on deep neural network | |
| Lemus et al. | Dosimetric assessment of patient dose calculation on a deep learning‐based synthesized computed tomography image for adaptive radiotherapy | |
| CN113941100A (en) | Method and apparatus for generating deliverable radiotherapy plan according to three-dimensional spatial dose distribution | |
| US20160213948A1 (en) | System and method to create test objects for radiation therapy | |
| Lang et al. | Hybrid-supervised deep learning for domain transfer 3D protoacoustic Image reconstruction | |
| Xie et al. | Inpainting truncated areas of CT images based on generative adversarial networks with gated convolution for radiotherapy | |
| CN114344735A (en) | Radiation dose monitoring and verifying method based on scattered ray characteristics | |
| US11887301B2 (en) | System and method for automatic delineation of scanned images | |
| Peng et al. | Recent advancements of artificial intelligence in particle therapy | |
| CN116848549A (en) | Detection of image structures via dimension-reduction projection | |
| Jeon et al. | Preliminary application of synthetic computed tomography image generation from magnetic resonance image using deep-learning in breast cancer patients | |
| US20240173571A1 (en) | Systems and methods for dose verification | |
| Folkerts | Topics in Cancer Radiotherapy: Automated Treatment Planning and Quality Assurance | |
| CN116091517A (en) | Medical image processing method, medical image processing device, storage medium and computer program product |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |