CN116825283B - Nuclear medicine treatment individuation dosage evaluation method and device based on transfer learning - Google Patents
Nuclear medicine treatment individuation dosage evaluation method and device based on transfer learning Download PDFInfo
- Publication number
- CN116825283B CN116825283B CN202310470415.4A CN202310470415A CN116825283B CN 116825283 B CN116825283 B CN 116825283B CN 202310470415 A CN202310470415 A CN 202310470415A CN 116825283 B CN116825283 B CN 116825283B
- Authority
- CN
- China
- Prior art keywords
- network
- training
- dose
- nuclear medicine
- parameters
- 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.)
- Active
Links
- 238000009206 nuclear medicine Methods 0.000 title claims abstract description 54
- 238000013526 transfer learning Methods 0.000 title claims abstract description 50
- 238000011282 treatment Methods 0.000 title claims abstract description 40
- 238000011156 evaluation Methods 0.000 title claims abstract description 27
- 238000012549 training Methods 0.000 claims abstract description 80
- 238000000034 method Methods 0.000 claims abstract description 62
- 238000009826 distribution Methods 0.000 claims abstract description 34
- 238000013210 evaluation model Methods 0.000 claims abstract description 17
- 239000011159 matrix material Substances 0.000 claims description 60
- 230000008569 process Effects 0.000 claims description 35
- 238000012360 testing method Methods 0.000 claims description 22
- 230000006870 function Effects 0.000 claims description 20
- 230000000694 effects Effects 0.000 claims description 19
- 239000000463 material Substances 0.000 claims description 19
- 238000002560 therapeutic procedure Methods 0.000 claims description 12
- 238000005070 sampling Methods 0.000 claims description 9
- 238000013508 migration Methods 0.000 claims description 8
- 230000005012 migration Effects 0.000 claims description 8
- 238000011176 pooling Methods 0.000 claims description 8
- 238000012795 verification Methods 0.000 claims description 8
- 238000004590 computer program Methods 0.000 claims description 7
- 230000000153 supplemental effect Effects 0.000 claims description 4
- 230000004913 activation Effects 0.000 claims description 3
- 238000010030 laminating Methods 0.000 claims 1
- 230000007704 transition Effects 0.000 claims 1
- 238000013528 artificial neural network Methods 0.000 abstract description 29
- 238000005516 engineering process Methods 0.000 abstract description 9
- 230000008021 deposition Effects 0.000 abstract description 8
- 238000004891 communication Methods 0.000 description 8
- 238000012545 processing Methods 0.000 description 7
- 238000004364 calculation method Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 5
- 206010028980 Neoplasm Diseases 0.000 description 3
- 230000008901 benefit Effects 0.000 description 3
- 238000013527 convolutional neural network Methods 0.000 description 3
- 239000003814 drug Substances 0.000 description 3
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 2
- 231100000987 absorbed dose Toxicity 0.000 description 2
- 238000003491 array Methods 0.000 description 2
- 230000000295 complement effect Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 229940079593 drug Drugs 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 210000000056 organ Anatomy 0.000 description 2
- 238000002600 positron emission tomography Methods 0.000 description 2
- 238000011084 recovery Methods 0.000 description 2
- 238000004088 simulation Methods 0.000 description 2
- 238000002603 single-photon emission computed tomography Methods 0.000 description 2
- 238000000638 solvent extraction Methods 0.000 description 2
- 238000000342 Monte Carlo simulation Methods 0.000 description 1
- 238000012952 Resampling Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 201000011510 cancer Diseases 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000009977 dual effect Effects 0.000 description 1
- 238000000802 evaporation-induced self-assembly Methods 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 239000013307 optical fiber Substances 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 230000002093 peripheral effect Effects 0.000 description 1
- 239000000047 product Substances 0.000 description 1
- 230000002285 radioactive effect Effects 0.000 description 1
- 230000001950 radioprotection Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 239000013589 supplement Substances 0.000 description 1
- 238000003325 tomography Methods 0.000 description 1
Landscapes
- Radiation-Therapy Devices (AREA)
Abstract
The application relates to a nuclear medicine treatment individuation dosage evaluation method and device based on transfer learning, wherein the method comprises the following steps: selecting case samples with the difference of the physical parameters of the existing cases meeting preset conditions; generating training sets of patients of different physique based on the case samples; training a 3D-Unet network by using a training set, and after constructing a nuclear medicine treatment individuation dose evaluation model based on the network DeepDose _T, performing supplementary training on unfixed parameters of the network during transfer learning to obtain specific network parameters suitable for learning distribution samples so as to generate the nuclear medicine treatment individuation dose evaluation model after the transfer learning. Therefore, the problems that in the related technology, the neural network application has high dependence on a training set and limited generalization capability, meanwhile, the difference of the body types of patients can cause different distances of source target voxels, the energy deposition distribution of nuclides is further influenced, the prediction result of a network obtained by case training of common physical parameters is inaccurate, and the like are solved.
Description
Technical Field
The application relates to the technical field of modern medicine, in particular to a nuclear medicine treatment individuation dose evaluation method and device based on transfer learning.
Background
With the development of quantitative measurement technology, the spatial distribution description of the activity of the radioactive drug in the nuclear medicine treatment is finer and finer, and the scale of the calculation of the dose in the nuclear medicine is finer and finer after the gradual development from the organ scale to the sub-organ scale, to the voxel scale and the cell scale. In studies in the field of radioprotection, dose levels of the illuminated personnel are often assessed at organ-scale doses and at effective doses. In nuclear medicine treatment, tumor and normal tissue drugs have uneven activity distribution, resulting in uneven absorbed dose distribution, and "cold spots" formed by uneven dose may allow the tumor to regrow from surviving cancer cells, so organ-averaged absorbed dose is not the most useful data for predicting clinical treatment outcome. The source item positioning at the voxel level can be realized by using molecular functional imaging equipment such as SPECT (single photon emission computed tomography), PET (positron emission tomography) and the like, so that the calculation of the voxel-level dose distribution is possible.
In the related art, source item distribution and CT HU value distribution are used as input, meng Ka is used for simulation to obtain voxel level 3D dose distribution as an accurate value, a convolutional neural network is trained to realize dose prediction, and the result shows that the application of the neural network to realize 3D dose calculation can keep the advantage of accurate simulation calculation of Meng Ka, and meanwhile, the defect of long calculation time is overcome.
However, in the related art, the neural network application has high dependence on the training set, the generalization capability is limited, the neural network application is difficult to be accurately applied to the prediction of the test sample, meanwhile, the difference of the body types of the patient can cause different distances of source target voxels, the energy deposition distribution of nuclides is further influenced, the prediction result of the network obtained by case training of common physical parameters is inaccurate, and the problem needs to be solved.
Disclosure of Invention
The application provides a nuclear medicine treatment individuation dosage evaluation method and device based on transfer learning, which are used for solving the problems that in the related technology, the neural network application has high dependence on a training set, the generalization capability is limited, the neural network application is difficult to accurately apply to the prediction of a test sample, meanwhile, the difference of body types of patients can cause different distances of source and target voxels, the energy deposition distribution of nuclides is further influenced, the prediction result of a network obtained by case training of common physical parameters is inaccurate, and the like.
An embodiment of the first aspect of the present application provides a method for evaluating a personalized dose of a nuclear medicine therapy based on transfer learning, including the steps of: selecting case samples with the difference of the physical parameters of the existing cases meeting preset conditions; generating a training set of different physical patients based on the case samples; and training the 3D-Unet network by using the training set, and after constructing a nuclear medicine treatment individuation dose evaluation model based on the network DeepDose _T, performing supplementary training on unfixed parameters of the network during transfer learning to obtain specific network parameters suitable for learning distribution samples so as to generate the nuclear medicine treatment individuation dose evaluation model after the transfer learning.
Optionally, in an embodiment of the present application, the unfixed parameter is a deep network parameter of the encoding process.
Optionally, in an embodiment of the present application, the generating a training set of different physical patients based on the case samples includes: and (3) reducing the activity matrix and the material matrix of the sample while maintaining the resolution ratio of the voxels unchanged, and sampling the difference value after reducing the preset multiple to obtain the activity matrix and the material matrix of the small-size sample reaching the target small size.
Optionally, in an embodiment of the present application, when performing the transfer learning, the performing supplemental training on the unfixed parameters of the network to obtain specific network parameters suitable for learning the distribution sample includes: and respectively selecting a plurality of layers of the convolution layer parameters in the coding process for training, obtaining an optimal network structure based on the verification set, and testing by using the test set to obtain the specific network parameters.
Optionally, in an embodiment of the present application, the training the 3D-Unet network using the training set includes: taking the L1-norm between the real dose rate matrix and the network predicted dose rate matrix as a loss function, wherein the loss function is as follows:
Wherein y i and The dose truth value and the dose predicted value of the ith voxel are respectively represented, and n is the number of voxels.
An embodiment of the second aspect of the present application provides a nuclear medicine therapy personalized dose evaluation device based on transfer learning, including: the selection module is used for selecting case samples with the difference of the physical parameters of the existing cases meeting the preset conditions; the generation module is used for generating training sets of patients with different physique based on the case samples; and the evaluation module is used for training the 3D-Unet network by using the training set, and after constructing the nuclear medicine treatment individuation dose evaluation model based on the network DeepDose _T, performing supplementary training on unfixed parameters of the network during transfer learning to obtain specific network parameters suitable for learning distribution samples so as to generate the nuclear medicine treatment individuation dose evaluation model after the transfer learning.
Optionally, in an embodiment of the present application, the unfixed parameter is a deep network parameter of the encoding process.
Optionally, in one embodiment of the present application, the generating module includes: and the generating unit is used for reducing the activity matrix and the material matrix of the sample while keeping the resolution of the voxels unchanged, and sampling the difference after reducing the preset multiple to obtain the activity matrix and the material matrix of the small-size sample reaching the target small size.
Optionally, in one embodiment of the present application, the evaluation module includes: the training unit is used for respectively selecting a plurality of layers of the convolutional layer parameters in the coding process to train, obtaining an optimal network structure based on the verification set, and testing by utilizing the test set to obtain the specific network parameters.
Optionally, in one embodiment of the present application, the evaluation module includes: an evaluation unit, configured to take an L1-norm between a real dose rate matrix and a network predicted dose rate matrix as a loss function, where the loss function is:
Wherein y i and The i-th represents the dose truth value and the dose predicted value of the voxels, respectively, and n is the number of voxels.
An embodiment of a third aspect of the present application provides an electronic device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor, the processor executing the program to implement the method for personalized dose assessment of nuclear medicine treatment based on transfer learning as described in the above embodiments.
A fourth aspect of the present application provides a computer readable storage medium storing a computer program which when executed by a processor implements a method for personalized dose assessment of a nuclear medicine treatment based on transfer learning as above.
According to the embodiment of the application, the 3D-Unet network can be trained by using the training set, and the unfixed parameters of the network are subjected to complementary training when the migration learning is performed, so that the specific network parameters suitable for learning the distribution sample are obtained, the influence of the physical difference of a patient on the prediction result of the neural network is eliminated, and the accuracy of the target voxel dose prediction far from the source is improved. Therefore, the problems that in the related technology, the neural network application has high dependence on a training set, the generalization capability is limited, the neural network application is difficult to accurately apply to the prediction of a test sample, meanwhile, the difference of the body types of patients can cause different distances of source target voxels, the energy deposition distribution of nuclides is further influenced, the prediction result of the network obtained by case training of common physical parameters is inaccurate, and the like are solved.
Additional aspects and advantages of the application will be set forth in part in the description which follows and, in part, will be obvious from the description, or may be learned by practice of the application.
Drawings
The foregoing and/or additional aspects and advantages of the application will become apparent and readily appreciated from the following description of the embodiments, taken in conjunction with the accompanying drawings, in which:
FIG. 1 is a flow chart of a method for personalized dose assessment of nuclear medicine treatment based on transfer learning according to an embodiment of the present application;
FIG. 2 is a schematic diagram of a DeepDose _T transfer learning network architecture for a transfer learning-based nuclear medicine therapy personalized dose assessment method according to one embodiment of the application;
FIG. 3 is a schematic structural diagram of a nuclear medicine therapy personalized dose assessment device based on transfer learning according to an embodiment of the present application;
fig. 4 is a schematic structural diagram of an electronic device according to an embodiment of the present application.
Detailed Description
Embodiments of the present application are described in detail below, examples of which are illustrated in the accompanying drawings, wherein like or similar reference numerals refer to like or similar elements or elements having like or similar functions throughout. The embodiments described below by referring to the drawings are illustrative and intended to explain the present application and should not be construed as limiting the application.
The following describes a method and a device for evaluating the personalized dose of a nuclear medicine treatment based on transfer learning according to an embodiment of the present application with reference to the accompanying drawings. Aiming at the problems that in the related technology mentioned in the background technology, the neural network application has high dependence on a training set, the generalization capability is limited, the neural network application is difficult to accurately apply to the prediction of test samples, meanwhile, the difference of patient sizes can cause different distances of source target voxels, the energy deposition distribution of nuclides is further influenced, the prediction result is inaccurate by using a network obtained by case training of common physical parameters, the application provides a nuclear medicine treatment individuation dose evaluation method based on transfer learning, in the method, a 3D-Unet network can be trained by using the training set, and in the transfer learning, the unfixed parameters of the network are subjected to complementary training to obtain specific network parameters suitable for learning distribution samples, so that the influence of the physical differences of patients on the prediction result of the neural network is eliminated, and the accuracy of target voxel dose prediction far from a source is improved. Therefore, the problems that in the related technology, the neural network application has high dependence on a training set, the generalization capability is limited, the neural network application is difficult to accurately apply to the prediction of a test sample, meanwhile, the difference of the body types of patients can cause different distances of source target voxels, the energy deposition distribution of nuclides is further influenced, the prediction result of the network obtained by case training of common physical parameters is inaccurate, and the like are solved.
Specifically, fig. 1 is a schematic flow chart of a method for evaluating personalized doses of nuclear medicine treatment based on transfer learning according to an embodiment of the present application.
As shown in fig. 1, the method for personalized dose evaluation of nuclear medicine treatment based on transfer learning comprises the following steps:
In step S101, a case sample that satisfies a preset condition with respect to the existing case physique parameter difference is selected.
It can be understood that the case samples meeting the preset conditions in the embodiment of the present application may be case samples having a large difference from the existing case physique parameters.
In the actual execution process, the embodiment of the application can select the case samples with larger difference with the existing case physique parameters, thereby providing basis for generating training sets of different physique patients according to the existing case samples, further eliminating the influence of the physique difference of the patients on the prediction result of the neural network and improving the accuracy of target voxel dosage prediction far from the source.
In step S102, a training set of different physical patients is generated based on the case samples.
As a possible implementation manner, the embodiment of the present application may perform three-dimensional equal-scale reduction on the existing normal-size sample to generate a small-size sample, where the neural network input in the embodiment of the present application only has a three-dimensional data matrix, and voxel size information is not introduced. When the whole body model is reduced by directly reducing the size of the voxels, the source term matrix and the body model matrix are unchanged before and after the size is changed, namely the data input into the network are consistent, and the predicted value output by the network is the same as the original value. However, due to the change of the voxel size, the dose distribution truth value calculated by the Monte Carlo method is changed, so that the obtained predicted value and the truth value are inevitably different, namely, the situation that samples with the same input and different outputs are introduced into a training set occurs, and therefore, the training set of patients with different physique is generated based on case samples, and the influence of the physique difference of the patients on the neural network predicted result can be eliminated.
Optionally, in one embodiment of the present application, generating training sets of different physical patients based on the case samples includes: and (3) reducing the activity matrix and the material matrix of the sample while maintaining the resolution ratio of the voxels unchanged, and sampling the difference value after reducing the preset multiple to obtain the activity matrix and the material matrix of the small-size sample reaching the target small size.
It is understood that the shrinking preset times in the embodiments of the present application may be, but are not limited to, a times.
For example, the embodiment of the application can reduce the activity matrix (a) and the material matrix (M) of the original sample while maintaining the resolution of the voxels unchanged, and perform interpolation resampling after reducing a times to obtain the activity matrix (a S) and the material matrix (M S) of the small-size sample reaching the target small size, wherein the voxel size and the voxel number of the small-size sample matrix are consistent with those of the original matrix, but the volume occupied by the patient is reduced by a 3 times.
The method and the device can reduce the activity matrix and the material matrix of the sample, and further obtain the activity matrix and the material matrix of the small-size sample with the small size reaching the target, thereby further eliminating the influence of the physical difference of the patient on the prediction result of the neural network.
In step S103, the 3D-Unet network is trained by using the training set, and after the personalized dose evaluation model of the nuclear medicine treatment based on the network DeepDose _t is constructed, additional training is performed on unfixed parameters of the network during the transfer learning, so as to obtain specific network parameters suitable for learning distribution samples, so as to generate the personalized dose evaluation model of the nuclear medicine treatment after the transfer learning.
It will be appreciated that the 3D-Unet network in the embodiments of the present application was proposed and applied in 2016 by the university of Freiburg, germany, to the medical image-based segmentation problem. 2D-Unet is commonly used in the field of image processing to divide images, and because the processing object is a single picture and the processing object of 3D-Unet is a 3D matrix, not only can the characteristic information of an XY plane be extracted, but also the information of the Z direction can be considered, and the method is suitable for the characteristics that axial information needs to be considered in biomedical tomography. In addition, the medical image used in the embodiment of the application has the characteristics of small gray scale difference, fuzzy boundary, low resolution and the like, voxel-level information is needed to be used for accurate mapping, meanwhile, the tissue organ structure of the human body is relatively uniform, the regularity is good, and the information on the characteristic layer can be beneficial to structural recovery. The Unet network combines the low-level features and the high-level features by using a cross-link mode, and combines the dual requirements of structure recovery and fine edge information reconstruction, so that the method becomes an excellent choice for processing the problems in the field of medical images.
Furthermore, the Unet network structure is characterized in that compared with a shallow convolution layer in the encoding process, the perception domain of the network is smaller, so that the extracted features contain more position-related information, and the alignment of the voxel level feature map and the original map is facilitated. For convolutional layers in the decoding process, the extracted features are abstract features that contain more context information, since their perceptual domain is larger and more advantageous to provide low resolution information. The difference of source and target voxel positions caused by different body types in the embodiment of the application belongs to the global relation represented by the deep features, so that the convolutional layer parameters of the coding process are fixed during migration learning, and the deep network parameters are opened for further learning.
In the actual implementation process, the embodiment of the application can be based on a 3D-Unet network commonly used in the biomedical image field, build a network DeepDose _T suitable for the targets in the embodiment of the application, train the 3D-Unet network by using a training set with enough sample size, fix a part of network parameters, and after constructing a nuclear medicine treatment individuation dose evaluation model based on the network DeepDose _T, supplement training on unfixed parameters of the network by using training sets with different distributions with small sample size during migration learning to obtain specific network parameters suitable for learning distribution samples so as to generate a nuclear medicine treatment individuation dose evaluation model after migration learning.
According to the embodiment of the application, the unfixed parameters of the network can be subjected to supplementary training to obtain the specific network parameters suitable for learning the distribution samples, so that the problem of individuation difference in the nuclear medicine dose prediction of the neural network is solved, the influence of the physical difference of a patient on the prediction result of the neural network is eliminated, and the accuracy of the dose prediction of the target voxels far from the source is improved.
Optionally, in one embodiment of the present application, the unfixed parameter is a deep network parameter of the encoding process.
In some embodiments, additional training can be performed through unfixed parameters such as deep network parameters in the encoding process, so as to provide basis for obtaining specific network parameters suitable for learning distribution samples, further eliminate influence of physical differences of patients on neural network prediction results, and improve accuracy of target voxel dose prediction far from a source.
Optionally, in one embodiment of the present application, when performing the transfer learning, performing supplemental training on the unfixed parameters of the network to obtain specific network parameters suitable for learning the distribution sample, including: and respectively selecting a plurality of layers of the convolution layer parameters in the coding process for training, obtaining an optimal network structure based on the verification set, and testing by using the test set to obtain specific network parameters.
In the actual execution process, when performing migration learning, the embodiment of the application can respectively select a plurality of layers of the convolutional layer parameters of the coding process for training, for example, deepDose _t27, deepDose _t34 and DeepDose _t41 can be respectively trained under three conditions of 27, 34 and 41 fixed layers, an optimal network structure is obtained based on a verification set, and a test set is utilized for testing, so that the specific network parameters are obtained, the problem of individuation difference in the nuclear medicine dose prediction of the neural network is solved, and the influence of physical differences of patients on the prediction result of the neural network is eliminated.
Optionally, in one embodiment of the present application, training the 3D-Unet network with the training set includes: taking the L1-norm between the real dose rate matrix and the network predicted dose rate matrix as a loss function, wherein the loss function is as follows:
Wherein y i and The dose truth value and the dose predicted value of the ith voxel are respectively represented, and n is the number of voxels.
In actual implementation, embodiments of the present application may provide for the partitioning of a material matrix (M, N x 24) and an activity matrix (a, N x 24) are each used as a channel, combined into the input of the network, the input dimension is (NX) 24×24×24×2). Each convolution block (block) in the encoding process comprises a 3 x 3 convolution layer (Convolutional layer), a normalized BN layer, a ReLU layer for activation, and a maximum pooling layer (MaxPooling) after every two convolution blocks, wherein the pooling layer has a size of 2 x 2, each direction step length is 2, two convolution blocks and a pooling layer are cooperated to form a first order, the feature layers of the same order have the same resolution (the number of voxels in XYZ directions), and finally the feature layers with 108 dimensions are obtained through the encoding of 8 convolution blocks of 3 orders. Up-sampling is performed in each stage of the decoding process using a 2 x 2 deconvolution layer (Deconvolutional layer), with a step size of 2, followed by two convolutional blocks, while, the output features of each stage in the encoding process are combined into the corresponding resolution stage in the decoding process, and the final convolution layer obtains single-channel output (Nx24×24×24×1) to obtain a predicted block dose rate matrix (Dpre).
Further, the embodiment of the application can take the L1-norm between the real dose rate matrix (DMC) and the CNN predicted dose rate matrix (Dpre) as a loss function, adopts an Adam optimizer to optimize, uses GPU (Tesla V100-PCIe-16GB, NVIDIA) for training the network, has the batch size of 300, trains 300 epochs altogether, and has the formula as follows:
Wherein y i and The dose truth value and the dose predicted value of the ith voxel are respectively represented, and n is the number of voxels.
The embodiment of the application can take the L1-norm between the real dose rate matrix and the dose rate matrix predicted by the network as a loss function, thereby further improving the accuracy of predicting the dose of the target voxels far from the source.
Specifically, the working principle of the nuclear medicine therapy personalized dose evaluation method based on the migration learning according to the embodiment of the application can be elaborated with reference to fig. 2.
As shown in fig. 2, the DeepDose _t transfer learning network structure in the embodiment of the present application includes 3 downward encoding processes and 3 upward sampling decoding processes, and finally returns to the original resolution, so that further investigation can be performed on the number of transfer learning layers, three cases of fixed layer numbers N of 27, 34 and 41 are respectively selected, deepDose _t27, deepDose _t34 and DeepDose _t41 are respectively trained, an optimal network structure is found based on the verification set, and the test set is tested.
Embodiments of the present application may provide for the partitioning of a material matrix (M, N x 24) and an activity matrix (a, N x 24) are each used as a channel, combined into the input of the network, the input dimension is (NX) 24×24×24×2). Each convolution block (block) in the encoding process comprises a 3 x 3 convolution layer (Convolutional layer), a normalized BN layer, a ReLU layer for activation, and a maximum pooling layer (MaxPooling) after every two convolution blocks, wherein the pooling layer has a size of 2x 2, each direction step length is 2, two convolution blocks and a pooling layer are cooperated to form a first order, the feature layers of the same order have the same resolution (the number of voxels in XYZ directions), and finally the feature layers with 108 dimensions are obtained through the encoding of 8 convolution blocks of 3 orders. Up-sampling is performed in each stage of the decoding process using a 2x 2 deconvolution layer (Deconvolutional layer), with a step size of 2, followed by two convolutional blocks, while, the output features of each stage in the encoding process are combined into the corresponding resolution stage in the decoding process, and the final convolution layer obtains single-channel output (Nx24×24×24×1) to obtain a predicted block dose rate matrix (Dpre).
Further, the embodiment of the application can take the L1-norm between the real dose rate matrix (DMC) and the CNN predicted dose rate matrix (Dpre) as a loss function, adopts an Adam optimizer to optimize, uses GPU (Tesla V100-PCIe-16GB, NVIDIA) for training the network, has the batch size of 300, trains 300 epochs altogether, and has the formula as follows:
Wherein y i and The dose truth value and the dose predicted value of the ith voxel are respectively represented, and n is the number of voxels.
According to the nuclear medicine treatment individuation dose evaluation method based on the transfer learning, which is provided by the embodiment of the application, the 3D-Unet network can be trained by utilizing the training set, and the unfixed parameters of the network are subjected to supplementary training when the transfer learning is carried out, so that the specific network parameters suitable for learning the distribution sample are obtained, the influence of the physical difference of a patient on the neural network prediction result is eliminated, and the accuracy of target voxel dose prediction far from a source is improved. Therefore, the problems that in the related technology, the neural network application has high dependence on a training set, the generalization capability is limited, the neural network application is difficult to accurately apply to the prediction of a test sample, meanwhile, the difference of the body types of patients can cause different distances of source target voxels, the energy deposition distribution of nuclides is further influenced, and the prediction result of the network obtained by case training of common physical parameters is inaccurate are solved.
Next, a nuclear medicine therapy individuation dose evaluation device based on transfer learning according to an embodiment of the present application will be described with reference to the accompanying drawings.
Fig. 3 is a schematic structural diagram of a nuclear medicine therapy personalized dose estimation device based on transfer learning according to an embodiment of the application.
As shown in fig. 3, the nuclear medicine therapy personalized dose evaluation device 10 based on transfer learning includes: a selection module 100, a generation module 200 and an evaluation module 300.
Specifically, the selection module 100 is configured to select a case sample that meets a preset condition with respect to an existing case physique parameter difference.
A generation module 200 for generating training sets of different physical patients based on the case samples.
The evaluation module 300 is configured to train the 3D-Unet network by using the training set, and perform supplemental training on unfixed parameters of the network during the transfer learning after constructing the personalized dose evaluation model for the nuclear medicine treatment based on the network DeepDose _t, so as to obtain specific network parameters suitable for learning the distribution sample, so as to generate the personalized dose evaluation model for the nuclear medicine treatment after the transfer learning.
Optionally, in one embodiment of the present application, the unfixed parameter is a deep network parameter of the encoding process.
Optionally, in one embodiment of the present application, the generating module 200 includes: and a generating unit.
The generation unit is used for reducing the activity matrix and the material matrix of the sample while keeping the resolution of the voxels unchanged, and sampling the difference after reducing the preset multiple to obtain the activity matrix and the material matrix of the small-size sample reaching the target small size.
Optionally, in one embodiment of the present application, the evaluation module 300 includes: training unit.
The training unit is used for respectively selecting a plurality of layers of the convolutional layer parameters of the coding process to train, obtaining an optimal network structure based on the verification set, and testing by utilizing the test set to obtain specific network parameters.
Optionally, in one embodiment of the present application, the evaluation module 300 includes: and an evaluation unit.
The evaluation unit is used for taking the L1-norm between the real dose rate matrix and the dose rate matrix predicted by the network as a loss function, wherein the loss function is as follows:
Wherein y i and The dose truth value and the dose predicted value of the ith voxel are respectively represented, and n is the number of voxels.
It should be noted that the foregoing explanation of the embodiment of the method for evaluating a personalized dose of a nuclear medicine treatment based on transfer learning is also applicable to the device for evaluating a personalized dose of a nuclear medicine treatment based on transfer learning of this embodiment, and will not be repeated here.
According to the nuclear medicine treatment individuation dosage evaluation device based on the transfer learning, which is provided by the embodiment of the application, the 3D-Unet network can be trained by utilizing the training set, and the unfixed parameters of the network are subjected to supplementary training when the transfer learning is carried out, so that the specific network parameters suitable for learning the distribution sample are obtained, the influence of the physical difference of a patient on the prediction result of the neural network is eliminated, and the accuracy of target voxel dosage prediction far from a source is improved. Therefore, the problems that in the related technology, the neural network application has high dependence on a training set, the generalization capability is limited, the neural network application is difficult to accurately apply to the prediction of a test sample, meanwhile, the difference of the body types of patients can cause different distances of source target voxels, the energy deposition distribution of nuclides is further influenced, and the prediction result of the network obtained by case training of common physical parameters is inaccurate are solved.
Fig. 4 is a schematic structural diagram of an electronic device according to an embodiment of the present application. The electronic device may include:
memory 401, processor 402, and a computer program stored on memory 401 and executable on processor 402.
The processor 402, when executing the program, implements the method for personalized dose assessment of nuclear medicine treatments based on transfer learning provided in the above embodiments.
Further, the electronic device further includes:
A communication interface 403 for communication between the memory 401 and the processor 402.
A memory 401 for storing a computer program executable on the processor 402.
Memory 401 may comprise high-speed RAM memory or may also include non-volatile memory (non-volatile memory), such as at least one disk memory.
If the memory 401, the processor 402, and the communication interface 403 are implemented independently, the communication interface 403, the memory 401, and the processor 402 may be connected to each other by a bus and perform communication with each other. The bus may be an industry standard architecture (Industry Standard Architecture, abbreviated ISA) bus, an external device interconnect (PERIPHERAL COMPONENT, abbreviated PCI) bus, or an extended industry standard architecture (Extended Industry Standard Architecture, abbreviated EISA) bus, among others. The buses may be divided into address buses, data buses, control buses, etc. For ease of illustration, only one thick line is shown in fig. 4, but not only one bus or one type of bus.
Alternatively, in a specific implementation, if the memory 401, the processor 402, and the communication interface 403 are integrated on a chip, the memory 401, the processor 402, and the communication interface 403 may complete communication with each other through internal interfaces.
Processor 402 may be a central processing unit (Central Processing Unit, abbreviated as CPU), or an Application SPECIFIC INTEGRATED Circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the application.
The present embodiment also provides a computer-readable storage medium having stored thereon a computer program which, when executed by a processor, implements the nuclear medicine therapy individualization dose assessment method based on transfer learning as above.
In the description of the present specification, a description referring to terms "one embodiment," "some embodiments," "examples," "specific examples," or "some examples," etc., means that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the present application. In this specification, schematic representations of the above terms are not necessarily directed to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or N embodiments or examples. Furthermore, the different embodiments or examples described in this specification and the features of the different embodiments or examples may be combined and combined by those skilled in the art without contradiction.
Furthermore, the terms "first," "second," and the like, are used for descriptive purposes only and are not to be construed as indicating or implying a relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defining "a first" or "a second" may explicitly or implicitly include at least one such feature. In the description of the present application, "N" means at least two, for example, two, three, etc., unless specifically defined otherwise.
Any process or method descriptions in flow charts or otherwise described herein may be understood as representing modules, segments, or portions of code which include one or more executable instructions for implementing specific logical functions or steps of the process, and additional implementations are included within the scope of the preferred embodiment of the present application in which functions may be executed out of order from that shown or discussed, including substantially concurrently or in reverse order from that shown or discussed, depending on the functionality involved, as would be understood by those reasonably skilled in the art of the embodiments of the present application.
Logic and/or steps represented in the flowcharts or otherwise described herein, e.g., a ordered listing of executable instructions for implementing logical functions, can be embodied in any computer-readable medium for use by or in connection with an instruction execution system, apparatus, or device, such as a computer-based system, processor-containing system, or other system that can fetch the instructions from the instruction execution system, apparatus, or device and execute the instructions. For the purposes of this description, a "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of the computer-readable medium would include the following: an electrical connection (electronic device) having one or N wires, a portable computer cartridge (magnetic device), a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). In addition, the computer readable medium may even be paper or other suitable medium on which the program is printed, as the program may be electronically captured, via optical scanning of the paper or other medium, then compiled, interpreted or otherwise processed in a suitable manner if necessary, and then stored in a computer memory.
It is to be understood that portions of the present application may be implemented in hardware, software, firmware, or a combination thereof. In the above-described embodiments, the N steps or methods may be implemented in software or firmware stored in a memory and executed by a suitable instruction execution system. As with the other embodiments, if implemented in hardware, may be implemented using any one or combination of the following techniques, as is well known in the art: discrete logic circuits having logic gates for implementing logic functions on data signals, application specific integrated circuits having suitable combinational logic gates, programmable Gate Arrays (PGAs), field Programmable Gate Arrays (FPGAs), and the like.
Those of ordinary skill in the art will appreciate that all or a portion of the steps carried out in the method of the above-described embodiments may be implemented by a program to instruct related hardware, where the program may be stored in a computer readable storage medium, and where the program, when executed, includes one or a combination of the steps of the method embodiments.
In addition, each functional unit in the embodiments of the present application may be integrated in one processing module, or each unit may exist alone physically, or two or more units may be integrated in one module. The integrated modules may be implemented in hardware or in software functional modules. The integrated modules may also be stored in a computer readable storage medium if implemented in the form of software functional modules and sold or used as a stand-alone product.
The above-mentioned storage medium may be a read-only memory, a magnetic disk or an optical disk, or the like. While embodiments of the present application have been shown and described above, it will be understood that the above embodiments are illustrative and not to be construed as limiting the application, and that variations, modifications, alternatives and variations may be made to the above embodiments by one of ordinary skill in the art within the scope of the application.
Claims (8)
1. A method for personalized dose assessment of nuclear medicine treatment based on transfer learning, comprising the steps of:
selecting case samples with the difference of the physical parameters of the existing cases meeting preset conditions;
generating a training set of different physical patients based on the case samples; and
Training a 3D-Unet network by using the training set, and after constructing a nuclear medicine treatment individuation dose evaluation model based on the network DeepDose _T, performing supplementary training on unfixed parameters of the network during transfer learning to obtain specific network parameters suitable for learning distribution samples so as to generate the nuclear medicine treatment individuation dose evaluation model after the transfer learning;
the generating training sets of different physical patients based on the case samples comprises:
Reducing the activity matrix and the material matrix of the sample while keeping the resolution of the voxels unchanged, and sampling the difference after reducing the preset multiple to obtain the activity matrix and the material matrix of the small-size sample reaching the target small size;
the unfixed parameters are deep network parameters in the encoding process;
The constructing of the nuclear medicine treatment individuation dosage evaluation model based on the network DeepDose _T comprises a coding process and a decoding process;
the encoding process includes:
Merging the activity matrix and the material matrix into a network input of the network DeepDose _t-based nuclear medicine treatment personalized dose assessment model;
Each convolution block comprises a convolution layer, a normalized BN layer, a ReLU layer for activation, and a max pooling layer is used after every two of said convolution blocks;
Laminating the convolution block and the pooling layer to obtain a characteristic layer;
The decoding process includes:
up-sampling using a deconvolution layer;
continuing the convolution block;
The original resolution of the voxel is obtained.
2. The method according to claim 1, wherein the performing of the transfer learning to perform the supplemental training on the unfixed parameters of the network to obtain specific network parameters suitable for learning the distribution sample comprises:
and respectively selecting a plurality of layers of the convolution layer parameters in the coding process for training, obtaining an optimal network structure based on the verification set, and testing by using the test set to obtain the specific network parameters.
3. The method of claim 1, wherein training the 3D-Unet network using the training set comprises:
taking the L1-norm between the real dose rate matrix and the network predicted dose rate matrix as a loss function, wherein the loss function is as follows:
，
Wherein, And/>The dose truth value and the dose predicted value of the ith voxel are respectively represented, and n is the number of voxels.
4. A migration learning based nuclear medicine therapy personalized dose assessment apparatus for implementing a migration learning based nuclear medicine therapy personalized dose assessment method according to any one of claims 1 to 3, the apparatus comprising:
The selection module is used for selecting case samples with the difference of the physical parameters of the existing cases meeting the preset conditions;
the generation module is used for generating training sets of patients with different physique based on the case samples; and
The evaluation module is used for training the 3D-Unet network by using the training set, and after constructing the nuclear medicine treatment individuation dose evaluation model based on the network DeepDose _T, performing supplementary training on unfixed parameters of the network during transfer learning to obtain specific network parameters suitable for learning distribution samples so as to generate the nuclear medicine treatment individuation dose evaluation model after the transfer learning.
5. The apparatus of claim 4, wherein the evaluation module comprises:
the training unit is used for respectively selecting a plurality of layers of the convolutional layer parameters in the coding process to train, obtaining an optimal network structure based on the verification set, and testing by utilizing the test set to obtain the specific network parameters.
6. The apparatus of claim 4, wherein the evaluation module comprises:
an evaluation unit, configured to take an L1-norm between a real dose rate matrix and a network predicted dose rate matrix as a loss function, where the loss function is:
，
Wherein, And/>The dose truth value and the dose predicted value of the ith voxel are respectively represented, and n is the number of voxels.
7. An electronic device, comprising: a memory, a processor and a computer program stored on the memory and executable on the processor, the processor executing the program to implement the method of transition learning based nuclear medicine therapy individualization dose assessment of any one of claims 1-3.
8. A computer-readable storage medium, on which a computer program is stored, characterized in that the program is executed by a processor for implementing the method for personalized dose assessment of nuclear medicine treatments based on transfer learning as claimed in any one of claims 1-3.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310470415.4A CN116825283B (en) | 2023-04-27 | 2023-04-27 | Nuclear medicine treatment individuation dosage evaluation method and device based on transfer learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310470415.4A CN116825283B (en) | 2023-04-27 | 2023-04-27 | Nuclear medicine treatment individuation dosage evaluation method and device based on transfer learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116825283A CN116825283A (en) | 2023-09-29 |
| CN116825283B true CN116825283B (en) | 2024-06-25 |
Family
ID=88121183
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310470415.4A Active CN116825283B (en) | 2023-04-27 | 2023-04-27 | Nuclear medicine treatment individuation dosage evaluation method and device based on transfer learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116825283B (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110995475A (en) * | 2019-11-20 | 2020-04-10 | 国网湖北省电力有限公司信息通信公司 | Power communication network fault detection method based on transfer learning |
| CN114266939A (en) * | 2021-12-23 | 2022-04-01 | 太原理工大学 | Brain extraction method based on ResTLU-Net model |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111239137B (en) * | 2020-01-09 | 2021-09-10 | 江南大学 | Grain quality detection method based on transfer learning and adaptive deep convolution neural network |
| US20220108171A1 (en) * | 2020-10-02 | 2022-04-07 | Google Llc | Training neural networks using transfer learning |
| AU2020103613A4 (en) * | 2020-11-23 | 2021-02-04 | Agricultural Information and Rural Economic Research Institute of Sichuan Academy of Agricultural Sciences | Cnn and transfer learning based disease intelligent identification method and system |
| CN115965076A (en) * | 2021-10-09 | 2023-04-14 | 中国移动通信集团终端有限公司 | Network model training method and equipment |
| CN114038535A (en) * | 2021-10-09 | 2022-02-11 | 海创时代(深圳)医疗科技有限公司 | Radiotherapy dose prediction device and method based on deep learning |
-
2023
- 2023-04-27 CN CN202310470415.4A patent/CN116825283B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110995475A (en) * | 2019-11-20 | 2020-04-10 | 国网湖北省电力有限公司信息通信公司 | Power communication network fault detection method based on transfer learning |
| CN114266939A (en) * | 2021-12-23 | 2022-04-01 | 太原理工大学 | Brain extraction method based on ResTLU-Net model |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116825283A (en) | 2023-09-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Gaser et al. | CAT–A computational anatomy toolbox for the analysis of structural MRI data | |
| CN109993726B (en) | Medical image detection method, device, equipment and storage medium | |
| Fu et al. | Deep learning approaches using 2D and 3D convolutional neural networks for generating male pelvic synthetic computed tomography from magnetic resonance imaging | |
| US10751548B2 (en) | Automated image segmentation using DCNN such as for radiation therapy | |
| Chen et al. | U‐net‐generated synthetic CT images for magnetic resonance imaging‐only prostate intensity‐modulated radiation therapy treatment planning | |
| US20200388035A1 (en) | Systems and methods for medical acquisition processing and machine learning for anatomical assessment | |
| US10282588B2 (en) | Image-based tumor phenotyping with machine learning from synthetic data | |
| CN111640100B (en) | Tumor image processing method and device, electronic equipment and storage medium | |
| CA2640802A1 (en) | Method and apparatus for accelerated elastic registration of multiple scans of internal properties of a body | |
| Sun et al. | Imaging study of pseudo-CT synthesized from cone-beam CT based on 3D CycleGAN in radiotherapy | |
| CN110751187B (en) | Training method of abnormal area image generation network and related product | |
| Vu et al. | Deep convolutional neural networks for automatic segmentation of thoracic organs‐at‐risk in radiation oncology–use of non‐domain transfer learning | |
| US20220189028A1 (en) | Automatic detection of lesions in medical images using 2d and 3d deep learning networks | |
| US11908568B2 (en) | System and methods for radiographic image quality assessment and protocol optimization | |
| Holmes et al. | Creation of an anthropomorphic CT head phantom for verification of image segmentation | |
| Farjam et al. | Deep learning‐based synthetic CT generation for MR‐only radiotherapy of prostate cancer patients with 0.35 T MRI linear accelerator | |
| JP2023133132A (en) | Transformer feature sharing-based pet image reconstruction method, apparatus and device and medium | |
| Kläser et al. | Deep boosted regression for MR to CT synthesis | |
| Leu et al. | Generation of pseudo-CT using high-degree polynomial regression on dual-contrast pelvic MRI data | |
| Meaney et al. | Deep learning characterization of brain tumours with diffusion weighted imaging | |
| Lan et al. | Potential roles of transformers in brain tumor diagnosis and treatment | |
| Zhou et al. | Multimodality MRI synchronous construction based deep learning framework for MRI-guided radiotherapy synthetic CT generation | |
| CN110610527B (en) | SUV computing method, device, equipment, system and computer storage medium | |
| CN116825283B (en) | Nuclear medicine treatment individuation dosage evaluation method and device based on transfer learning | |
| Zhao et al. | [Retracted] Application of Artificial Intelligence in Radiotherapy of Nasopharyngeal Carcinoma with Magnetic Resonance Imaging |
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 | ||
| GR01 | Patent grant |