CN115880440A - Magnetic particle three-dimensional reconstruction imaging method based on generation of countermeasure network - Google Patents
Magnetic particle three-dimensional reconstruction imaging method based on generation of countermeasure network Download PDFInfo
- Publication number
- CN115880440A CN115880440A CN202310047876.0A CN202310047876A CN115880440A CN 115880440 A CN115880440 A CN 115880440A CN 202310047876 A CN202310047876 A CN 202310047876A CN 115880440 A CN115880440 A CN 115880440A
- Authority
- CN
- China
- Prior art keywords
- image
- dimensional
- network
- mpi
- attention
- 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.)
- Granted
Links
- 238000003384 imaging method Methods 0.000 title claims abstract description 49
- 239000006249 magnetic particle Substances 0.000 title claims abstract description 36
- 238000009792 diffusion process Methods 0.000 claims abstract description 85
- 238000000034 method Methods 0.000 claims abstract description 44
- 238000005070 sampling Methods 0.000 claims abstract description 16
- 238000001914 filtration Methods 0.000 claims abstract description 7
- 230000006870 function Effects 0.000 claims description 126
- 230000004913 activation Effects 0.000 claims description 71
- 230000007246 mechanism Effects 0.000 claims description 35
- 239000002245 particle Substances 0.000 claims description 31
- 230000004927 fusion Effects 0.000 claims description 30
- 238000013527 convolutional neural network Methods 0.000 claims description 26
- 238000012545 processing Methods 0.000 claims description 15
- 230000009466 transformation Effects 0.000 claims description 13
- 230000008569 process Effects 0.000 claims description 12
- 238000012549 training Methods 0.000 claims description 12
- 210000002569 neuron Anatomy 0.000 claims description 10
- 239000011159 matrix material Substances 0.000 claims description 9
- 230000003042 antagnostic effect Effects 0.000 claims description 8
- 238000005457 optimization Methods 0.000 claims description 4
- 230000005484 gravity Effects 0.000 claims description 3
- 238000013528 artificial neural network Methods 0.000 claims description 2
- 230000001131 transforming effect Effects 0.000 claims description 2
- 238000002059 diagnostic imaging Methods 0.000 abstract description 2
- 238000010586 diagram Methods 0.000 description 4
- 238000012360 testing method Methods 0.000 description 3
- 238000013461 design Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000006467 substitution reaction Methods 0.000 description 2
- 238000003325 tomography Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 239000003814 drug Substances 0.000 description 1
- 229940079593 drug Drugs 0.000 description 1
- 230000005855 radiation Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 230000008685 targeting Effects 0.000 description 1
Images
Landscapes
- Image Analysis (AREA)
Abstract
The invention belongs to the technical field of medical imaging, and particularly relates to a magnetic particle three-dimensional reconstruction imaging method, system and device based on a generation countermeasure network, aiming at solving the problems of long reconstruction time and low imaging resolution of the existing MPI high-resolution image. The method comprises the following steps: acquiring a multi-view sparse two-dimensional MPI image of a simulated body of an object to be imaged and reconstructed, and performing down-sampling to obtain a down-sampled sparse two-dimensional MPI image; inputting the downsampled sparse two-dimensional MPI image into a trained neighborhood point average diffusion search self-attention generation confrontation network generation model to obtain a dense two-dimensional MPI image; reconstructing each dense two-dimensional MPI image through a filtering back projection reconstruction algorithm to finally obtain a three-dimensional MPI image of a reconstructed object to be imaged; the invention accelerates the reconstruction speed of MPI images, improves the imaging resolution and ensures that the magnetic particle imaging equipment has greater application prospect in the medical field.
Description
Technical Field
The invention belongs to the technical field of medical imaging, and particularly relates to a magnetic particle three-dimensional reconstruction imaging method, system and device based on a generation countermeasure network.
Background
Magnetic particle imaging is a novel imaging technology at present, and has the advantages of no radiation, high sensitivity and the like. At present, magnetic particle imaging equipment has wide application prospects in clinical research, including focus early detection, drug targeting treatment and other forward directions. In the existing magnetic particle imaging apparatus, how to quickly realize scanning of an object and high-resolution three-dimensional image reconstruction is a challenging problem at present. At present, part of devices implement tomography scanning on an object by rotating a device coil, scan two-dimensional images of the object at different angles by physically rotating an imaged coil, and then perform filtered back-projection reconstruction on the scanned images. However, the imaging technology needs to perform coil rotation movement, and the service life of the equipment is shortened due to the fact that the rotating speed is too high. Secondly, if the time resolution of the scanning is to be improved, multiple scans are required, and the process is time-consuming and labor-consuming. Therefore, how to improve the reconstruction speed and the imaging resolution of imaging is a difficult problem which needs to be faced at present under the condition of considering the construction cost of hardware equipment. Based on the method, the invention provides a magnetic particle three-dimensional reconstruction imaging method based on generation of a countermeasure network.
Disclosure of Invention
In order to solve the above problems in the prior art, that is, to solve the problems of long reconstruction time and low imaging resolution of the existing MPI three-dimensional high resolution image, the invention provides a magnetic particle three-dimensional reconstruction imaging method based on a generation countermeasure network, which comprises the following steps:
s100, acquiring a multi-view sparse two-dimensional MPI image of a simulated body of a to-be-imaged reconstructed object, and performing down-sampling to obtain a down-sampled sparse two-dimensional MPI image; the multi-view sparse two-dimensional MPI image is a plurality of MPI projection images which are collected in sequence according to a set angle rotation;
step S200, inputting the down-sampled sparse two-dimensional MPI image into a trained neighborhood point average diffusion search self-attention generation confrontation network generation model to obtain a dense two-dimensional MPI image;
s300, reconstructing each dense two-dimensional MPI image through a filtering back projection reconstruction algorithm to finally obtain a three-dimensional MPI image of a to-be-imaged reconstructed object;
the neighborhood point average diffusion search self-attention generation countermeasure network comprises a generation model and a discrimination model;
the generation model comprises five diffusion search attention mechanism convolution modules; the diffusion search attention mechanism convolution module is constructed on the basis of a neighborhood point average diffusion convolution sub-network, a self-attention fusion network and an activation function layer which are sequentially connected; the activation function adopted by the activation function layer in the first four diffusion search attention machine convolution modules is Leaky ReLU, and the activation function adopted by the activation function layer in the fifth diffusion search attention machine convolution module is hyperbolic tangent function;
the neighborhood point average diffusion convolution sub-network is configured to sequentially encode the down-sampled sparse two-dimensional MPI image through an encoding function to obtain a sequentially encoded image; performing regional convolution operation on each sequential coded image to acquire particle concentration information, angle information and correlation degree information among the images of each sequential coded image;
the self-attention fusion network is configured to multiply particle concentration information, angle information and correlation degree information among images of each sequence coding image with corresponding weight matrixes to perform linear transformation, perform fusion through a multi-head attention mechanism, and perform up-sampling after fusion to obtain a dense two-dimensional MPI image;
the discrimination model comprises five convolutional neural network modules; the convolutional neural network module is constructed on the basis of a convolutional layer, a regularization layer and an activation function layer which are connected in sequence; the activation function adopted by the activation function layers of the first four convolutional neural network modules is Leaky ReLU, and the activation function adopted by the activation function layer of the fifth convolutional neural network module is softmax.
In some preferred embodiments, the training process of the neighborhood point average diffusion search self-attention generation countermeasure network is as follows:
a100, acquiring a multi-view sparse two-dimensional MPI image of a simulated body of a to-be-imaged reconstructed object through MPI imaging equipment, and performing down-sampling to obtain a down-sampled sparse two-dimensional MPI image; obtaining a training data set based on the down-sampled sparse two-dimensional MPI image and a corresponding truth label;
step A200, inputting the down-sampled sparse two-dimensional MPI image into a pre-constructed neighborhood point average diffusion search self-attention generation confrontation network generation model to obtain a dense two-dimensional MPI image;
step A300, inputting the dense two-dimensional MPI image and a truth label corresponding to the dense two-dimensional MPI image into the neighborhood point average diffusion search self-attention to generate a discrimination model of an anti-network, and obtaining a discrimination result of the dense two-dimensional MPI image;
step A400, based on the discrimination result, combining each dense two-dimensional MPI image, each sparse two-dimensional MPI image and the corresponding truth value label thereof, calculating a total loss value through a pre-constructed loss function, and updating the network parameters of the generation model and the discrimination model of the self-attention generation countermeasure network of the neighborhood point average diffusion search;
step A500, training the generator network and the discriminator network of the anti-network generated by the neighborhood point average diffusion search self-attention circularly until the trained neighborhood point average diffusion search self-attention generation anti-network is obtained.
In some preferred embodiments, let the number of angles generated beAt an angle ofWhen it comes toThe coding function of the sparse two-dimensional MPI image is:
In some preferred embodiments, the region convolution operation is performed on the sequentially encoded image by: processing the concentration value of the pixel point in the sequential coding image, and calculating by adopting a3 multiplied by 3 convolution kernel after the processing;
the method for processing the density value of the pixel point in the sequential coding image comprises the following steps:
wherein ,representing the first of said sequentially encoded picturesGo to the firstThe number of pixels in a column is,representing a neighborhood of pointsThe average of all the values within the range,representThe minimum of all the values in the neighborhood of the point,representThe maximum of all the values in the neighborhood of the point,representThe variable values in the neighborhood of points are composed of points adjacent in the transverse direction, the longitudinal direction and the diagonal line.
In some preferred embodiments, the particle density information, the angle information, and the inter-image correlation information of each sequentially encoded image are multiplied by a corresponding weight matrix to perform linear transformation, and the method includes:
wherein ,representing self-attention mechanism, vectorThe information on the concentration of the particles is represented,(ii) a Vector quantityRepresentThe information on the angle of the light source,(ii) a Vector quantityInformation indicating the degree of association between the images,,representing a transformation of a matrix into a one-dimensional column vector, vectorA sequentially encoded image representing an input neighborhood point average diffusion convolution sub-network;、、respectively representing weight matrixes corresponding to particle density information, angle information and inter-picture correlation information of sequentially encoded pictures,representing each sequentially coded pictureNeurons in all output layersThe sum of the output results from 1 to N,representing each sequentially coded pictureThrough the first stepThe output value of each of the neurons is,representing respective sequentially coded picturesThrough the first stepThe output value of each of the neurons is,representing each sequentially coded pictureThrough the first stepThe output value of each neuron accounts for the total output valueSpecific gravity of (a).
In some preferred embodiments, the particle concentration information, the angle information and the correlation information between the images of each sequentially encoded image are fused by a multi-head attention mechanism, and the method comprises the following steps:
wherein ,is a function of the splicing function,is shown asThe output of each attention mechanism sub-module,an overall weight matrix representing the output of the multi-headed attention mechanism,information obtained by fusing particle density information, angle information, and inter-image correlation information of sequentially encoded images.
In some preferred embodiments, the total loss value is calculated by a pre-constructed loss function by:
wherein ,as a function of the total loss, the loss,representing a pixel loss function and a counter loss function,the angular order sorting optimization function is represented,which is indicative of an error in the pixel information,representing the loss of the generative model and,representing sparse twoThe dimension of the MPI image is measured,the first of representation generationThe two-dimensional MPI image is stretched and compacted,the first of representation generationThe two-dimensional MPI image is stretched and compacted,is shown asA true value label corresponding to the sparse two-dimensional MPI image,a coefficient representing a loss of a pixel,the coefficient representing the resistance to the loss is,indicating the probability of being judged as true by the discrimination network,the output of the generative model is represented,a function representing the difference in the angle is shown,in the form of a function of the hyperbolic tangent,indicating the predicted value of the angle at the corresponding input,,representing arbitrary two dense two-dimensional MPI imagesThe function of the judgment of the angle,andrespectively represent the firstAngle labels corresponding to the generated dense two-dimensional MPI images,parameters representing richness settings for different mock-containing information.
In a second aspect of the present invention, a magnetic particle three-dimensional reconstruction imaging system based on generation of a countermeasure network is presented, the system comprising: the image reconstruction system comprises an image acquisition module 100, an image reconstruction module 200 and a three-dimensional reconstruction module 300;
the image acquisition module 100 is configured to acquire a multi-view sparse two-dimensional MPI image of a phantom of an object to be imaged and reconstructed, and perform downsampling to obtain a downsampled sparse two-dimensional MPI image; the multi-view sparse two-dimensional MPI images are a plurality of MPI projection images which are sequentially collected according to a set angle rotation;
the image reconstruction module 200 is configured to input the downsampled sparse two-dimensional MPI image into a trained neighborhood point average diffusion search self-attention generation confrontation network generation model to obtain a dense two-dimensional MPI image;
the three-dimensional reconstruction module 300 is configured to reconstruct each dense two-dimensional MPI image through a filtering back projection reconstruction algorithm, and finally obtain a three-dimensional MPI image of a reconstructed object to be imaged;
the neighborhood point average diffusion search self-attention generation countermeasure network comprises a generation model and a discrimination model;
the generation model comprises five diffusion search attention mechanism convolution modules; the diffusion search attention mechanism convolution module is constructed on the basis of a neighborhood point average diffusion convolution sub-network, a self-attention fusion network and an activation function layer which are sequentially connected; the activation function adopted by the activation function layer in the first four diffusion search attention machine convolution modules is Leaky ReLU, and the activation function adopted by the activation function layer in the fifth diffusion search attention machine convolution module is hyperbolic tangent function;
the neighborhood point average diffusion convolution sub-network is configured to sequentially encode the down-sampled sparse two-dimensional MPI image through an encoding function to obtain a sequentially encoded image; performing regional convolution operation on each sequential coded image to acquire particle concentration information, angle information and correlation degree information among the images of each sequential coded image;
the self-attention fusion network is configured to multiply particle concentration information, angle information and correlation degree information among images of each sequence coding image with corresponding weight matrixes to perform linear transformation, perform fusion through a multi-head attention mechanism, and perform up-sampling after fusion to obtain a dense two-dimensional MPI image;
the discrimination model comprises five convolution neural network modules; the convolutional neural network module is constructed on the basis of a convolutional layer, a regularization layer and an activation function layer which are connected in sequence; the activation function adopted by the activation function layers of the first four convolutional neural network modules is Leaky ReLU, and the activation function adopted by the activation function layer of the fifth convolutional neural network module is softmax.
In a third aspect of the present invention, a storage device is provided, in which a plurality of programs are stored, the programs being adapted to be loaded and executed by a processor to implement the above-mentioned method for three-dimensional reconstruction imaging based on magnetic particles generating a countermeasure network.
In a fourth aspect of the invention, a processing arrangement is provided, comprising a processor, a storage device; a processor adapted to execute various programs; a storage device adapted to store a plurality of programs; the program is suitable for being loaded and executed by a processor to realize the magnetic particle three-dimensional reconstruction imaging method based on the generation countermeasure network.
The invention has the beneficial effects that:
the invention improves the reconstruction speed and the imaging resolution of MPI imaging;
the method integrates the neighborhood point average diffusion search self-attention mechanism into the generation of the countermeasure network, optimizes by utilizing the angle sequence ordering optimization function, accelerates the MPI image reconstruction speed, improves the MPI image reconstruction quality, makes up the hardware defects existing in the current equipment, and enables the magnetic particle imaging equipment to have a larger application prospect in the medical field.
Drawings
Other features, objects and advantages of the present application will become more apparent upon reading of the following detailed description of non-limiting embodiments thereof, made with reference to the accompanying drawings in which:
FIG. 1 is a schematic flow chart of a magnetic particle three-dimensional reconstruction imaging method based on generation of a countermeasure network according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of a training flow of a magnetic particle three-dimensional reconstruction imaging method based on generation of a countermeasure network according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of a network module of an attention mechanism in accordance with an embodiment of the present invention;
FIG. 4 is a schematic network structure diagram of a magnetic particle three-dimensional reconstruction imaging method based on generation of a countermeasure network according to an embodiment of the present invention;
fig. 5 is a schematic diagram of a framework of a magnetic particle three-dimensional reconstruction imaging system based on generation of a countermeasure network according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, 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 some, but not all embodiments of the present invention. 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 present application will be described in further detail with reference to the following drawings and examples. It is to be understood that the specific embodiments described herein are merely illustrative of the relevant invention and not restrictive of the invention. It should be noted that, for convenience of description, only the portions related to the related invention are shown in the drawings.
It should be noted that the embodiments and features of the embodiments in the present application may be combined with each other without conflict. The present application will be described in detail below with reference to the embodiments with reference to the attached drawings.
The invention provides a magnetic particle three-dimensional reconstruction imaging method based on generation of a countermeasure network, as shown in figure 1, the method comprises the following steps:
s100, acquiring a multi-view sparse two-dimensional MPI image of a simulated body of a to-be-imaged reconstructed object, and performing down-sampling to obtain a down-sampled sparse two-dimensional MPI image; the multi-view sparse two-dimensional MPI images are a plurality of MPI projection images which are sequentially collected according to a set angle rotation;
step S200, inputting the down-sampled sparse two-dimensional MPI image into a trained neighborhood point average diffusion search self-attention generation confrontation network generation model to obtain a dense two-dimensional MPI image;
s300, reconstructing each dense two-dimensional MPI image through a filtering back projection reconstruction algorithm to finally obtain a three-dimensional MPI image of a to-be-imaged reconstructed object;
the neighborhood point average diffusion search self-attention generation antagonistic network comprises a generation model and a discrimination model;
the generation model comprises five diffusion search attention mechanism convolution modules; the diffusion search attention mechanism convolution module is constructed on the basis of a neighborhood point average diffusion convolution sub-network, a self-attention fusion network and an activation function layer which are sequentially connected; the activation function adopted by the activation function layer in the first four diffusion search attention machine convolution modules is Leaky ReLU, and the activation function adopted by the activation function layer in the fifth diffusion search attention machine convolution module is hyperbolic tangent function;
the neighborhood point average diffusion convolution sub-network is configured to sequentially encode the down-sampled sparse two-dimensional MPI image through an encoding function to obtain a sequentially encoded image; performing regional convolution operation on each sequential coded image to acquire particle concentration information, angle information and correlation degree information among the images of each sequential coded image;
the self-attention fusion network is configured to multiply particle concentration information, angle information and correlation degree information among images of each sequence coding image with corresponding weight matrixes to perform linear transformation, perform fusion through a multi-head attention mechanism, and perform up-sampling after fusion to obtain a dense two-dimensional MPI image;
the discrimination model comprises five convolutional neural network modules; the convolutional neural network module is constructed on the basis of a convolutional layer, a regularization layer and an activation function layer which are connected in sequence; the activation function adopted by the activation function layers of the first four convolutional neural network modules is Leaky ReLU, and the activation function adopted by the activation function layer of the fifth convolutional neural network module is softmax.
In order to more clearly describe the magnetic particle three-dimensional reconstruction imaging method based on generation of the countermeasure network, the following describes each step in the embodiment of the method in detail with reference to the accompanying drawings.
In the existing commercial equipment for realizing Magnetic Particle Imaging (MPI) Imaging process, in order to image the distribution of particles inside an object, the object needs to be scanned, but the time for reconstructing an image based on a physical scanning mode of coil rotation is long at present, multi-angle tomography is limited by the rotation speed of the coil, the hardware cost of the equipment is greatly increased, and the service life of the equipment is shortened. In order to effectively reduce reconstruction time and save hardware cost, the invention designs a Multi-angle Three-Dimensional Magnetic Particle Imaging (MV-3D-MPI) method and a system based on a generated discriminant model. The invention provides a neighborhood point average diffusion search self-attention generation antagonistic network model, aiming at realizing the rapid reconstruction of an MPI high-resolution image and improving the imaging resolution.
In the following embodiments, a training process of the neighborhood point average diffusion search self-attention generation countermeasure network is described, and then a test process of the neighborhood point average diffusion search self-attention generation countermeasure network is described in detail.
1. The training process of the neighborhood point average diffusion search self-attention generation countermeasure network comprises the following specific steps:
a100, acquiring a multi-view sparse two-dimensional MPI image of a simulated body of a to-be-imaged reconstructed object through MPI imaging equipment, and performing down-sampling to obtain a down-sampled sparse two-dimensional MPI image; obtaining a training data set based on the down-sampled sparse two-dimensional MPI image and a corresponding truth label;
in the embodiment, a multi-view dense MPI projection image is acquired by an own MPI device, and the 15000 image is used as a training set; for a phantom, preferably acquiring one image every 5 ° (in other embodiments, the images may be acquired at set angle intervals according to actual requirements), and acquiring projection MPI images of 36 angles in total; the pixel size is changed to 12 × 12 by an arbitrary conventional downsampling method, and the downsampled image is set as an input target of the network.
Step A200, inputting the down-sampled sparse two-dimensional MPI image into a pre-constructed neighborhood point average diffusion search self-attention generation confrontation network generation model to obtain a dense two-dimensional MPI image;
in this embodiment, the neighborhood point average diffusion search self-attention generation countermeasure network includes a generation model and a discrimination model;
the generation model comprises five diffusion search attention mechanism convolution modules; the diffusion search attention mechanism convolution module is constructed on the basis of a neighborhood point average diffusion convolution sub-network, a self-attention fusion network and an activation function layer which are sequentially connected; the activation function adopted by the activation function layer in the first four diffusion search attention machine convolution modules is Leaky ReLU, and the activation function adopted by the activation function layer in the fifth diffusion search attention machine convolution module is hyperbolic tangent function;
the neighborhood point average diffusion convolution sub-network is configured to sequentially encode the down-sampled sparse two-dimensional MPI image through an encoding function to obtain a sequentially encoded image; performing regional convolution operation on each sequential coded image to acquire particle concentration information, angle information and correlation degree information among the images of each sequential coded image;
the self-attention fusion network is configured to multiply particle concentration information, angle information and correlation degree information among images of each sequence coding image with corresponding weight matrixes to perform linear transformation, perform fusion through a multi-head attention mechanism, and perform up-sampling after fusion to obtain a dense two-dimensional MPI image;
in the invention, because the input has the sequential relation of angles, in order to enable the model to learn the relevance of adjacent angles and the global relevance of different angles, the input layer of the network adopts sequential coding, the input of the model is set to comprise 6 images of different angles, one image is taken every 30 degrees in [0 DEG and 180 DEG ], and the angle isWhile to the secondCoding function of sparse two-dimensional MPI imageComprises the following steps:
wherein , the number of the generated angles is represented, and sine and cosine functions in the formula can ensure that the constructed model learns the relative position relation between different images.
Aiming at the characteristics of the MPI image, the neighborhood point diffusion search value is mainly used for more fully modeling the relationship between a certain pixel value of the MPI image and the neighborhood of the pixel value, and the density value of the image at a part of points is related to the density values of other surrounding pixel points due to the influence of the magnetic particle response of the magnetic field free region by the nearby particles. Performing a region convolution operation on the sequentially encoded image, the method comprising: processing the density value of the pixel point in the sequential coding image, and preferably adopting a3 x 3 convolution kernel to carry out operation after processing;
the method for processing the density value of the pixel point in the sequential coding image comprises the following steps:
to be provided withThe method for processing the density value by taking the point as an example comprises the following steps:
wherein ,representing the first of said sequentially encoded picturesGo to the firstThe number of pixels of a column is,representing a neighborhood of pointsThe average of all the values within the range,to representThe minimum of all the values in the neighborhood of the point,to representThe maximum of all the values in the neighborhood of the point,representThe variable values in the neighborhood of points are composed of points that are adjacent in the lateral, longitudinal, and diagonal directions.
The invention provides a self-attention fusion network module for fusing key information extracted from images by different sub-networks in a network model generation part. Let the input of each sub-network be the pre-processed imageProcessing input by three sub-networks to obtain different types of information, multiplying the obtained three types of different information by corresponding weight matrixes respectively for linear transformation, wherein the corresponding weight matrixes are、、Then, different types of information are fused through a self-attention fusion network module, and the particle concentration information, the angle information and the correlation degree information among the images of each sequence coding image are multiplied by a corresponding weight matrix for linear transformation, wherein the method comprises the following steps:
wherein ,indicating a self-attention mechanism, vectorThe information on the concentration of the particles is represented,(ii) a Vector quantityThe information on the angle is represented by the angle information,(ii) a Vector quantityInformation indicating the degree of association between the images,,representing a transformation of a matrix into a one-dimensional column vector, vectorA sequentially encoded image representing an input neighborhood point average diffusion convolution sub-network;、、respectively representing weight matrixes corresponding to particle density information, angle information and inter-picture correlation information of sequentially encoded pictures,representing each sequentially coded pictureNeurons in all output layersThe sum of the output results from 1 to N,representing respective sequentially coded picturesThrough the first stepThe output value of each of the neurons is,representing each sequentially coded pictureThrough the first passThe output value of each of the neurons is,representing each sequentially coded pictureThrough the first stepThe output value of each neuron accounts for the total output valueSpecific gravity of (a).
Fusing particle concentration information, angle information and correlation degree information among images of each sequence coding image through a multi-head attention mechanism, wherein the method comprises the following steps:
wherein ,is a function of the splicing function,denotes the firstThe output of each attention mechanism sub-module,an overall weight matrix representing the output of a multi-headed attention mechanism,the information obtained by fusing particle concentration information, angle information and image correlation information of sequentially encoded images is spliced with the output of each attention submoduleMultiplying, linear transforming, and up-sampling to obtain the output of generated model, Representing an upsampling function.
Step A300, inputting the intensive two-dimensional MPI image and the corresponding truth label into the neighborhood point average diffusion search self-attention to generate a discrimination model of a countermeasure network, and acquiring a discrimination result of the intensive two-dimensional MPI image;
the discrimination model comprises five convolutional neural network modules; the convolutional neural network module is constructed on the basis of a convolutional layer, a regularization layer and an activation function layer which are connected in sequence; the activation function adopted by the activation function layers of the first four convolutional neural network modules is Leaky ReLU, and the activation function adopted by the activation function layer of the fifth convolutional neural network module is softmax.
The convolution layer carries out convolution operation on input data of the discrimination model, then carries out regularization through the regularization layer, and finally inputs the input data into the activation function layer to carry out function operation.
Step A400, based on the discrimination result, combining each dense two-dimensional MPI image, each sparse two-dimensional MPI image and the corresponding truth value label thereof, calculating a total loss value through a pre-constructed loss function, and updating the network parameters of the generation model and the discrimination model of the self-attention generation countermeasure network of the neighborhood point average diffusion search;
in this embodiment, the total loss value is calculated by a pre-constructed loss function, and the method includes:
wherein ,as a function of the total loss, the loss,representing a pixel loss function and a counter loss function,the angular order sorting optimization function is represented,which is indicative of an error in the pixel information,representing the loss of the generative model,represents a sparse two-dimensional MPI image,representation generation of the firstThe two-dimensional MPI image is stretched and compacted,the first of representation generationThe two-dimensional MPI image is stretched and compacted,is shown asA true value label corresponding to the sparse two-dimensional MPI image,a coefficient representing a loss of a pixel,coefficient representing resistance loss, in the present embodimentAndit is preferably set to 1 and,representing the probability of whether the discrimination by the discrimination network is true,the output of the generative model is represented,a function representing the difference in the angle is shown,in the form of a function of the hyperbolic tangent,representing the predicted value of the angle at the corresponding input,,representing arbitrary two dense two-dimensional MPI imagesThe angle judgment function takes the value of 1 if the judgment value is true, otherwise takes the value of 0,andrespectively representAngle labels corresponding to the generated dense two-dimensional MPI images;parameters representing richness settings of information contained in different imitations, in the invention, if the collected imitations are letter imitations, the parameters are set for the letter imitationsPreferably set to 1, for more complex Chinese character imitationsPreferably set to 2, for a phantom specifically used to evaluate resolution since the details are rich,preferably set to 3. When the image pairs are taken, the image pairs taken by each phantom image are supervised by taking the phantom as a unit, and the wrong sorting is punished. The inner diameter of the filling of the three sets of mimetics is preferably 0.5mm with 10 mg/ml Perimag solution. Three sets of phantom field of view (FO)V) are preferably 6cm × 6cm × 6cm (letters), 6cm × 6cm × 10 cm (helix, acquisition time of one sparse projection is 3 minutes), 6cm × 6cm × cm (container, acquisition time of one sparse projection is 2 minutes), respectively.
Step A500, circularly training a generator network and a discriminator network of the neighborhood point average diffusion search self-attention generation antagonistic network until a trained neighborhood point average diffusion search self-attention generation antagonistic network is obtained;
2. testing process for neighborhood point average diffusion search self-attention generation countermeasure network
S100, acquiring a multi-view sparse two-dimensional MPI image of a phantom of a to-be-imaged reconstructed object, and performing down-sampling to obtain a down-sampled sparse two-dimensional MPI image; the multi-view sparse two-dimensional MPI image is a plurality of MPI projection images which are collected in sequence according to a set angle rotation;
in the present embodiment, 200 images acquired by the MPI apparatus are taken as a test set.
Step S200, inputting the down-sampled sparse two-dimensional MPI image into a trained neighborhood point average diffusion search self-attention generation confrontation network generation model to obtain a dense two-dimensional MPI image;
in the present embodiment, the antagonistic network is generated based on the trained neighborhood point average diffusion search self-attention, and preferably, the sparse two-dimensional MPI image is generated from any angle of [0 °,5 °,10 °,15 °,20 °,25 °, 30 °,35 °,40 °,45 °,50 °,55 °,60 °,65 °, \ 8230;, 170 °,175 ° ] and the generated dense two-dimensional MPI image has a pixel size of 48 × 48.
S300, reconstructing each dense two-dimensional MPI image through a filtering back projection reconstruction algorithm to finally obtain a three-dimensional MPI image of a to-be-imaged reconstructed object;
a magnetic particle three-dimensional reconstruction imaging system based on generation of a countermeasure network according to a second embodiment of the present invention, as shown in fig. 4, includes: the image reconstruction system comprises an image acquisition module 100, an image reconstruction module 200 and a three-dimensional reconstruction module 300;
the image acquisition module 100 is configured to acquire a multi-view sparse two-dimensional MPI image of a phantom of an object to be imaged and reconstructed, and perform downsampling to obtain a downsampled sparse two-dimensional MPI image; the multi-view sparse two-dimensional MPI image is a plurality of MPI projection images which are collected in sequence according to a set angle rotation;
the image reconstruction module 200 is configured to input the downsampled sparse two-dimensional MPI image into a trained neighborhood point average diffusion search self-attention generation confrontation network generation model to obtain a dense two-dimensional MPI image;
the three-dimensional reconstruction module 300 is configured to reconstruct each dense two-dimensional MPI image by a filtered back projection reconstruction algorithm, and finally obtain a three-dimensional MPI image of a to-be-imaged reconstructed object;
the neighborhood point average diffusion search self-attention generation countermeasure network comprises a generation model and a discrimination model;
the generation model comprises five diffusion search attention mechanism convolution modules; the diffusion search attention mechanism convolution module is constructed on the basis of a neighborhood point average diffusion convolution sub-network, a self-attention fusion network and an activation function layer which are sequentially connected; the activation function adopted by the activation function layer in the first four diffusion search attention machine convolution modules is Leaky ReLU, and the activation function adopted by the activation function layer in the fifth diffusion search attention machine convolution module is hyperbolic tangent function;
the neighborhood point average diffusion convolution sub-network is configured to sequentially encode the down-sampled sparse two-dimensional MPI image through an encoding function to obtain a sequentially encoded image; performing regional convolution operation on each sequential coded image to acquire particle concentration information, angle information and correlation degree information among the images of each sequential coded image;
the self-attention fusion network is configured to multiply particle concentration information, angle information and correlation degree information among images of each sequence coding image with corresponding weight matrixes to perform linear transformation, perform fusion through a multi-head attention mechanism, and perform up-sampling after fusion to obtain a dense two-dimensional MPI image;
the discrimination model comprises five convolutional neural network modules; the convolutional neural network module is constructed on the basis of a convolutional layer, a regularization layer and an activation function layer which are connected in sequence; the activation function adopted by the activation function layers of the first four convolutional neural network modules is LeakyReLU, and the activation function adopted by the activation function layer of the fifth convolutional neural network module is softmax.
It can be clearly understood by those skilled in the art that, for convenience and brevity of description, the specific working process and related description of the system described above may refer to the corresponding process in the foregoing method embodiment, and details are not described herein again.
It should be noted that, the magnetic particle three-dimensional reconstruction imaging system based on generation of a countermeasure network provided in the foregoing embodiment is only illustrated by dividing the functional modules, and in practical applications, the above functions may be allocated to different functional modules according to needs, that is, the modules or steps in the embodiments of the present invention are further decomposed or combined, for example, the modules in the embodiments may be combined into one module, or may be further split into multiple sub-modules, so as to complete all or part of the above described functions. The names of the modules and steps involved in the embodiments of the present invention are only for distinguishing the modules or steps, and are not to be construed as unduly limiting the present invention.
A storage device according to a third embodiment of the present invention stores a plurality of programs, and the programs are adapted to be loaded by a processor and to implement a method for three-dimensional reconstruction imaging of magnetic particles based on a generative countermeasure network as described above.
A processing apparatus according to a fourth embodiment of the present invention includes a processor, a storage device; a processor adapted to execute various programs; a storage device adapted to store a plurality of programs; the program is suitable to be loaded and executed by a processor to implement a magnetic particle three-dimensional reconstruction imaging method based on generation of a countermeasure network as described above.
It can be clearly understood by those skilled in the art that, for convenience and brevity of description, the specific working processes and related descriptions of the storage device and the processing device described above may refer to the corresponding processes in the foregoing method examples, and are not described herein again.
Those of skill in the art would appreciate that the various illustrative modules, method steps, and modules described in connection with the embodiments disclosed herein may be implemented as electronic hardware, computer software, or combinations of both, and that programs corresponding to the software modules, method steps may be located in Random Access Memory (RAM), memory, read Only Memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art. To clearly illustrate this interchangeability of electronic hardware and software, various illustrative components and steps have been described above generally in terms of their functionality. Whether these functions are performed in electronic hardware or software depends upon the particular application and design constraints imposed on the solution. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present invention.
The terms "first," "second," "third," and the like are used for distinguishing between similar elements and not necessarily for describing or implying a particular order or sequence.
So far, the technical solutions of the present invention have been described in connection with the preferred embodiments shown in the drawings, but it is easily understood by those skilled in the art that the scope of the present invention is obviously not limited to these specific embodiments. Equivalent changes or substitutions of related technical features can be made by those skilled in the art without departing from the principle of the invention, and the technical scheme after the changes or substitutions can fall into the protection scope of the invention.
Claims (10)
1. A magnetic particle three-dimensional reconstruction imaging method based on generation of a countermeasure network is characterized by comprising the following steps:
s100, acquiring a multi-view sparse two-dimensional MPI image of a simulated body of a to-be-imaged reconstructed object, and performing down-sampling to obtain a down-sampled sparse two-dimensional MPI image; the multi-view sparse two-dimensional MPI image is a plurality of MPI projection images which are collected in sequence according to a set angle rotation;
step S200, inputting the downsampled sparse two-dimensional MPI image into a trained neighborhood point average diffusion search self-attention generation confrontation network generation model to obtain a dense two-dimensional MPI image;
s300, reconstructing each dense two-dimensional MPI image through a filtering back projection reconstruction algorithm to finally obtain a three-dimensional MPI image of a to-be-imaged reconstructed object;
the neighborhood point average diffusion search self-attention generation antagonistic network comprises a generation model and a discrimination model;
the generation model comprises five diffusion search attention mechanism convolution modules; the diffusion search attention mechanism convolution module is constructed on the basis of a neighborhood point average diffusion convolution sub-network, a self-attention fusion network and an activation function layer which are sequentially connected; the activation function adopted by the activation function layers in the first four diffusion search attention mechanism convolution modules is Leaky ReLU, and the activation function adopted by the activation function layers in the fifth diffusion search attention mechanism convolution module is hyperbolic tangent function;
the neighborhood point average diffusion convolution sub-network is configured to sequentially encode the down-sampled sparse two-dimensional MPI image through an encoding function to obtain a sequentially encoded image; performing regional convolution operation on each sequential coded image to acquire particle concentration information, angle information and correlation degree information among the images of each sequential coded image;
the self-attention fusion network is configured to multiply particle concentration information, angle information and correlation degree information among images of each sequence coding image with corresponding weight matrixes to perform linear transformation, perform fusion through a multi-head attention mechanism, and perform up-sampling after fusion to obtain a dense two-dimensional MPI image;
the discrimination model comprises five convolution neural network modules; the convolutional neural network module is constructed on the basis of a convolutional layer, a regularization layer and an activation function layer which are connected in sequence; the activation function adopted by the activation function layers of the first four convolutional neural network modules is Leaky ReLU, and the activation function adopted by the activation function layer of the fifth convolutional neural network module is softmax.
2. The magnetic particle three-dimensional reconstruction imaging method based on generation of the confrontation network as claimed in claim 1, wherein the training process of the neighborhood point average diffusion search self-attention generation confrontation network is as follows:
a100, acquiring a multi-view sparse two-dimensional MPI image of a simulated body of a to-be-imaged reconstructed object through MPI imaging equipment, and performing down-sampling to obtain a down-sampled sparse two-dimensional MPI image; obtaining a training data set based on the down-sampled sparse two-dimensional MPI image and a corresponding truth label;
step A200, inputting the down-sampled sparse two-dimensional MPI image into a pre-constructed neighborhood point average diffusion search self-attention generation confrontation network generation model to obtain a dense two-dimensional MPI image;
step A300, inputting the intensive two-dimensional MPI image and the corresponding truth label into the neighborhood point average diffusion search self-attention to generate a discrimination model of a countermeasure network, and acquiring a discrimination result of the intensive two-dimensional MPI image;
step A400, based on the discrimination result, combining each dense two-dimensional MPI image, each sparse two-dimensional MPI image and the corresponding truth value label thereof, calculating a total loss value through a pre-constructed loss function, and updating the network parameters of the generation model and the discrimination model of the self-attention generation countermeasure network of the neighborhood point average diffusion search;
step A500, training the generator network and the discriminator network of the neighborhood point average diffusion search self-attention generation antagonistic network in a circulating manner until the trained neighborhood point average diffusion search self-attention generation antagonistic network is obtained.
4. The magnetic particle three-dimensional reconstruction imaging method based on the generation countermeasure network of claim 1, wherein the sequential coding image is subjected to a region convolution operation, and the method comprises the following steps: processing the density value of the pixel point in the sequential coding image, and calculating by adopting a3 multiplied by 3 convolution kernel after processing;
the method for processing the density value of the pixel point in the sequential coding image comprises the following steps:
wherein ,indicating a ^ th or greater in the sequentially encoded picture>Line is on the fifth or fifth side>Pixel point of the column->Representing a neighborhood of pointsAverage of all values in>Represents->Minimum of all values in the neighborhood of points, </or >>RepresentMaximum of all values in a point neighborhood>Represents->Values of variables in the neighborhood of points.
5. The method for generating magnetic particle three-dimensional reconstruction imaging based on the countermeasure network as claimed in claim 1, wherein the particle concentration information, angle information and correlation degree information between images of each sequential coding image are multiplied by the corresponding weight matrix for linear transformation, and the method comprises:
wherein ,represents a self-attention mechanism, vector->Represents the particle concentration information, is>(ii) a Vector->Represents angle information, and->(ii) a Vector->Information indicating the degree of association between the images,,/>means for transforming the matrix into a one-dimensional column vector, vector @>A sequentially encoded image representing an input neighborhood point average diffusion convolution sub-network; />、/>、/>Respectively representing the weight matrixes corresponding to the particle concentration information, the angle information and the correlation degree information among the images of the sequential coding image, and then>Indicates that each sequentially encoded picture pick>Neurons in all output layers->The sum of the output results from 1 to N>Representing sequentially coded pictures>Past the second decision->The output value of each neuron is greater than or equal to>Indicates that each sequentially encoded picture pick>Past a first>The output value of each of the neurons is,indicates that each sequentially encoded picture pick>Past a first>The output value of each neuron accounts for the total output value->Specific gravity of (a).
6. The method for generating magnetic particle three-dimensional reconstruction imaging based on confrontation network according to claim 5, characterized in that the particle concentration information, angle information and correlation information between images of each sequential coding image are fused by a multi-head attention mechanism, and the method comprises:
wherein ,is a splicing function->Indicates the fifth->An attention device controls the output of the submodule, < > or >>A total weight matrix representing the output of the multi-head attention device>Information obtained by fusing particle density information, angle information, and inter-image correlation information of sequentially encoded images.
7. The magnetic particle three-dimensional reconstruction imaging method based on the generation countermeasure network as claimed in claim 2, characterized in that the total loss value is calculated by a pre-constructed loss function, and the method is as follows:
wherein ,in a total loss function>A function representing pixel loss and a penalty function, based on a comparison of the pixel loss and the penalty function>Represents an angular sequential order optimization function, < > or >>Represents a pixel information error, and->Represents a loss of the generative model>Represents a sparse two-dimensional MPI image, <' >>Indicates a generated fifth>Opened and closed two-dimensional MPI image->Indicates the fifth occurrence>The two-dimensional MPI image is stretched and compacted,indicates the fifth->Based on a truth label corresponding to the sparse two-dimensional MPI image>Coefficient representing a pixel loss>Represents a factor counteracting the loss>Representing the probability of a decision by the decision network being true or not, based on the presence or absence of a predetermined condition>The output of the generative model is represented,represents an angle difference function>Is a hyperbolic tangent function->A predictor value representing an angle at the corresponding input, <' > or>,/>Representing any two dense two-dimensional MPI images->The function of the judgment of the angle, and />Respectively denote a fifth->Angle labels corresponding to the densely generated two-dimensional MPI images>Parameters representing richness settings for different mockups containing information.
8. A magnetic particle three-dimensional reconstruction imaging system based on generation of a countermeasure network, the system comprising: the image reconstruction system comprises an image acquisition module 100, an image reconstruction module 200 and a three-dimensional reconstruction module 300;
the image acquisition module 100 is configured to acquire a multi-view sparse two-dimensional MPI image of a phantom of an object to be imaged and reconstructed, and perform downsampling to obtain a downsampled sparse two-dimensional MPI image; the multi-view sparse two-dimensional MPI image is a plurality of MPI projection images which are collected in sequence according to a set angle rotation;
the image reconstruction module 200 is configured to input the downsampled sparse two-dimensional MPI image into a trained generation model of a neighborhood point average diffusion search self-attention generation countermeasure network to obtain a dense two-dimensional MPI image;
the three-dimensional reconstruction module 300 is configured to reconstruct each dense two-dimensional MPI image through a filtering back projection reconstruction algorithm, and finally obtain a three-dimensional MPI image of a reconstructed object to be imaged;
the neighborhood point average diffusion search self-attention generation countermeasure network comprises a generation model and a discrimination model;
the generation model comprises five diffusion search attention mechanism convolution modules; the diffusion search attention mechanism convolution module is constructed on the basis of a neighborhood point average diffusion convolution sub-network, a self-attention fusion network and an activation function layer which are sequentially connected; the activation function adopted by the activation function layer in the first four diffusion search attention machine convolution modules is Leaky ReLU, and the activation function adopted by the activation function layer in the fifth diffusion search attention machine convolution module is hyperbolic tangent function;
the neighborhood point average diffusion convolution sub-network is configured to sequentially encode the downsampled sparse two-dimensional MPI image through an encoding function to obtain a sequentially encoded image; performing regional convolution operation on each sequential coded image to acquire particle concentration information, angle information and correlation degree information among the images of each sequential coded image;
the self-attention fusion network is configured to multiply particle concentration information, angle information and correlation degree information among images of each sequence coding image with corresponding weight matrixes to perform linear transformation, perform fusion through a multi-head attention mechanism, and perform up-sampling after fusion to obtain a dense two-dimensional MPI image;
the discrimination model comprises five convolutional neural network modules; the convolutional neural network module is constructed on the basis of a convolutional layer, a regularization layer and an activation function layer which are connected in sequence; the activation function adopted by the activation function layers of the first four convolutional neural network modules is Leaky ReLU, and the activation function adopted by the activation function layer of the fifth convolutional neural network module is softmax.
9. A storage device, in which a plurality of programs are stored, wherein the programs are adapted to be loaded and executed by a processor to implement a method for three-dimensional reconstruction imaging of magnetic particles based on generation of a countermeasure network according to any of claims 1 to 7.
10. A processing device comprising a processor, a storage device; a processor adapted to execute various programs; a storage device adapted to store a plurality of programs; characterized in that the program is adapted to be loaded and executed by a processor to implement a method of magnetic particle three-dimensional reconstruction imaging based on generation of a countermeasure network as claimed in any of claims 1-7.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310047876.0A CN115880440B (en) | 2023-01-31 | 2023-01-31 | Magnetic particle three-dimensional reconstruction imaging method based on generation countermeasure network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310047876.0A CN115880440B (en) | 2023-01-31 | 2023-01-31 | Magnetic particle three-dimensional reconstruction imaging method based on generation countermeasure network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN115880440A true CN115880440A (en) | 2023-03-31 |
CN115880440B CN115880440B (en) | 2023-04-28 |
Family
ID=85758588
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202310047876.0A Active CN115880440B (en) | 2023-01-31 | 2023-01-31 | Magnetic particle three-dimensional reconstruction imaging method based on generation countermeasure network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN115880440B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116563412A (en) * | 2023-06-26 | 2023-08-08 | 中国科学院自动化研究所 | MPI image reconstruction method, system and equipment based on sparse system matrix |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20210015395A1 (en) * | 2019-07-15 | 2021-01-21 | Institute Of Automation, Chinese Academy Of Sciences | Ffl-based magnetic particle imaging three-dimensional reconstruction method, system, and device |
WO2021017372A1 (en) * | 2019-08-01 | 2021-02-04 | 中国科学院深圳先进技术研究院 | Medical image segmentation method and system based on generative adversarial network, and electronic equipment |
CN113808234A (en) * | 2021-11-08 | 2021-12-17 | 北京航空航天大学 | Rapid magnetic particle imaging reconstruction method based on undersampling |
CN113850883A (en) * | 2021-10-14 | 2021-12-28 | 北京航空航天大学 | Magnetic particle imaging reconstruction method based on attention mechanism |
CN114792287A (en) * | 2022-03-25 | 2022-07-26 | 南京航空航天大学 | Medical ultrasonic image super-resolution reconstruction method based on multi-image fusion |
CN115409945A (en) * | 2022-09-01 | 2022-11-29 | 中国科学院自动化研究所 | Three-dimensional magnetic particle imaging system and method for carrying out quantitative analysis by fusing imaging parameters |
CN115526946A (en) * | 2022-10-13 | 2022-12-27 | 中国科学院自动化研究所 | Method, system and equipment for denoising magnetic particle imaging image based on feature fusion |
-
2023
- 2023-01-31 CN CN202310047876.0A patent/CN115880440B/en active Active
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20210015395A1 (en) * | 2019-07-15 | 2021-01-21 | Institute Of Automation, Chinese Academy Of Sciences | Ffl-based magnetic particle imaging three-dimensional reconstruction method, system, and device |
WO2021017372A1 (en) * | 2019-08-01 | 2021-02-04 | 中国科学院深圳先进技术研究院 | Medical image segmentation method and system based on generative adversarial network, and electronic equipment |
CN113850883A (en) * | 2021-10-14 | 2021-12-28 | 北京航空航天大学 | Magnetic particle imaging reconstruction method based on attention mechanism |
CN113808234A (en) * | 2021-11-08 | 2021-12-17 | 北京航空航天大学 | Rapid magnetic particle imaging reconstruction method based on undersampling |
CN114792287A (en) * | 2022-03-25 | 2022-07-26 | 南京航空航天大学 | Medical ultrasonic image super-resolution reconstruction method based on multi-image fusion |
CN115409945A (en) * | 2022-09-01 | 2022-11-29 | 中国科学院自动化研究所 | Three-dimensional magnetic particle imaging system and method for carrying out quantitative analysis by fusing imaging parameters |
CN115526946A (en) * | 2022-10-13 | 2022-12-27 | 中国科学院自动化研究所 | Method, system and equipment for denoising magnetic particle imaging image based on feature fusion |
Non-Patent Citations (2)
Title |
---|
JIWON KIM 等: "Accurate Image Super-Resolution Using Very Deep Convolutional Networks", IEEE:COMPUTE .\\R SOCIETY * |
查体博 等: "基于改进超分辨率生成对抗网络的图像重建算法", 基于改进超分辨率生成对抗网络的图像重建算法 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116563412A (en) * | 2023-06-26 | 2023-08-08 | 中国科学院自动化研究所 | MPI image reconstruction method, system and equipment based on sparse system matrix |
CN116563412B (en) * | 2023-06-26 | 2023-10-20 | 中国科学院自动化研究所 | MPI image reconstruction method, system and equipment based on sparse system matrix |
Also Published As
Publication number | Publication date |
---|---|
CN115880440B (en) | 2023-04-28 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Shen et al. | Patient-specific reconstruction of volumetric computed tomography images from a single projection view via deep learning | |
Liu et al. | TomoGAN: low-dose synchrotron x-ray tomography with generative adversarial networks: discussion | |
Anirudh et al. | Lose the views: Limited angle CT reconstruction via implicit sinogram completion | |
Pan et al. | Multi-domain integrative Swin transformer network for sparse-view tomographic reconstruction | |
CN109300166A (en) | The method and apparatus and storage medium of CT image reconstruction | |
Topal et al. | Deep learning-based inaccuracy compensation in reconstruction of high resolution XCT data | |
CN114170146A (en) | Image processing method, image processing device, electronic equipment and computer readable storage medium | |
CN115880440A (en) | Magnetic particle three-dimensional reconstruction imaging method based on generation of countermeasure network | |
Lahiri et al. | Sparse-view cone beam CT reconstruction using data-consistent supervised and adversarial learning from scarce training data | |
Liu et al. | Tomogan: Low-dose x-ray tomography with generative adversarial networks | |
Xia et al. | Deep residual neural network based image enhancement algorithm for low dose CT images | |
CN114612318A (en) | Three-dimensional modeling method, system and equipment based on cultural relic CT image contour line | |
Ferreira et al. | GAN-based generation of realistic 3D volumetric data: A systematic review and taxonomy | |
Elss et al. | Deep-learning-based CT motion artifact recognition in coronary arteries | |
CN116503506B (en) | Image reconstruction method, system, device and storage medium | |
WO2021184195A1 (en) | Medical image reconstruction method, and medical image reconstruction network training method and apparatus | |
US11933870B2 (en) | Contrast and/or system independent motion detection for magnetic resonance imaging | |
Wang et al. | Osnet & mneto: Two types of general reconstruction architectures for linear computed tomography in multi-scenarios | |
CN114862982A (en) | Hybrid domain unsupervised finite angle CT reconstruction method based on generation countermeasure network | |
Guo et al. | Iterative Image Reconstruction for Limited‐Angle CT Using Optimized Initial Image | |
Sabnam et al. | Application of generative adversarial networks in image, face reconstruction and medical imaging: challenges and the current progress | |
Guo et al. | Noise-resilient deep learning for integrated circuit tomography | |
Cheng et al. | Super-resolution reconstruction for parallel-beam SPECT based on deep learning and transfer learning: a preliminary simulation study | |
Han et al. | BiRPN-YOLOvX: A weighted bidirectional recursive feature pyramid algorithm for lung nodule detection | |
Wang et al. | Optimization algorithm of CT image edge segmentation using improved convolution neural network |
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 | ||
GR01 | Patent grant |