CN116228905A - CT imaging method and system for cloud image center with multiple imaging geometries - Google Patents
CT imaging method and system for cloud image center with multiple imaging geometries Download PDFInfo
- Publication number
- CN116228905A CN116228905A CN202310098648.6A CN202310098648A CN116228905A CN 116228905 A CN116228905 A CN 116228905A CN 202310098648 A CN202310098648 A CN 202310098648A CN 116228905 A CN116228905 A CN 116228905A
- Authority
- CN
- China
- Prior art keywords
- imaging
- module
- network
- domain filtering
- image
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000003384 imaging method Methods 0.000 title claims abstract description 129
- 238000000034 method Methods 0.000 title claims abstract description 51
- 238000013170 computed tomography imaging Methods 0.000 title claims abstract description 39
- 238000002591 computed tomography Methods 0.000 claims abstract description 49
- 238000001914 filtration Methods 0.000 claims description 74
- 238000012549 training Methods 0.000 claims description 32
- 239000013598 vector Substances 0.000 claims description 24
- 238000005070 sampling Methods 0.000 claims description 17
- 238000013507 mapping Methods 0.000 claims description 14
- 230000001133 acceleration Effects 0.000 claims description 5
- 238000012545 processing Methods 0.000 claims description 5
- 238000004088 simulation Methods 0.000 claims description 5
- 238000010606 normalization Methods 0.000 claims description 4
- 238000007781 pre-processing Methods 0.000 claims description 4
- 230000006798 recombination Effects 0.000 claims description 4
- 238000005215 recombination Methods 0.000 claims description 4
- 229910052704 radon Inorganic materials 0.000 claims description 3
- SYUHGPGVQRZVTB-UHFFFAOYSA-N radon atom Chemical compound [Rn] SYUHGPGVQRZVTB-UHFFFAOYSA-N 0.000 claims description 3
- 230000006870 function Effects 0.000 description 10
- 230000008569 process Effects 0.000 description 9
- 238000013135 deep learning Methods 0.000 description 7
- 238000013528 artificial neural network Methods 0.000 description 4
- 238000012360 testing method Methods 0.000 description 4
- 238000011161 development Methods 0.000 description 3
- 210000001015 abdomen Anatomy 0.000 description 2
- 230000005540 biological transmission Effects 0.000 description 2
- 238000004891 communication Methods 0.000 description 2
- 230000007812 deficiency Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 206010063385 Intellectualisation Diseases 0.000 description 1
- 241000237502 Ostreidae Species 0.000 description 1
- 238000009825 accumulation Methods 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 210000003484 anatomy Anatomy 0.000 description 1
- 230000002238 attenuated effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 238000003759 clinical diagnosis Methods 0.000 description 1
- 239000012141 concentrate Substances 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000005684 electric field Effects 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 235000020636 oyster Nutrition 0.000 description 1
- 238000012805 post-processing Methods 0.000 description 1
- 230000005855 radiation Effects 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B6/00—Apparatus or devices for radiation diagnosis; Apparatus or devices for radiation diagnosis combined with radiation therapy equipment
- A61B6/02—Arrangements for diagnosis sequentially in different planes; Stereoscopic radiation diagnosis
- A61B6/03—Computed tomography [CT]
- A61B6/032—Transmission computed tomography [CT]
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B6/00—Apparatus or devices for radiation diagnosis; Apparatus or devices for radiation diagnosis combined with radiation therapy equipment
- A61B6/52—Devices using data or image processing specially adapted for radiation diagnosis
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B6/00—Apparatus or devices for radiation diagnosis; Apparatus or devices for radiation diagnosis combined with radiation therapy equipment
- A61B6/52—Devices using data or image processing specially adapted for radiation diagnosis
- A61B6/5211—Devices using data or image processing specially adapted for radiation diagnosis involving processing of medical diagnostic data
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T11/00—2D [Two Dimensional] image generation
- G06T11/003—Reconstruction from projections, e.g. tomography
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Physics & Mathematics (AREA)
- Medical Informatics (AREA)
- Biomedical Technology (AREA)
- General Health & Medical Sciences (AREA)
- Theoretical Computer Science (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Surgery (AREA)
- Veterinary Medicine (AREA)
- Radiology & Medical Imaging (AREA)
- Optics & Photonics (AREA)
- Heart & Thoracic Surgery (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- High Energy & Nuclear Physics (AREA)
- Animal Behavior & Ethology (AREA)
- Pathology (AREA)
- Public Health (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Pulmonology (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Apparatus For Radiation Diagnosis (AREA)
Abstract
A CT imaging method and system for a cloud image center of multiple imaging geometries, wherein the CT imaging method for a cloud image center of multiple imaging geometries comprises 4 steps. The CT imaging method comprises the steps of constructing an imaging geometric modulation sub-network and an intelligent reconstruction sub-network to obtain an optimized intelligent imaging network, inputting projection data acquired by a CT scanning device into the optimized intelligent imaging network, and finally obtaining an optimized CT image. The invention can be suitable for reconstructing CT images with different imaging geometries and dosages, and obtaining high-quality reconstructed images, thereby greatly improving the flexibility.
Description
Technical Field
The invention relates to the technical field of deep learning and computed tomography, in particular to a CT imaging method for a cloud image center with various imaging geometries and a CT imaging system for the cloud image center with various imaging geometries.
Background
Computed Tomography (CT) has a great advantage in terms of not causing invasive damage when reconstructing the anatomy in the patient's body, low-dose CT imaging has been widely used in clinical diagnosis because patients suffer from potential health risks from radiation during CT scanning.
In recent years, the rapid development of deep learning technology has revealed a broad prospect for medical industry intellectualization. In light of this, some CT reconstruction methods based on deep learning are proposed and gradually become the mainstream. However, the current reconstruction methods based on the deep learning network have the limitation that training can only be performed by using a data set with specific imaging geometry and dose level, and cannot process multiple imaging geometry data of different devices at the same time, which reduces the flexibility of the deep learning reconstruction method.
Recently, along with development of big data and cloud computing, a cloud image center is becoming a trend of future medical image technology development, and the cloud image center generally concentrates detection data of different CT image devices, so that in order to reconstruct a CT image efficiently, a traditional deep learning CT reconstruction method based on single imaging geometry is not applicable any more.
Therefore, in order to solve the deficiencies of the prior art, it is necessary to provide a CT imaging method for a cloud image center with multiple imaging geometries and a CT imaging system for a cloud image center with multiple imaging geometries.
Disclosure of Invention
One of the objects of the present invention is to provide a CT imaging method for a cloud image center of multiple imaging geometries, which avoids the disadvantages of the prior art. The CT imaging method for the cloud image center with various imaging geometries can be suitable for reconstructing and imaging CT images with different imaging geometries and dosages, and can obtain high-quality reconstructed images, so that the flexibility can be greatly improved.
The above object of the present invention is achieved by the following technical measures:
provided is a CT imaging method for a cloud image center of various imaging geometries, comprising the following steps:
the method comprises the steps of (1) constructing an imaging geometric modulation sub-network for generating modulation feature vectors and an intelligent reconstruction sub-network capable of dynamically adjusting filtering according to the modulation feature vectors, and then embedding the imaging geometric modulation sub-network into the intelligent reconstruction sub-network, wherein the imaging geometric modulation sub-network comprises a geometric mapping module and an imaging geometric modulation module, and the intelligent reconstruction sub-network is provided with a chord domain filtering module, an image domain filtering module and a back projection module;
step (2), carrying out downsampling acceleration and upsampling recombination to obtain a preliminary intelligent imaging network;
training the preliminary intelligent imaging network obtained in the step (2) to obtain an optimized intelligent imaging network;
and (4) inputting projection data acquired by the CT scanning device into an optimized intelligent imaging network to obtain an optimized CT image.
Preferably, the step (1) specifically includes:
step (1.1), constructing a geometric mapping module by using a multi-layer perceptron with a plurality of fully connected layers, and constructing two geometric mapping modules altogether, wherein the geometric mapping module maps vector omega input to higher dimensionality to obtain a high-dimensional modulation feature vector beta and a high-dimensional modulation feature vector gamma, and the vector omega is constructed by all imaging geometries and doses;
step (1.2), respectively constructing a chord domain filtering module, an image domain filtering module and a back projection module, wherein the back projection module is positioned between the last convolution layer of the chord domain filtering module and the first convolution layer of the image domain filtering module;
step (1.3), constructing an imaging geometric modulation module by using a plurality of multi-layer perceptrons, and constructing two imaging geometric modulation modules altogether, wherein the imaging geometric modulation module modulates a high-dimensional modulation eigenvector beta and a high-dimensional modulation eigenvector gamma to obtain an eigenvector graph
As the chord domain filtering module and the image domain filteringInput of the module, said->
Represented by the formula (I),
wherein fl And outputting a characteristic map for the convolution layer of the chord domain filtering module and the convolution layer of the image domain filtering module.
Preferably, the chord domain filtering module and the image domain filtering module are respectively constructed by a convolution residual error network.
Preferably, the back projection module performs fourier transform, inverse fourier transform, interpolation and inverse radon transform processing on the chord-graph data of the chord-graph domain filtering module, and then maps the processed chord-graph data to the image domain filtering module.
Preferably, the step (2) specifically includes:
step (2.1), constructing two downsampling modules, wherein one downsampling module is deployed behind a first convolution layer of the chord domain filtering module, the other downsampling module is deployed behind the first convolution layer of the image domain filtering module, and the downsampling module reduces input data by taking a scale factor as alpha, wherein alpha is more than or equal to 2 and less than or equal to 16;
and (2.2) constructing two up-sampling modules, wherein one up-sampling module is deployed before the last convolution layer of the chord domain filtering module, the other up-sampling module is deployed before the last convolution layer of the image domain filtering module, and the up-sampling module amplifies input data according to a scale factor alpha to obtain a preliminary intelligent imaging network.
Preferably, the above-mentioned Ω is represented by formula (II),
Ω(N s ,N d ,N i ,D v ,D d ,DSD,DSO,I 0 sigma) … … formula (II),
wherein ,Ns Projection view angle number, N, of CT scan d For the number of CT detector units, N i To reconstruct the number of pixels of the image, D v To reconstruct the pixel size of the image, D d For the size of the CT detector unit, DSD is the distance from the source to the detector, DSO is the distance from the source to the center of rotation, I 0 For the amount of incident photons, σ is the standard deviation of Gaussian noise.
Preferably, the step (3) specifically includes:
training the preliminary intelligent imaging network obtained in the step (2) by using normal dose data to obtain a pre-training model;
and (3.2) continuing training the pre-training model obtained in the step (3.1) by using low-dose data so as to finely tune and optimize parameters of the intelligent imaging network and obtain the optimized intelligent imaging network.
Preferably, in the step (3.1), L is used 2 Training the preliminary intelligent imaging network by using a loss function, L 2 The loss function is represented by formula (III):
wherein ,
is L 2 Loss function constraint of paradigm, μ * For reconstructing an image, μ is a reference image.
Preferably, the step (4) is specifically to perform linear preprocessing on projection data acquired by the CT scanning device, and then input the processed projection data into the optimized intelligent imaging network obtained in the step (3) to obtain a filtered and reconstructed optimized CT image.
Preferably, the linear pretreatment is normalization.
Preferably, the low dose data is dose data obtained by scanning under a low dose scanning protocol of low tube voltage or low tube current or low dose data obtained by simulation.
Another object of the present invention is to provide a cloud image center CT imaging system with multiple imaging geometries that avoids the deficiencies of the prior art. The cloud image center CT imaging system with various imaging geometries can be suitable for oyster reconstruction imaging of CT images with different imaging geometries and dosages and can obtain high-quality reconstructed images.
The above object of the present invention is achieved by the following technical measures:
a CT imaging system for a cloud image center of a plurality of imaging geometries is provided, and the CT imaging method for the cloud image center of the plurality of imaging geometries is adopted.
The invention discloses a CT imaging method and a CT imaging system for a cloud image center with multiple imaging geometries, wherein the CT imaging method for the cloud image center with multiple imaging geometries comprises the following steps: the method comprises the steps of (1) constructing an imaging geometric modulation sub-network for generating modulation feature vectors and an intelligent reconstruction sub-network capable of dynamically adjusting filtering according to the modulation feature vectors, and then embedding the imaging geometric modulation sub-network into the intelligent reconstruction sub-network, wherein the imaging geometric modulation sub-network comprises a geometric mapping module and an imaging geometric modulation module, and the intelligent reconstruction sub-network is provided with a chord domain filtering module, an image domain filtering module and a back projection module; step (2), carrying out downsampling acceleration and upsampling recombination to obtain a preliminary intelligent imaging network; training the preliminary intelligent imaging network obtained in the step (2) to obtain an optimized intelligent imaging network; and (4) inputting projection data acquired by the CT scanning device into an optimized intelligent imaging network to obtain an optimized CT image. The invention can be suitable for reconstructing CT images with different imaging geometries and dosages, and obtaining high-quality reconstructed images, thereby greatly improving the flexibility.
Drawings
The invention is further illustrated by the accompanying drawings, which are not to be construed as limiting the invention in any way.
FIG. 1 is a block diagram of an optimized intelligent imaging network of the present invention.
Fig. 2 is the reconstruction result of example 2.
Detailed Description
The technical scheme of the invention is further described with reference to the following examples.
Example 1
A CT imaging method for a cloud image center of multiple imaging geometries, comprising the steps of:
the method comprises the steps of (1) constructing an imaging geometric modulation sub-network for generating modulation feature vectors and an intelligent reconstruction sub-network capable of dynamically adjusting filtering according to the modulation feature vectors, and then embedding the imaging geometric modulation sub-network into the intelligent reconstruction sub-network, wherein the imaging geometric modulation sub-network comprises a geometric mapping module and an imaging geometric modulation module, and the intelligent reconstruction sub-network is provided with a chord domain filtering module, an image domain filtering module and a back projection module;
step (2), carrying out downsampling acceleration and upsampling recombination to obtain a preliminary intelligent imaging network;
step (3), training the preliminary intelligent imaging network obtained in the step (2), and obtaining an optimized intelligent imaging network, as shown in fig. 1;
and (4) inputting projection data acquired by the CT scanning device into an optimized intelligent imaging network to obtain an optimized CT image.
Wherein, step (1) specifically includes:
step (1.1), constructing a geometric mapping module by using a multi-layer perceptron with a plurality of full-connection layers, constructing two geometric mapping modules altogether, wherein the geometric mapping module maps vector omega input to higher dimensionality to obtain a high-dimensional modulation feature vector beta and a high-dimensional modulation feature vector gamma, and the vector omega is constructed by all imaging geometries and doses;
step (1.2), respectively constructing a chord domain filtering module, an image domain filtering module and a back projection module, wherein the back projection module is positioned between the last convolution layer of the chord domain filtering module and the first convolution layer of the image domain filtering module;
step (1.3), constructing an imaging geometric modulation module by using a plurality of multi-layer perceptrons, constructing two imaging geometric modulation modules altogether, and modulating the high-dimensional modulation eigenvector beta and the high-dimensional modulation eigenvector gamma by the imaging geometric modulation moduleSign map f l ～ As input to the chord domain filtering module and the image domain filtering module, fl l ～ Represented by the formula (I),
wherein fl The characteristic map is output by a convolution layer of the chord domain filtering module and a convolution layer of the image domain filtering module. The imaging geometric modulation module receives the high-dimensional modulation vector from the geometric mapping module and the intermediate feature map from the chord domain and image domain filtering module, and performs the above processing after two parts of beta and gamma with the same channel number of the high-dimensional modulation vector and the feature map.
The chord domain filtering module and the image domain filtering module are respectively constructed by a convolution residual error network. It should be noted that, the chord domain filtering module and the image domain filtering module both include a series of convolution layers, a regularization layer and an activation function layer.
The chord chart domain filtering module specifically processes input chord chart data (namely projection data acquired in the step (4)) by using a convolution residual error network, and the chord chart data passes through the chord chart domain filtering module and outputs a filtered chord chart. The chord domain filtering module plays a role similar to the traditional reconstruction filtering by using a neural network in the reconstruction process.
The image domain filtering module specifically processes the image domain data obtained by back projection by using a convolution residual error network, and the image domain data passes through the module to output a final reconstructed image. The image domain filtering module corresponds to the post-processing operation of the image in the conventional reconstruction method.
The back projection module performs fourier transform, inverse fourier transform, interpolation and inverse radon transform processing on the chord-graph data of the chord-graph domain filtering module, and then maps the processed chord-graph data to the image domain filtering module. It should be noted that, the back projection module performs a series of operations on the chord-graph data, and back projects the chord-graph data to the image domain filtering module, where the back projection module does not train network parameters, but performs gradient back transmission in the neural network training process, specifically, in the gradient back transmission process, the back projection module performs front projection operation on the gradient returned by the image domain filtering module, and uses the front projection operation as the gradient returned to the chord domain filtering module.
The step (2) of the invention specifically comprises the following steps:
step (2.1), constructing two downsampling modules, wherein one downsampling module is deployed behind a first convolution layer of a chord domain filtering module, the other downsampling module is deployed behind the first convolution layer of an image domain filtering module, and the downsampling module reduces input data by taking a scale factor as alpha, wherein alpha is more than or equal to 2 and less than or equal to 16;
and (2.2) constructing two up-sampling modules, wherein one up-sampling module is deployed before the last convolution layer of the chord domain filtering module, the other up-sampling module is deployed before the last convolution layer of the image domain filtering module, and the up-sampling module amplifies input data according to a scaling factor alpha to obtain the preliminary intelligent imaging network.
It should be noted that the downsampling module may scale the input data by a scale factor of 2 to 16, thereby speeding up the overall network training speed, and adaptively scale the features by the scale factor according to the size of the input data without losing any detail. The downsampling module generates different sampling indexes by different sampling rates, and the sampling indexes sample data at equal intervals.
The present embodiment describes the processing procedure of the downsampling module with a scale factor of 2, and the downsampling module reshapes a feature map of size c×w×h into a feature map of size c×w×h by recombining pixels in the original feature map without losing any detail information
Is described.
The up-sampling module amplifies the input data according to a scale factor of 2 to 16, and recoversData detail, the downsampled data is recombined into pixels in the original feature map without losing any detail information. Similarly, describing the upsampling module process with a scale factor of 2, the pixels in the original feature map are resized to a size of
Is reshaped into an original signature of size c×w×h.
Before the back projection operation is executed, an up-sampling module is added at the tail end of a chord domain filtering module to restore the chord domain data to the normal size, and then the back projection operation is executed; and the normal chord domain data is used for back projection operation, so that error accumulation caused by downsampling is avoided, and meanwhile, a downsampling module is added in front of an image domain filtering module so as to finish subsequent acceleration.
The step (3) of the invention specifically comprises the following steps:
training the preliminary intelligent imaging network obtained in the step (2) by using normal dose data to obtain a pre-training model;
step (3.2), training the pre-training model obtained in the step (3.1) continuously by using low-dose data so as to finely adjust and optimize parameters of the intelligent imaging network, and obtaining an optimized intelligent imaging network; the low dose data of the invention is dose data obtained by scanning under a low dose scanning protocol of low tube voltage or low tube current or low dose data obtained by simulation.
Wherein in step (3.1), L is used 2 Training a preliminary intelligent imaging network by using a loss function, L 2 The loss function is represented by formula (III):
wherein ,
is L 2 Loss function constraint of paradigm, μ * For reconstructing an image, μ is a reference image. The pre-training of the invention using normal dose data allows the network to have good rebuilding performance.
The training stop determination method in step (3.1) may be as follows: 1. in the training process of the neural network, whether the error calculated by the loss function is stable or not and converges to a smaller value is not the important point of the invention, and the size of the value can be determined by a person skilled in the art according to the actual situation. 2. Through test data test, high-quality CT images (better results in visual analysis and quantitative indexes) can be reconstructed from chord chart data of different geometries, and when one of the two conditions is reached, training is stopped and a pre-training model is obtained.
The step (4) of the invention is to perform linear preprocessing on projection data acquired by a CT scanning device, and then input the projection data into the optimized intelligent imaging network obtained in the step (3) to obtain a filtered and reconstructed optimized CT image, wherein the linear preprocessing is normalization.
Wherein Ω is represented by formula (II),
Ω(N s ,N d ,N i ,D v ,D d ,DSD,DSO,I 0 sigma) … … formula (II),
wherein ,Ns Projection view angle number, N, of CT scan d For the number of CT detector units, N i To reconstruct the number of pixels of the image, D v To reconstruct the pixel size of the image, D d For the size of the CT detector unit, DSD is the distance from the source to the detector, DSO is the distance from the source to the center of rotation, I 0 For the amount of incident photons, σ is the standard deviation of Gaussian noise.
The imaging geometry and dose of the invention are the number of projection views of CT scanning, the number of CT detector units, the number of pixels of the reconstructed image, the size of the CT detector units, the distance from the ray source to the detector, the distance from the ray source to the rotation center, the incident photon quantity and the standard deviation of Gaussian noise.
The imaging geometry and the dose are known when projection data are acquired. Omega of the present invention is the normalization of all parameters and doses in the imaging geometry and dose, each parameter being a dimension, thus constructed as a vector. For omega and projection data of the invention to be used as the input of the optimized intelligent imaging network, the projection data is input into the intelligent reconstruction sub-network for filtering reconstruction; omega is input into the imaging geometric modulation sub-network and mapped into a high-dimensional vector for modulating a high-dimensional feature map in the reconstruction process of the intelligent imaging sub-network.
The CT imaging method for the cloud image center with various imaging geometries can be suitable for reconstructing and imaging CT images with different imaging geometries and dosages, and can obtain high-quality reconstructed images, so that the flexibility can be greatly improved.
Example 2
An application of a CT imaging method for a cloud image center with multiple imaging geometries.
(1) Data construction
The projection data of this example is 3000 CT images collected from a public dataset, including 1000 head images, 1000 chest images, and 1000 abdomen images.
The example data is 3000 CT images collected from a public dataset, including 1000 head images, 1000 chest images, and 1000 abdomen images. The high/low dose projection data is obtained by performing simulation on CT images of different parts, and the imaging geometry and dose used by the simulation are shown in the following table.
90% of these data are used to train the neural network and 10% are used to test verify optimal intelligent imaging network performance.
(2) Process of implementation
S1, constructing a preliminary intelligent imaging network of the embodiment 1;
s2, pre-training the preliminary intelligent imaging network by using the non-added poisson-Gaussian noise (namely normal dose data), stopping training when the network loss function converges to a smaller value, and obtaining a pre-training model, wherein the value can be determined according to actual conditions.
And S3, fine tuning the pre-training model on the basis of the S2 by using low-dose data, and stopping training when the network loss function is converged to a smaller value again to obtain the optimized intelligent imaging network.
And S4, verifying and optimizing the performance of the intelligent imaging network by using the test data, and calculating a quantitative index of the reconstruction result to obtain the figure 2.
As shown in fig. 2, the result of the embodiment of reconstructing projection data of different imaging geometries and dose levels by using the cloud image center CT imaging method for multiple imaging geometries in the present invention is more excellent in image quality and obvious in noise removal effect compared with the result of the conventional filtered back projection method; the three indexes of peak signal-to-noise ratio (PSNR), structural Similarity Index (SSIM) and Root Mean Square Error (RMSE) are calculated between the demonstration result and the reference image, and the three indexes show that the cloud image center CT imaging method for various imaging geometries is remarkably improved compared with the traditional filtering back projection method.
Example 3
A cloud image center CT imaging system of multiple imaging geometries employs the CT imaging method of embodiment 1 for a cloud image center of multiple imaging geometries.
The system of the invention comprises a CT device, a communication medium, a computer device and a display.
The CT apparatus can emit X-rays to a patient or object to be detected, and the X-rays penetrate a human body and can be collected by a detector.
The communication medium is used to transmit information data including, but not limited to, projection data acquired by the CT scanning device, and program instructions sent by an operator.
Computer devices are used to acquire and reconstruct CT projection data, typically in the form of high performance computing servers, typically including a central processor, a graphics computing unit and may integrate storage devices for storing important data, including raw CT projection data, deep learning technique (e.g., deep learning model) data, reconstructed CT images, and the like.
The display is used to display image data including CT projection data, reconstructed CT images, residual images, and the like.
The CT device is provided with an X-ray bulb tube, an X-ray detector, a high-voltage generator and a rotating frame. The X-ray bulb is used to emit an X-ray beam to scan an object. The X-ray detector is configured to receive an X-ray beam attenuated by the object. The high voltage generator is used for providing a high voltage electric field required by the X-ray bulb. The rotating frame is used for fixing the bulb tube and the detector, and the rotating frame drives the bulb tube and the detector to rotate around the detected object.
The cloud image center CT imaging system with various imaging geometries can be suitable for reconstructing and imaging CT images with different imaging geometries and dosages, and can obtain high-quality reconstructed images, so that the flexibility can be greatly improved.
Finally, it should be noted that the above embodiments are only for illustrating the technical solution of the present invention and not for limiting the scope of the present invention, and although the present invention has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical solution of the present invention may be modified or substituted equally without departing from the spirit and scope of the technical solution of the present invention.
Claims (10)
1. A CT imaging method for a cloud image center of multiple imaging geometries, comprising the steps of:
the method comprises the steps of (1) constructing an imaging geometric modulation sub-network for generating modulation feature vectors and an intelligent reconstruction sub-network capable of dynamically adjusting filtering according to the modulation feature vectors, and then embedding the imaging geometric modulation sub-network into the intelligent reconstruction sub-network, wherein the imaging geometric modulation sub-network comprises a geometric mapping module and an imaging geometric modulation module, and the intelligent reconstruction sub-network is provided with a chord domain filtering module, an image domain filtering module and a back projection module;
step (2), carrying out downsampling acceleration and upsampling recombination to obtain a preliminary intelligent imaging network;
training the preliminary intelligent imaging network obtained in the step (2) to obtain an optimized intelligent imaging network;
and (4) inputting projection data acquired by the CT scanning device into an optimized intelligent imaging network to obtain an optimized CT image.
2. The CT imaging method for a cloud image center of multiple imaging geometries according to claim 1, wherein said step (1) specifically comprises:
step (1.1), constructing a geometric mapping module by using a multi-layer perceptron with a plurality of fully connected layers, and constructing two geometric mapping modules altogether, wherein the geometric mapping module maps vector omega input to higher dimensionality to obtain a high-dimensional modulation feature vector beta and a high-dimensional modulation feature vector gamma, and the vector omega is constructed by all imaging geometries and doses;
step (1.2), respectively constructing a chord domain filtering module, an image domain filtering module and a back projection module, wherein the back projection module is positioned between the last convolution layer of the chord domain filtering module and the first convolution layer of the image domain filtering module;
step (1.3), constructing an imaging geometric modulation module by using a plurality of multi-layer perceptrons, constructing two imaging geometric modulation modules altogether, wherein the imaging geometric modulation module modulates a high-dimensional modulation eigenvector beta and a high-dimensional modulation eigenvector gamma to obtain an eigenvector f l ～ The f is used as the input of the chord domain filtering module and the image domain filtering module l ～ Represented by the formula (I),
f l ～ ＝βf l +γ … … formula (I);
wherein fl Features for and output with the convolution layer of the chord-domain filtering moduleA drawing.
3. The CT imaging method for a cloud image center of multiple imaging geometries of claim 2, wherein: the chord chart domain filtering module and the image domain filtering module are respectively constructed by a convolution residual error network;
the back projection module performs fourier transform, inverse fourier transform, interpolation and inverse radon transform processing on the chord-graph data of the chord-graph domain filtering module, and then maps the processed chord-graph data to the image domain filtering module.
4. The CT imaging method for a cloud image center of multiple imaging geometries according to claim 2, wherein said step (2) specifically comprises:
step (2.1), constructing two downsampling modules, wherein one downsampling module is deployed behind a first convolution layer of the chord domain filtering module, the other downsampling module is deployed behind the first convolution layer of the image domain filtering module, and the downsampling module reduces input data by taking a scale factor as alpha, wherein alpha is more than or equal to 2 and less than or equal to 16;
and (2.2) constructing two up-sampling modules, wherein one up-sampling module is deployed before the last convolution layer of the chord domain filtering module, the other up-sampling module is deployed before the last convolution layer of the image domain filtering module, and the up-sampling module amplifies input data according to a scale factor alpha to obtain a preliminary intelligent imaging network.
5. The CT imaging method for a cloud image center of multiple imaging geometries of claim 2, wherein: the Ω is represented by formula (II),
Ω(N s ，N d ，N i ，D v ，D d ，DSD，DSO，I 0 sigma) … … formula (II),
wherein ,Ns Projection view angle number, N, of CT scan d For the number of CT detector units, N i To reconstruct the number of pixels of the image, D v To reconstruct the pixel size of the image, D d For the size of the CT detector unit, DSD is the distance from the source to the detector, DSO is the distance from the source to the center of rotation, I 0 For the amount of incident photons, σ is the standard deviation of Gaussian noise.
6. The CT imaging method for a cloud image center of multiple imaging geometries as recited in claim 4, wherein said step (3) specifically comprises:
training the preliminary intelligent imaging network obtained in the step (2) by using normal dose data to obtain a pre-training model;
and (3.2) continuing training the pre-training model obtained in the step (3.1) by using low-dose data so as to finely tune and optimize parameters of the intelligent imaging network and obtain the optimized intelligent imaging network.
7. The CT imaging method for a cloud image center of multiple imaging geometries of claim 6, wherein: in the step (3.1), L is used 2 Training the preliminary intelligent imaging network by using a loss function, L 2 The loss function is represented by formula (III):
wherein ,
is L 2 Loss function constraint of paradigm, μ * For reconstructing an image, μ is a reference image.
8. The CT imaging method for a cloud image center of multiple imaging geometries of claim 7, wherein: the step (4) is specifically to perform linear preprocessing on projection data acquired by a CT scanning device, and then input the projection data into the optimized intelligent imaging network obtained in the step (3) to obtain a filtered and reconstructed optimized CT image.
9. The CT imaging modality for a cloud image center of multiple imaging geometries of claim 8, wherein: the linear pretreatment is normalization;
the low dose data is dose data obtained by scanning under a low dose scanning protocol of low tube voltage or low tube current or low dose data obtained by simulation.
10. A CT imaging system for a cloud image center of multiple imaging geometries, characterized by: CT imaging method employing a cloud image center for a plurality of imaging geometries as claimed in any of claims 1 to 9.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310098648.6A CN116228905A (en) | 2023-02-10 | 2023-02-10 | CT imaging method and system for cloud image center with multiple imaging geometries |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310098648.6A CN116228905A (en) | 2023-02-10 | 2023-02-10 | CT imaging method and system for cloud image center with multiple imaging geometries |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116228905A true CN116228905A (en) | 2023-06-06 |
Family
ID=86576253
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310098648.6A Pending CN116228905A (en) | 2023-02-10 | 2023-02-10 | CT imaging method and system for cloud image center with multiple imaging geometries |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116228905A (en) |
-
2023
- 2023-02-10 CN CN202310098648.6A patent/CN116228905A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| JP7234064B2 (en) | Iterative image reconstruction framework | |
| Kuanar et al. | Low dose abdominal CT image reconstruction: an unsupervised learning based approach | |
| US11126914B2 (en) | Image generation using machine learning | |
| CN111325686B (en) | Low-dose PET three-dimensional reconstruction method based on deep learning | |
| JP6855223B2 (en) | Medical image processing device, X-ray computer tomographic imaging device and medical image processing method | |
| Yang et al. | Low-dose CT denoising via sinogram inner-structure transformer | |
| US12064281B2 (en) | Method and system for denoising CT images using a neural network | |
| JP2021013725A (en) | Medical apparatus | |
| CN112102428B (en) | CT cone beam scanning image reconstruction method, scanning system and storage medium | |
| CN113052936A (en) | Single-view CT reconstruction method integrating FDK and deep learning | |
| Zhang et al. | CT super-resolution using multiple dense residual block based GAN | |
| CN117813055A (en) | Multi-modality and multi-scale feature aggregation for synthesis of SPECT images from fast SPECT scans and CT images | |
| Li et al. | Eliminating CT radiation for clinical PET examination using deep learning | |
| CN113344876B (en) | Deformable registration method between CT and CBCT | |
| Whiteley et al. | FastPET: Near real-time PET reconstruction from histo-images using a neural network | |
| US20220375142A1 (en) | Few-view ct image reconstruction system | |
| CN117876261A (en) | CBCT scattering correction imaging method based on deep learning | |
| Yang et al. | Transfer learning framework for low‐dose CT reconstruction based on marginal distribution adaptation in multiscale | |
| US20220327750A1 (en) | Medical image processing method, medical image processing apparatus, and computer readable non-volatile storage medium storing medical image processing program | |
| CN116228905A (en) | CT imaging method and system for cloud image center with multiple imaging geometries | |
| Li et al. | A cascade-based dual-domain data correction network for sparse view CT image reconstruction | |
| CN110176045A (en) | A method of dual-energy CT image is generated by single energy CT image | |
| Zhong et al. | Super-resolution image reconstruction from sparsity regularization and deep residual-learned priors | |
| KR102329938B1 (en) | Method for processing conebeam computed tomography image using artificial neural network and apparatus therefor | |
| JP2024507766A (en) | Contrast enhancement using machine learning |
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 |