CN112381741A - Tomography image reconstruction method based on SPECT data sampling and noise characteristics - Google Patents
Tomography image reconstruction method based on SPECT data sampling and noise characteristics Download PDFInfo
- Publication number
- CN112381741A CN112381741A CN202011333288.6A CN202011333288A CN112381741A CN 112381741 A CN112381741 A CN 112381741A CN 202011333288 A CN202011333288 A CN 202011333288A CN 112381741 A CN112381741 A CN 112381741A
- Authority
- CN
- China
- Prior art keywords
- spect
- image
- projection data
- sampling
- noise
- 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
- 238000002603 single-photon emission computed tomography Methods 0.000 title claims abstract description 82
- 238000005070 sampling Methods 0.000 title claims abstract description 69
- 238000000034 method Methods 0.000 title claims abstract description 46
- 238000003325 tomography Methods 0.000 title abstract description 7
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 38
- 230000009467 reduction Effects 0.000 claims abstract description 27
- 238000012545 processing Methods 0.000 claims abstract description 14
- 238000013528 artificial neural network Methods 0.000 claims abstract description 13
- 238000012805 post-processing Methods 0.000 claims abstract description 5
- 230000008569 process Effects 0.000 claims description 7
- 238000007476 Maximum Likelihood Methods 0.000 claims description 6
- 230000005540 biological transmission Effects 0.000 claims description 3
- 239000011159 matrix material Substances 0.000 claims description 3
- 238000012549 training Methods 0.000 description 15
- 230000000694 effects Effects 0.000 description 10
- 238000003384 imaging method Methods 0.000 description 10
- 238000010586 diagram Methods 0.000 description 6
- 230000006870 function Effects 0.000 description 4
- 229940121896 radiopharmaceutical Drugs 0.000 description 4
- 239000012217 radiopharmaceutical Substances 0.000 description 4
- 230000002799 radiopharmaceutical effect Effects 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 3
- 238000002600 positron emission tomography Methods 0.000 description 3
- 238000002059 diagnostic imaging Methods 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 238000004088 simulation Methods 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 1
- 230000004075 alteration Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000000052 comparative effect Effects 0.000 description 1
- 230000006835 compression Effects 0.000 description 1
- 238000007906 compression Methods 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 201000010099 disease Diseases 0.000 description 1
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000001737 promoting effect Effects 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/70—Denoising; Smoothing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T11/00—2D [Two Dimensional] image generation
- G06T11/003—Reconstruction from projections, e.g. tomography
- G06T11/008—Specific post-processing after tomographic reconstruction, e.g. voxelisation, metal artifact correction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10072—Tomographic images
- G06T2207/10108—Single photon emission computed tomography [SPECT]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Nuclear Medicine (AREA)
- Image Processing (AREA)
Abstract
The tomographic image reconstruction method based on SPECT data sampling and noise characteristics comprises the following steps: a, evaluating the noise level of SPECT original projection data, and selecting a first convolution neural network matched with the noise level to perform noise reduction processing on the SPECT original projection data; b, applying a statistical iterative reconstruction algorithm to the projection data subjected to noise reduction to obtain a preliminary reconstructed image; step C, applying a second convolutional neural network matched with the number of the sampling angles of the SPECT projection data to carry out post-processing on the primary reconstructed image; and D, further applying an image iterative reconstruction algorithm based on compressed sensing to obtain a final reconstructed image based on the reconstructed image after artifact removal and the SPECT original projection data. The reconstruction method can shorten the time for acquiring the SPECT tomography and improve the efficiency particularly for the sparsely sampled SPECT data.
Description
Technical Field
The invention relates to the technical field of medical imaging, in particular to a tomographic image reconstruction method based on SPECT data sampling and noise characteristics.
Background
SPECT (Single-Photon Emission Computed Tomography) is a medical imaging device whose basic working principle is as follows: the gamma camera is used for rotating around an imaging target (patient), gamma photons emitted by the radiopharmaceutical in the target body and collimated by the collimator are detected at different angles to form two-dimensional projection data, and the three-dimensional spatial distribution of the radiopharmaceutical in the imaging target body is reconstructed by applying a tomographic image reconstruction algorithm to the projection data acquired at all sampling angles.
Subject to imaging physics, SPECT requires acquisition at each projection angle for a certain time, typically between 20-60 seconds, to accumulate gamma photon counts, thereby increasing the signal-to-noise ratio of the projection data, even though the noise level of SPECT projection data is still much higher than other similar radiological imaging devices such as PET (positron emission tomography) and X-ray CT. To meet the angular sampling requirement of tomography, in a whole-body imaging application, the conventional SPECT needs to complete 60-angle projection data sampling within 360-degree angle around the axial direction of an imaging target to complete tomography of a bed, covering about 40 cm of axial field of view, and consuming about 15-20 minutes. By analogy, tomographic imaging with 3 beds covering about a 1.2 meter axial field of view requires about 1 hour, which is unacceptable from the standpoint of clinical work efficiency and patient tolerance. Therefore, the image reconstruction method which can keep the image quality unchanged basically on the premise of reducing the data acquisition time can play an important role in promoting the popularization of the clinical application of SPECT tomography.
In recent years, deep learning techniques represented by convolutional neural networks are widely applied to the fields of CT and PET medical image noise reduction, artifact removal introduced by sparse sampling and the like, and a good effect is achieved. But the method has the limitation that the adaptability to differentiated data is poor, and when the noise level and artifact characteristics of actual application data and training neural network data are close, good noise reduction and artifact removal effects can be achieved; otherwise, the noise reduction and artifact removal effects are reduced to different degrees.
For SPECT imaging, reducing the acquisition time can be achieved by reducing the sampling time per angle, reducing the number of sampling angles, and a combination of both, with different selections of acquisition time and number of sampling angles resulting in greater differences in the characteristics of the data and image noise and sparsely sampled artifacts. In addition, in clinical practice, the noise level of SPECT raw projection data and images is also affected by many factors: such as different types or doses of radiopharmaceuticals, differential distribution of radiopharmaceuticals between different patients or different parts of the same patient, and parameter settings of conventional image reconstruction algorithms. Therefore, a single, fixed-parameter convolutional neural network cannot well complete the functions of noise reduction and artifact removal of the SPECT sparse sampling tomographic reconstruction.
Disclosure of Invention
In view of the above-mentioned drawbacks, the present invention aims to provide a tomographic image reconstruction method based on SPECT data sampling and noise characteristics, and particularly to solve one or more of the above-mentioned problems for sparsely sampled SPECT data, so as to shorten the SPECT tomographic imaging acquisition time and improve the efficiency.
In order to achieve the purpose, the invention adopts the following technical scheme:
the tomographic image reconstruction method based on SPECT data sampling and noise characteristics comprises the following steps:
a, evaluating the noise level of SPECT original projection data by using a Poisson noise model, and selecting a first convolution neural network matched with the noise level to perform noise reduction processing on the SPECT original projection data;
b, applying a statistical iterative reconstruction algorithm based on a physical model and fixed parameters to the projection data after noise reduction to obtain a primary reconstructed image;
step C, applying a second convolutional neural network matched with the number of the sampling angles of the SPECT projection data to carry out post-processing on the primary reconstructed image, and removing artifacts caused by sparse sampling;
and D, further applying an image iterative reconstruction algorithm based on compressed sensing to obtain a final reconstructed image based on the reconstructed image after artifact removal and the SPECT original projection data.
To further illustrate, the method for evaluating the noise level of the SPECT raw projection data in step a is: calculating the median or average of all pixel values in the SPECT raw projection data greater than zero
And classifying it by noise level according to the following method:
first noise level:
second noise level:
third noise level:
fourth noise level:
further, the first convolutional neural network applied in step a performs noise reduction on two-dimensional projection data at different angles respectively or performs joint noise reduction on three-dimensional data composed of projection data at multiple angles.
Preferably, the statistical iterative algorithm in step B adopts a maximum likelihood iterative or ordered subset accelerated maximum likelihood iterative reconstruction algorithm.
Specifically, in step C, the number V of sampling angles of the SPECT raw projection data is first classified according to the following method:
first sampling angle range: v ═ {8,9,10 };
second sampling angle range: v ═ {12,14,16 };
third sampling angle range: v ═ 18,20,24 };
fourth sampling angle range: v ═ 30,32,36, 40.
Specifically, in the step C, the second convolutional neural network performs artifact removal processing on each two-dimensional image layer in the tomographic image, or performs overall processing on the three-dimensional tomographic image.
Specifically, the image iterative reconstruction algorithm in the step D has a formula:
wherein x is a SPECT target image vector to be reconstructed, xpC, an image vector after artifact removal processing in the step c, y is a SPECT original projection data vector, A is a system transmission matrix for carrying out physical modeling on the SPECT original projection data acquisition process,
for the final reconstructed image, tv (x) is a fully-variational model, i.e., the L2 norm of the image bi-directional spatial gradient, which is formulated when x is a vector representation of a two-dimensional image of size M × N:
wherein M and N are natural numbers, N is 1, 2, …, N-1,
is the L1 norm of the image vector x,i.e. the sum of the absolute values of all pixel values, alpha is a weight parameter.
Further, the image iterative reconstruction algorithm in step D is: on the premise of meeting the consistency condition of the SPECT reconstruction target image and the SPECT original projection data, namely Ax is y, solving the solution of the sparsity cost function which can minimize the difference between the SPECT reconstruction target image and the artifact-removed image in the formula (1) to serve as a final reconstruction image.
Preferably, the solution formula (2) in step D is optimized, specifically, the artifact-removed image is used as an initial estimate, the SPECT target reconstructed image is updated by iteratively and alternately applying the consistency condition Ax-y in the formula (2) and the sparsity minimization condition in the formula (1), and a final reconstructed image is obtained after convergence.
The invention can achieve the following beneficial effects:
1. aiming at differentiated SPECT original projection data and noise characteristics thereof in clinical practical application, a first convolution neural network and a second convolution neural network for noise reduction and artifact removal of self-adaptive parameters are applied to realize optimized image quality;
2. under the condition of properly reducing the data acquisition time, the image quality is ensured not to be changed basically through the self-adaptive optimized noise reduction and artifact removal algorithm, so that the patient examination efficiency is improved;
3. the two convolutional neural networks are adopted to respectively process the problems of noise and sparse sampling artifacts, and compared with the scheme adopting a single network, the complexity and the training difficulty of network parameters are reduced.
4. By applying a compressed sensing technology and original projection data, the reconstructed image is further perfected on the basis of two convolutional neural networks, and the false details of the image, which are introduced due to the limitation of the convolutional neural networks, are avoided to the greatest extent.
Drawings
FIG. 1 is a flow chart of the present invention.
FIG. 2 is a schematic structural diagram of a first convolutional neural network MAP-NN for denoising projection data according to an embodiment of the present invention.
FIG. 3 is a schematic diagram of the noise reduction effect of the first convolutional neural network MAP-NN according to the present invention.
FIG. 4 is a diagram illustrating the effect of applying MAP-NN to reduce noise in projection data according to an embodiment of the present invention.
FIG. 5 is a schematic diagram of a U-Net network structure according to an embodiment of the present invention.
Fig. 6 is a schematic comparison diagram of the effect of removing the artifact by using the Unet network optimized for the preliminary reconstructed images of two sampling angles respectively and the effect of removing the artifact by using the conventional Unet network in an embodiment of the present invention.
FIG. 7 is a diagram illustrating the comparison of the reconstructed image obtained by applying the complete reconstruction method with other contrast methods according to an embodiment of the present invention.
Detailed Description
Reference will now be made in detail to embodiments of the present invention, examples of which are illustrated in the accompanying drawings, wherein like or similar reference numerals refer to the same or similar elements or elements having the same or similar function throughout. The embodiments described below with reference to the accompanying drawings are illustrative only for the purpose of explaining the present invention, and are not to be construed as limiting the present invention.
Furthermore, the terms "first", "second" and "first" are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated.
The tomographic image reconstruction method based on SPECT data sampling and noise characteristics comprises the following steps:
a, evaluating the noise level of SPECT original projection data by using a Poisson noise model, and selecting a first convolution neural network matched with the noise level to perform noise reduction processing on the SPECT original projection data;
b, applying a statistical iterative reconstruction algorithm based on a physical model and fixed parameters to the projection data after noise reduction to obtain a primary reconstructed image;
step C, applying a second convolutional neural network matched with the number of the sampling angles of the SPECT projection data to carry out post-processing on the primary reconstructed image, and removing artifacts caused by sparse sampling;
and D, further applying an image iterative reconstruction algorithm based on compressed sensing to obtain a final reconstructed image based on the reconstructed image after artifact removal and the SPECT original projection data.
The tomographic image reconstruction method provided by the application aims at differentiated SPECT data and noise characteristics in clinical practical application, applies noise reduction and artifact removal convolutional neural networks of adaptive parameters to realize optimized image quality, concretely, adopts two convolutional neural networks to respectively process the problems of noise and sparse sampling artifacts, and compared with a scheme adopting a single network, reduces the complexity and training difficulty of network parameters, simultaneously applies a compression sensing technology and original projection data, further perfects a reconstructed image on the basis of the convolutional neural networks, and avoids the false details of the image introduced due to the limitation of the convolutional neural networks to the maximum extent. The method is particularly suitable for sparsely sampled SPECT original projection data, solves the problem that the characteristics of data and image noise and sparsely sampled artifacts have large difference, and can shorten the SPECT tomography acquisition time, thereby improving the efficiency.
To further illustrate, the method for evaluating the noise level of the SPECT raw projection data in step a is: calculating the median or average of all pixel values in the SPECT raw projection data greater than zero
And classifying it by noise level according to the following method:
first noise level:
second noise level:
third noise level:
fourth noise level:
the noise reduction of the SPECT raw projection data is realized by selecting a first convolution neural network with network parameters matched with the estimated noise level, and different neural network parameters in the first convolution neural network are obtained by parameter training by applying a training data set of the corresponding noise level.
Preferably, the first convolutional neural network applied in step a performs noise reduction on two-dimensional projection data at different angles respectively or performs joint noise reduction on three-dimensional data composed of projection data at multiple angles.
Preferably, the statistical iterative algorithm in step B adopts a maximum likelihood iteration or an ordered subset accelerated maximum likelihood iterative reconstruction algorithm.
Physical processes such as gamma photon attenuation, scattering and collimator detector response in the SPECT imaging process are modeled, and meanwhile, the number of image iteration updates (the number of image iteration updates is the number of subsets and the number of full iterations) is kept fixed.
To explain further, in step C, the number V of sampling angles of the SPECT raw projection data is classified as follows:
first sampling angle range: v ═ {8,9,10 };
second sampling angle range: v ═ {12,14,16 };
third sampling angle range: v ═ 18,20,24 };
fourth sampling angle range: v ═ 30,32,36, 40.
The sparse artifact removing processing of the primary reconstruction image is realized by a second convolutional neural network with network parameters corresponding to the number of sampling angles of SPECT original projection data, and different neural network parameters in the second convolutional neural network are obtained by performing parameter training by applying a training data set with a corresponding sampling angle range.
Preferably, in the step C, the second convolutional neural network performs artifact removal processing on each two-dimensional image layer in the tomographic image, or performs overall processing on the three-dimensional tomographic image.
Further, the iterative reconstruction algorithm of the image in step D has the following formula:
wherein x is an image vector of a preliminary reconstructed image to be reconstructed, xpThe image vector after artifact removal processing in the step C, y is a SPECT original projection data vector, A is a system transmission matrix for carrying out physical modeling on the SPECT original projection data acquisition process,
for the final reconstructed image, tv (x) is a fully-variational model, i.e., the L2 norm of the image bi-directional spatial gradient, which is formulated when x is a vector representation of a two-dimensional image of size M × N:
wherein M and N are natural numbers, N is 1, 2, …, N-1,
is the L1 norm of the image vector x, i.e. the sum of the absolute values of all pixel values, and α is a weight parameter.
Preferably, the image iterative reconstruction algorithm in step D is: on the premise of meeting the consistency condition of the SPECT reconstruction target image and the SPECT original projection data, namely Ax is y, solving the solution of the sparsity cost function which can minimize the difference between the SPECT reconstruction target image and the artifact-removed image in the formula (1) to serve as a final reconstruction image.
Preferably, the solution formula (2) in step D is optimized, specifically, the artifact-removed image is used as an initial estimate, the SPECT target reconstructed image is updated by iteratively and alternately applying the consistency condition Ax ═ y in the formula (2) and the sparsity minimization condition in the formula (1), and the final reconstructed image is obtained after convergence.
Specific examples are as follows:
referring to FIG. 2, a first convolutional neural network MAP-NN structure is shown, in which a plurality of coder-decoder combinations with the same structure are included in the MAP-NN structure, and FIG. 3 shows the noise reduction effect of different numbers of coders/decoders. In the first convolution neural network MAP-NN training of this embodiment, the noise-free simulated projection data is used as the output of the network, and the simulated projection data added with poisson noise is used as the input of the network. The number of codec combinations T in training is 5. When training, a total of 4 networks of noise levels are trained,
average count rate representing pixels whose count is greater than 0:
first noise level:
second noise level:
third noise level:
fourth noise level:
in the use of the MAP-NN network trained in this embodiment, after training is completed, the number of encoder/decoder combinations D used for prediction using the model is 5. When the method is used, firstly, a program counts the counting rate of an image, and performs noise reduction on the SPECT original projection data according to the calling of the corresponding network weight, wherein the effect after the noise reduction can be seen in FIG. 4, from left to right: (1) the preliminary reconstructed image without noise reduction, (2) is the preliminary reconstructed image after noise reduction in step a.
In another embodiment, in the training of the second convolutional neural network U-Net, the present embodiment uses the simulation reconstruction result of sparse angular sampling as the input of the network, and uses the simulation reconstruction result of sufficient angular sampling as the output of the network to train the second convolutional neural network U-Net. During training, according to the sampling angle number used by an input image, four networks are trained in total:
first sampling angle range: v ═ {8,9,10 };
second sampling angle range: v ═ {12,14,16 };
third sampling angle range: v ═ 18,20,24 };
fourth sampling angle range: v ═ 30,32,36, 40.
When the second convolutional neural network U-Net is used, the program selects a corresponding network weight according to the number of reconstruction angles to remove artifacts from the preliminary reconstructed image, as shown in fig. 5, artifacts exist around a highlight region in the input image due to the detection technology, and after the output image is processed by the second convolutional neural network U-Net, the artifacts are removed to obtain an output image with higher quality, thereby avoiding the interference of the artifacts in the image on subsequent disease diagnosis.
For the comparative example group, the targeted sparsely sampled second convolutional neural network U-Net and the conventional same convolutional neural network are used for post-processing the same image, and the artifact removing effect is schematically shown in fig. 6, which is from left to right:
(1) is based on a conventional reconstructed image of 12 sampling angles;
(2) the method is an image which is subjected to artifact removal by a convolutional neural network U-Net trained based on 12 sampling angle data;
(3) the method is an image which is subjected to artifact removal by a convolutional neural network U-Net trained on the basis of various sampling angle data between 12 and 20;
(4) is a conventional reconstructed image (true value) based on 60 sampling angles.
The image obtained by the second convolutional neural network U-Net with the pertinence sparse sampling provided by the application under the condition of 12 sampling angle data training or 12-20 multiple sampling angle data training is very approximate to the conventional reconstructed image adopting 60 sampling angles in the conventional method, namely under the second convolutional neural network U-Net with the pertinence sparse sampling provided by the application, the number of the sampling angles is reduced, better image quality can be still ensured, the image quality is basically not changed compared with the images with the number of multiple sampling angles, the data acquisition time can be reduced, and the patient inspection efficiency is improved.
Compared with the conventional other methods, the complete tomographic image reconstruction method provided by the application has the comparison result as shown in fig. 7, from left to right:
(1) a conventional reconstructed image of 15 angular noisy projection data;
(2) applying a convolutional neural network to reduce noise aiming at projection data, and applying the convolutional neural network to the reconstruction result of the conventional statistical iterative algorithm to remove artifacts;
(3) the reconstructed image by using the method of the invention is that the image after compressed sensing reconstruction is applied on the basis of the step (2);
(4) the 60-angle low-noise projection data reconstructs an image (true value).
Combining the (2), (3) and (4), the quality of the image obtained by applying compressed sensing reconstruction on the basis of applying the convolutional neural network to remove artifacts after convolutional neural network noise reduction can be applied to projection data and applying the convolutional neural network to remove artifacts to the reconstruction result of the conventional statistical iterative algorithm is obviously improved; and the image quality of (3) is substantially the same as the image quality of (4).
In the description herein, references to the description of the term "one embodiment," "another embodiment," "an example," "a specific example," or "some examples," etc., mean that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the invention. In this specification, the schematic representations of the terms used above do not necessarily refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.
While embodiments of the invention have been shown and described, it will be understood by those of ordinary skill in the art that: various changes, modifications, substitutions and alterations can be made to the embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.
Claims (9)
1. The tomographic image reconstruction method based on SPECT data sampling and noise characteristics is characterized in that: the method comprises the following steps:
a, evaluating the noise level of SPECT original projection data by using a Poisson noise model, and selecting a first convolution neural network matched with the noise level to perform noise reduction processing on the SPECT original projection data;
b, applying a statistical iterative reconstruction algorithm based on a physical model and fixed parameters to the projection data after noise reduction to obtain a primary reconstructed image;
step C, applying a second convolutional neural network matched with the number of the sampling angles of the SPECT projection data to carry out post-processing on the primary reconstructed image, and removing artifacts caused by sparse sampling;
and D, further applying an image iterative reconstruction algorithm based on compressed sensing to obtain a final reconstructed image based on the reconstructed image after artifact removal and the SPECT original projection data.
2. The SPECT data sampling and noise characteristic-based tomographic image reconstruction method of claim 1, wherein: the method for evaluating the noise level of the SPECT raw projection data in step a is: calculating the median or average of all pixel values in the SPECT raw projection data greater than zero
And the method is as followsAnd (3) carrying out noise level classification:
first noise level:
second noise level:
third noise level:
fourth noise level:
3. the SPECT data sampling and noise characteristic-based tomographic image reconstruction method of claim 2, wherein: and B, respectively denoising the two-dimensional projection data of different angles or jointly denoising the three-dimensional data formed by the projection data of a plurality of angles by using the first convolution neural network applied in the step A.
4. The SPECT data sampling and noise characteristic-based tomographic image reconstruction method of claim 1, wherein: and B, adopting maximum likelihood iteration or ordered subset accelerated maximum likelihood iterative reconstruction algorithm for the statistical iterative algorithm in the step B.
5. The SPECT data sampling and noise characteristic-based tomographic image reconstruction method of claim 1, wherein: in step C, the number V of sampling angles of the SPECT raw projection data is first classified according to the following method:
first sampling angle range: v ═ {8,9,10 };
second sampling angle range: v ═ {12,14,16 };
third sampling angle range: v ═ 18,20,24 };
fourth sampling angle range: v ═ 30,32,36, 40.
6. The SPECT data sampling and noise characteristic-based tomographic image reconstruction method of claim 1, wherein: and in the step C, the second convolutional neural network respectively performs artifact removal processing on each two-dimensional image layer in the tomographic image or performs overall processing on the three-dimensional tomographic image.
7. The SPECT data sampling and noise characteristic-based tomographic image reconstruction method of claim 1, wherein: the image iterative reconstruction algorithm in the step D has the formula:
wherein x is a SPECT target image vector to be reconstructed, xpC, an image vector after artifact removal processing in the step c, y is a SPECT original projection data vector, A is a system transmission matrix for carrying out physical modeling on the SPECT original projection data acquisition process,
for the final reconstructed image, tv (x) is a fully-variational model, i.e., the L2 norm of the image bi-directional spatial gradient, which is formulated when x is a vector representation of a two-dimensional image of size M × N:
wherein M and N are natural numbers, N is 1, 2, …, N-1,
is the L1 norm of the image vector x, i.e. the sum of the absolute values of all pixel values, and α is a weight parameter.
8. The SPECT data sampling and noise characteristic-based tomographic image reconstruction method of claim 7, wherein: step D, the image iterative reconstruction algorithm is as follows: on the premise of meeting the consistency condition of the SPECT reconstruction target image and the SPECT original projection data, namely Ax is y, solving the solution of the sparsity cost function which can minimize the difference between the SPECT reconstruction target image and the artifact-removed image in the formula (1) to serve as a final reconstruction image.
9. The SPECT data sampling and noise characteristic-based tomographic image reconstruction method of claim 8, wherein: and D, optimizing the solving formula (2) in the step D, specifically, taking the image after artifact removal as initial estimation, updating the SPECT target reconstruction image by iteratively and alternately applying the consistency condition Ax-y in the formula (2) and the sparsity minimizing condition in the formula (1), and obtaining a final reconstruction image after convergence.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011333288.6A CN112381741B (en) | 2020-11-24 | 2020-11-24 | Tomography image reconstruction method based on SPECT data sampling and noise characteristics |
| PCT/CN2021/073324 WO2022110530A1 (en) | 2020-11-24 | 2021-01-22 | Tomographic image reconstruction method based on spect data sampling and noise characteristics |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011333288.6A CN112381741B (en) | 2020-11-24 | 2020-11-24 | Tomography image reconstruction method based on SPECT data sampling and noise characteristics |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112381741A true CN112381741A (en) | 2021-02-19 |
| CN112381741B CN112381741B (en) | 2021-07-16 |
Family
ID=74588367
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011333288.6A Active CN112381741B (en) | 2020-11-24 | 2020-11-24 | Tomography image reconstruction method based on SPECT data sampling and noise characteristics |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN112381741B (en) |
| WO (1) | WO2022110530A1 (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2021139835A3 (en) * | 2021-03-24 | 2021-11-11 | 苏州深透智能科技有限公司 | Spect imaging prediction model creation method and apparatus, and device and storage medium |
| CN113706419A (en) * | 2021-09-13 | 2021-11-26 | 上海联影医疗科技股份有限公司 | Image processing method and system |
| CN115187470A (en) * | 2022-06-10 | 2022-10-14 | 成都飞机工业（集团）有限责任公司 | Double-domain iterative noise reduction method based on 3D printing inner cavity |
| CN115423890A (en) * | 2022-09-15 | 2022-12-02 | 京心禾(北京)医疗科技有限公司 | Tomographic image iterative reconstruction method |
| US11540798B2 (en) | 2019-08-30 | 2023-01-03 | The Research Foundation For The State University Of New York | Dilated convolutional neural network system and method for positron emission tomography (PET) image denoising |
| CN115908204A (en) * | 2023-02-21 | 2023-04-04 | 北京唯迈医疗设备有限公司 | Method, apparatus and medium for noise reduction processing of medical images acquired using a radiological imaging device |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115856767B (en) * | 2023-02-08 | 2023-05-05 | 安徽大学 | Reconfigurable intelligent super-surface-assisted wave arrival direction estimation method |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104077763A (en) * | 2014-07-08 | 2014-10-01 | 中国科学院高能物理研究所 | TOF-PET image reconstruction method based on compressed sensing theory |
| CN108898642A (en) * | 2018-06-01 | 2018-11-27 | 安徽工程大学 | A kind of sparse angular CT imaging method based on convolutional neural networks |
| US20190104940A1 (en) * | 2017-10-06 | 2019-04-11 | Toshiba Medical Systems Corporation | Apparatus and method for medical image reconstruction using deep learning for computed tomography (ct) image noise and artifacts reduction |
| CN109685206A (en) * | 2018-09-30 | 2019-04-26 | 上海联影医疗科技有限公司 | The system and method for generating the neural network model for image procossing |
| CN110264559A (en) * | 2019-04-01 | 2019-09-20 | 佛山原子医疗设备有限公司 | A kind of bone tomographic image reconstructing process and system |
| CN110660123A (en) * | 2018-06-29 | 2020-01-07 | 清华大学 | Three-dimensional CT image reconstruction method and device based on neural network and storage medium |
| CN111292386A (en) * | 2020-01-15 | 2020-06-16 | 中国人民解放军战略支援部队信息工程大学 | CT projection metal trace completion metal artifact correction method based on U-net |
| US20200311878A1 (en) * | 2019-04-01 | 2020-10-01 | Canon Medical Systems Corporation | Apparatus and method for image reconstruction using feature-aware deep learning |
Family Cites Families (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10475214B2 (en) * | 2017-04-05 | 2019-11-12 | General Electric Company | Tomographic reconstruction based on deep learning |
| US11126914B2 (en) * | 2017-10-11 | 2021-09-21 | General Electric Company | Image generation using machine learning |
| CN110249365B (en) * | 2017-11-10 | 2023-05-30 | 上海联影医疗科技股份有限公司 | System and method for image reconstruction |
| US10977842B2 (en) * | 2018-06-04 | 2021-04-13 | Korea Advanced Institute Of Science And Technology | Method for processing multi-directional X-ray computed tomography image using artificial neural network and apparatus therefor |
| CN109559359B (en) * | 2018-09-27 | 2023-06-02 | 东南大学 | Artifact removal method for reconstructing image by sparse angle data based on deep learning |
| US10679385B1 (en) * | 2018-12-17 | 2020-06-09 | General Electric Company | System and method for statistical iterative reconstruction and material decomposition |
| CN110151210B (en) * | 2019-05-21 | 2022-11-22 | 上海联影医疗科技股份有限公司 | Medical image processing method, system, device and computer readable medium |
| CN111681297B (en) * | 2020-05-28 | 2024-02-27 | 上海联影智能医疗科技有限公司 | Image reconstruction method, computer device, and storage medium |
-
2020
- 2020-11-24 CN CN202011333288.6A patent/CN112381741B/en active Active
-
2021
- 2021-01-22 WO PCT/CN2021/073324 patent/WO2022110530A1/en active Application Filing
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104077763A (en) * | 2014-07-08 | 2014-10-01 | 中国科学院高能物理研究所 | TOF-PET image reconstruction method based on compressed sensing theory |
| US20190104940A1 (en) * | 2017-10-06 | 2019-04-11 | Toshiba Medical Systems Corporation | Apparatus and method for medical image reconstruction using deep learning for computed tomography (ct) image noise and artifacts reduction |
| CN108898642A (en) * | 2018-06-01 | 2018-11-27 | 安徽工程大学 | A kind of sparse angular CT imaging method based on convolutional neural networks |
| CN110660123A (en) * | 2018-06-29 | 2020-01-07 | 清华大学 | Three-dimensional CT image reconstruction method and device based on neural network and storage medium |
| CN109685206A (en) * | 2018-09-30 | 2019-04-26 | 上海联影医疗科技有限公司 | The system and method for generating the neural network model for image procossing |
| CN110264559A (en) * | 2019-04-01 | 2019-09-20 | 佛山原子医疗设备有限公司 | A kind of bone tomographic image reconstructing process and system |
| US20200311878A1 (en) * | 2019-04-01 | 2020-10-01 | Canon Medical Systems Corporation | Apparatus and method for image reconstruction using feature-aware deep learning |
| CN111292386A (en) * | 2020-01-15 | 2020-06-16 | 中国人民解放军战略支援部队信息工程大学 | CT projection metal trace completion metal artifact correction method based on U-net |
Non-Patent Citations (2)
| Title |
|---|
| KECHENG CHEN 等: "Low-Dose CT Image Blind Denoising with Graph Convolutional Networks", 《ICONIP 2020》 * |
| XIANGRUI YIN 等: "Domain Progressive 3D Residual Convolution Network to Improve Low-Dose CT Imaging", 《IEEE TRANSACTIONS ON MEDICAL IMAGING》 * |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11540798B2 (en) | 2019-08-30 | 2023-01-03 | The Research Foundation For The State University Of New York | Dilated convolutional neural network system and method for positron emission tomography (PET) image denoising |
| WO2021139835A3 (en) * | 2021-03-24 | 2021-11-11 | 苏州深透智能科技有限公司 | Spect imaging prediction model creation method and apparatus, and device and storage medium |
| CN113706419A (en) * | 2021-09-13 | 2021-11-26 | 上海联影医疗科技股份有限公司 | Image processing method and system |
| CN115187470A (en) * | 2022-06-10 | 2022-10-14 | 成都飞机工业（集团）有限责任公司 | Double-domain iterative noise reduction method based on 3D printing inner cavity |
| CN115423890A (en) * | 2022-09-15 | 2022-12-02 | 京心禾(北京)医疗科技有限公司 | Tomographic image iterative reconstruction method |
| CN115423890B (en) * | 2022-09-15 | 2023-09-19 | 京心禾(北京)医疗科技有限公司 | Tomographic image iterative reconstruction method |
| CN115908204A (en) * | 2023-02-21 | 2023-04-04 | 北京唯迈医疗设备有限公司 | Method, apparatus and medium for noise reduction processing of medical images acquired using a radiological imaging device |
| CN115908204B (en) * | 2023-02-21 | 2023-06-06 | 北京唯迈医疗设备有限公司 | Noise reduction processing method and device for medical image acquired by radiological imaging equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| WO2022110530A1 (en) | 2022-06-02 |
| CN112381741B (en) | 2021-07-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112381741B (en) | Tomography image reconstruction method based on SPECT data sampling and noise characteristics | |
| JP7234064B2 (en) | Iterative image reconstruction framework | |
| CN111325686B (en) | Low-dose PET three-dimensional reconstruction method based on deep learning | |
| CN108898642B (en) | Sparse angle CT imaging method based on convolutional neural network | |
| Qiao et al. | A motion-incorporated reconstruction method for gated PET studies | |
| JP2024054204A (en) | Neural network learning method, program, medical image processing method, and medical device | |
| Lim et al. | Improved low-count quantitative PET reconstruction with an iterative neural network | |
| Gao et al. | Robust principal component analysis-based four-dimensional computed tomography | |
| JP2023159080A (en) | Medical image processing apparatus and medical image processing system | |
| La Riviere et al. | Reduction of noise-induced streak artifacts in X-ray computed tomography through spline-based penalized-likelihood sinogram smoothing | |
| CN115605915A (en) | Image reconstruction system and method | |
| CN108209954B (en) | Emission type computed tomography image reconstruction method and system | |
| JP2021013725A (en) | Medical apparatus | |
| CN112258642B (en) | Low-dose PET data three-dimensional iterative updating reconstruction method based on deep learning | |
| AU2019271915A1 (en) | Method and system for motion correction in CT imaging | |
| Xue et al. | LCPR-Net: low-count PET image reconstruction using the domain transform and cycle-consistent generative adversarial networks | |
| CN109791701A (en) | The iterative image reconstruction that the dynamic of formation with the artifact to noise-inducing inhibits | |
| Ote et al. | List-mode PET image reconstruction using deep image prior | |
| Zhang et al. | DREAM-Net: Deep residual error iterative minimization network for sparse-view CT reconstruction | |
| Chen et al. | Low-dose CT image denoising model based on sparse representation by stationarily classified sub-dictionaries | |
| Li et al. | LU-Net: combining LSTM and U-Net for sinogram synthesis in sparse-view SPECT reconstruction | |
| Marin et al. | Deformable left‐ventricle mesh model for motion‐compensated filtering in cardiac gated SPECT | |
| Chang et al. | Deep learning image transformation under radon transform | |
| Hu et al. | Simultaneous attenuation correction, scatter correction, and denoising in pet imaging with deep learning | |
| Fessler | Iterative image reconstruction for CT |
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 |