CN110992290A - Training method and system for low-dose CT image denoising network - Google Patents
Training method and system for low-dose CT image denoising network Download PDFInfo
- Publication number
- CN110992290A CN110992290A CN201911249569.0A CN201911249569A CN110992290A CN 110992290 A CN110992290 A CN 110992290A CN 201911249569 A CN201911249569 A CN 201911249569A CN 110992290 A CN110992290 A CN 110992290A
- Authority
- CN
- China
- Prior art keywords
- layer
- convolution
- image
- dose
- low
- 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
- 238000012549 training Methods 0.000 title claims abstract description 133
- 238000000034 method Methods 0.000 title claims abstract description 41
- 230000004927 fusion Effects 0.000 claims abstract description 35
- 238000004590 computer program Methods 0.000 claims description 11
- 238000003860 storage Methods 0.000 claims description 9
- 238000010276 construction Methods 0.000 claims description 5
- 238000002591 computed tomography Methods 0.000 description 95
- 230000006870 function Effects 0.000 description 30
- 238000005070 sampling Methods 0.000 description 21
- 238000010586 diagram Methods 0.000 description 14
- 230000004913 activation Effects 0.000 description 11
- 238000012545 processing Methods 0.000 description 9
- 238000005457 optimization Methods 0.000 description 5
- 230000005855 radiation Effects 0.000 description 5
- 238000013170 computed tomography imaging Methods 0.000 description 3
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 230000006978 adaptation Effects 0.000 description 1
- 238000003759 clinical diagnosis Methods 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 238000013500 data storage Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 239000000835 fiber Substances 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 238000012804 iterative process Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000012827 research and development Methods 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 239000007787 solid Substances 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
- 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
- G06T2207/10081—Computed x-ray tomography [CT]
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (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)
- Apparatus For Radiation Diagnosis (AREA)
Abstract
The invention provides a training method of a low-dose CT image denoising network, which comprises the following steps: acquiring a training data set; establishing a low-dose CT image denoising network, wherein the low-dose CT image denoising network comprises a first convolution layer, a convolution module, a first fusion layer and a second convolution layer which are sequentially connected, the first fusion layer is used for fusing an input signal of the convolution module and an output signal of the convolution module, the convolution module comprises at least one convolution network, the at least one convolution network is sequentially connected, each convolution network comprises a channel layer, a third convolution layer and a second fusion layer which are sequentially connected, the channel layer comprises a first channel, the first channel comprises a fourth convolution layer and a first deconvolution layer, and the fourth convolution layer and the first deconvolution layer are alternately connected; and training the low-dose CT image denoising network by using the training data set. According to the training method of the low-dose CT image denoising network, the fourth convolution layer and the first deconvolution layer are alternately connected, so that information loss can be avoided.
Description
Technical Field
The invention relates to the technical field of low-dose CT image reconstruction, in particular to a training method and a training system for a low-dose CT image denoising network.
Background
Computed Tomography (CT) is an important imaging means for obtaining internal structural information of an object in a nondestructive manner, has many advantages of high resolution, high sensitivity, multiple levels and the like, is one of medical image diagnostic devices with the largest machine loading amount in China, and is widely applied to various medical clinical examination fields. However, as the use of X-rays is required during CT scanning, the problem of CT radiation dose is increasingly gaining attention as people become increasingly aware of the potential hazards of radiation. The rationale for using Low doses (As Low As reasonable Achievable, ALARA) requires that the radiation dose to the patient be minimized while meeting the clinical diagnosis. Therefore, the research and development of a new low-dose CT imaging method can ensure the CT imaging quality and reduce the harmful radiation dose, and has important scientific significance and application prospect in the field of medical diagnosis. However, in the existing low-dose CT imaging method, it is difficult to obtain a clear CT image under the condition of satisfying the low-dose CT radiation.
Disclosure of Invention
In order to solve the defects of the prior art, the invention provides a training method and a training system for a low-dose CT image denoising network, which can avoid information loss caused by continuous multiple convolution or continuous multiple deconvolution operation after continuous multiple convolution and better extract the details of an original image.
The specific technical scheme provided by the invention is as follows: a training method of a low-dose CT image denoising network is provided, and comprises the following steps:
acquiring a training data set, wherein the training data set comprises a plurality of training image block groups, each training image block group comprises a first image block and a second image block, and the first image block and the second image block are respectively image blocks positioned at the same position in a low-dose CT image and a standard-dose CT image block;
establishing a low-dose CT image denoising network, wherein the low-dose CT image denoising network comprises a first convolution layer, a convolution module, a first fusion layer and a second convolution layer which are sequentially connected, the first fusion layer is used for fusing an input signal of the convolution module and an output signal of the convolution module, the convolution module comprises at least one convolution network, the at least one convolution network is sequentially connected, each convolution network comprises a channel layer, a third convolution layer and a second fusion layer which are sequentially connected, the channel layer comprises a first channel, the first channel comprises a fourth convolution layer and a first deconvolution layer, the fourth convolution layer and the first deconvolution layer are alternately connected, and the second fusion layer is used for fusing an input signal of the convolution network and an output signal of the third convolution layer;
and training the low-dose CT image denoising network by using the training data set to obtain an updated low-dose CT image denoising network.
Furthermore, the fourth convolution layer and the first deconvolution layer are alternately connected in sequence.
Furthermore, the first deconvolution layer and the fourth convolution layer are alternately connected in sequence.
Further, the channel layer further includes a second channel and a splicing layer, the second channel is connected in parallel with the first channel, the splicing layer is connected between the channel layer and the third convolution layer, and the splicing layer is used for splicing an output signal of the first channel and an output signal of the second channel; the second channel comprises a fifth convolution layer and a second deconvolution layer, and the second deconvolution layer and the fifth convolution layer are sequentially and alternately connected.
Further, the channel layer further includes a second channel and a splicing layer, the second channel is connected in parallel with the first channel, the splicing layer is connected between the channel layer and the third convolution layer, and the splicing layer is used for splicing an output signal of the first channel and an output signal of the second channel; the second channel comprises a fifth convolution layer and a second deconvolution layer, and the second deconvolution layer and the fifth convolution layer are sequentially and alternately connected.
Further, training the low-dose CT image using the training data specifically includes:
inputting a first image block in the training image block groups into the low-dose CT image denoising network to obtain a plurality of output images;
constructing a loss function according to the plurality of output images and the second image blocks in the plurality of training image block groups respectively;
optimizing the minimum value of the loss function to obtain an optimized network parameter;
and updating the low-dose CT image denoising network by using the optimized network parameters.
Further, the loss function is:
where loss (θ) represents a loss function, n represents the number of training image block groups in the training data set, and G (X)i(ii) a θ) represents the ith output image, YiRepresenting the second image block in the ith training image block group.
The invention also provides a training system of the low-dose CT image denoising network, which comprises:
the training data set acquisition module is used for acquiring a training data set, the training data set comprises a plurality of training image block groups, each training image block group comprises a first image block and a second image block, and the first image block and the second image block are respectively image blocks positioned at the same position in a low-dose CT image and a standard-dose CT image block;
the network construction module is used for establishing a low-dose CT image denoising network, the low-dose CT image denoising network comprises a first convolution layer, a convolution module, a first fusion layer and a second convolution layer which are connected in sequence, the first fusion layer is used for fusing an input signal of the convolution module and an output signal of the convolution module, the convolution module comprises at least one convolution network, the convolution networks are connected in sequence, each convolution network comprises a channel layer, a third convolution layer and a second fusion layer which are connected in sequence, the channel layer includes a first channel including a fourth convolutional layer and a first deconvolution layer, the fourth convolutional layer and the first deconvolution layer are alternately connected, and the second fusion layer is used for fusing an input signal of the convolutional network and an output signal of the third convolutional layer;
and the training module is used for training the low-dose CT image denoising network by using the training data set to obtain an updated low-dose CT image denoising network.
The invention also provides a denoising method of the low-dose CT image, which comprises the following steps: and inputting the low-dose CT image to be denoised into an updated low-dose CT image denoising network obtained by the training method of the low-dose CT image denoising network to obtain the denoised low-dose CT image.
The present invention also provides a computer storage medium having stored therein a computer program which, when read and executed by one or more processors, implements the method of training a low-dose CT image denoising network as described above.
According to the training method of the low-dose CT image denoising network, the fourth convolution layer and the first deconvolution layer are alternately connected, so that information loss caused by continuous multiple convolution or continuous multiple deconvolution operation after continuous multiple convolution can be avoided, and the input of each convolution network is fused with the output signal of the third convolution layer in the convolution network, so that the details of an original image can be better extracted, the problem of distortion after multiple cascading is avoided, and the reconstructed image is clearer.
Drawings
The technical solution and other advantages of the present invention will become apparent from the following detailed description of specific embodiments of the present invention, which is to be read in connection with the accompanying drawings.
FIG. 1 is a flowchart illustrating a training method for a medium/low dose CT image denoising network according to an embodiment;
FIG. 2 is a schematic structural diagram of a medium/low dose CT image denoising network according to an embodiment;
FIG. 3 is a flowchart illustrating training a low-dose CT image denoising network using a training data set according to an embodiment of the present invention;
FIG. 4 is a schematic structural diagram of a medium/low dose CT image denoising network according to a second embodiment;
FIG. 5 is a schematic structural diagram of a denoising network for a middle-low dose CT image according to a third embodiment;
FIG. 6 is a schematic structural diagram of a denoising network for a middle-low dose CT image according to the fourth embodiment;
FIG. 7 is a schematic structural diagram of a training system of a medium-low dose CT image denoising network according to the fifth embodiment;
FIG. 8 is a diagram of a processor and a computer storage medium according to a seventh embodiment.
Detailed Description
Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. This invention may, however, be embodied in many different forms and should not be construed as limited to the specific embodiments set forth herein. Rather, these embodiments are provided to explain the principles of the invention and its practical application to thereby enable others skilled in the art to understand the invention for various embodiments and with various modifications as are suited to the particular use contemplated. In the drawings, like reference numerals will be used to refer to like elements throughout.
The training method of the low-dose CT image denoising network provided by the application comprises the following steps:
the method comprises the steps of obtaining a training data set, wherein the training data set comprises a plurality of training image block groups, each training image block group comprises a first image block and a second image block, the first image block and the second image block are respectively a low-dose CT image and a standard-dose CT image block which are positioned at the same position, and the first image block and the second image block are equal in size.
Establishing a low-dose CT image denoising network, wherein the low-dose CT image denoising network comprises a first convolution layer, a convolution module, a first fusion layer and a second convolution layer which are sequentially connected, the first fusion layer is used for fusing an input signal of the convolution module and an output signal of the convolution module, the convolution module comprises at least one convolution network, and the at least one convolution network is sequentially connected, namely the first convolution layer, the at least one convolution network, the first fusion layer and the second convolution layer are connected in series; each convolution network comprises a channel layer, a third convolution layer and a second fusion layer which are sequentially connected, the channel layer comprises a first channel, the first channel comprises a fourth convolution layer and a first deconvolution layer, the fourth convolution layer and the first deconvolution layer are alternately connected, and the second fusion layer is used for fusing an input signal of the convolution network and an output signal of the third convolution layer.
And training the low-dose CT image denoising network by using the training data set to obtain an updated low-dose CT image denoising network.
According to the method and the device, the fourth convolution layer and the first deconvolution layer are alternately connected, so that information loss caused by continuous multiple convolution or continuous multiple deconvolution operation after continuous multiple convolution can be avoided, and the input of each convolution network can be fused with the output signal of the third convolution layer in the convolution network, so that the details of an original image can be better extracted, the problem of distortion after multiple cascading is avoided, and the reconstructed image is clearer.
The training method of the low-dose CT image denoising network in the present application is described in detail below with several specific embodiments and with reference to the accompanying drawings.
Example one
Referring to fig. 1-2, the training method of the low-dose CT image denoising network in the embodiment includes the steps of:
s1, acquiring a training data set, where the training data set includes a plurality of training image block groups, each of the training image block groups includes a first image block and a second image block, the first image block and the second image block in each of the training image block groups are respectively a low-dose CT image and a standard-dose CT image, the first image block and the second image block in the same training image block group have the same size, and the first image blocks in different training image block groups may have the same size or may not have the same size.
For example, the training data set in this embodiment is:
D＝{(x1,y1),(x1,y1),......,(xi,yi),......,(xn,yn)}，
wherein x isiRepresenting the first image block, y, of the ith training image block groupiRepresenting second image blocks in the ith training image block group, n representing the number of the training image block groups in the training data set, the first image blocks in the n training image block groups being different, i.e. the first image blocks in the n training image block groups being selected from n image blocks at different positions in the same low-dose CT image, and correspondingly, the second image blocks in the n training image block groups being selected from n image blocks at different positions in the same standard-dose CT image, of course, in order to obtain better training results, the first images in the n training image block groups may also be selected from the image blocks at the same position in n different low-dose CT images, and correspondingly, the second image blocks in the n training image block groups being selected from the image blocks at the same position in n different standard-dose CT images, and the first images in the n training image block groups may also be selected from the image blocks at different positions in different low-dose CT images, correspondingly, the second image blocks in the n training image block groups are selected from different positions in different standard dose CT images.
It should be noted that the low-dose CT image and the standard-dose CT image in the data set for training in the present embodiment are selected from an existing sample set, wherein the existing sample set is a common sample set in the art, and are not illustrated here.
S2, establishing a low-dose CT image denoising network, as shown in FIG. 2, wherein the low-dose CT image denoising network comprises a first convolution layer 1, a convolution module 2, a first fusion layer 3 and a second convolution layer 4 which are connected in sequence.
Specifically, the first fusion layer 3 is configured to fuse the input signal of the convolution module 2 and the output signal of the convolution module 2, so that the input signal of the second convolution layer 4 retains the feature information of the input signal of the convolution module 2, that is, retains the original low-level feature information, thereby being capable of better extracting the details of the original image.
The convolution module 2 comprises at least one convolution network 20, and the at least one convolution network 20 is connected in sequence, namely, the first convolution layer 1, the at least one convolution network 20, the first fusion layer 3 and the second convolution layer 4 are connected in series. Fig. 1 exemplarily shows a case where the convolution module 2 includes 3 convolution networks 20, and it should be noted that this is merely shown as an example and is not used to limit the present application. As shown in fig. 1, 3 convolutional networks 20 are connected in sequence, that is, 3 convolutional networks 20 are connected in series, and each convolutional network 20 includes a channel layer, a third convolutional layer 23, and a second fusion layer 24 which are connected in sequence. The second fusion layer 24 is configured to fuse the input signal of the channel layer and the output signal of the third convolution layer 23, and the first fusion layer 3 and the second fusion layer 24 in this embodiment are respectively configured to sum the input signal of the convolution module 2 and the output signal of the convolution module 2, and the input signal of the channel layer and the output signal of the third convolution layer 23.
The channel layer in this embodiment includes a first channel, the first channel includes a fourth convolution layer 21 and a first deconvolution layer 22, and the fourth convolution layer 21 and the first deconvolution layer 22 are alternately connected in sequence, where it should be noted that, if the number of the fourth convolution layer 21 and the first deconvolution layer 22 is one (as shown in fig. 2), the first channel is the fourth convolution layer 21 and the first deconvolution layer 22 connected in sequence, and if the number of the fourth convolution layer 21 and the first deconvolution layer 22 is multiple, the first channel is the fourth convolution layer 21, the first deconvolution layer 22, … …, the fourth convolution layer 21, and the first deconvolution layer 22 connected in sequence. The fourth convolution layer 21 is used for down-sampling, the first anti-convolution layer 22 is used for up-sampling, and the fourth convolution layer 21 and the first anti-convolution layer 22 are sequentially and alternately connected, that is, the down-sampling and the up-sampling are sequentially and alternately performed, so that information loss caused by continuous down-sampling for multiple times or continuous up-sampling for multiple times after continuous down-sampling for multiple times is avoided.
For example, in the embodiment, the size of the convolution kernel of the first convolution layer 1 is 3 x 3, the number of the convolution kernels is 64, and the first image block can be converted into 64 channels by the first convolution layer 1, where it should be noted that the first convolution layer 1 further includes an activation function, and after performing a convolution operation on the first image block, the data after the convolution operation needs to be subjected to nonlinear processing by the activation function.
The convolution kernel of the fourth convolution layer 21 has a size of 3 gamma 64, and the number of channels of the fourth convolution layer 21 may be other values, and the number of channels of the fourth convolution layer 21 is generally selected to be 2mFor example, the number of channels of the fourth convolutional layer 21 is 8, 16, 32, etc., and the convolution kernel of the first deconvolution layer 22 has a size of 3 x 64. Similarly, the fourth convolutional layer 21 further includes an activation function, and the fourth convolutional layer 21 performs nonlinear processing on the data after the convolution operation by the activation function.
Here, when the number of convolutional networks 20 is plural, the parameters of the fourth convolutional layer 21, the first deconvolution layer 22, and the third convolutional layer 24 in the plural convolutional networks 20 may be the same or different.
And S3, training the low-dose CT image denoising network by using the training data set to obtain an updated low-dose CT image denoising network.
Referring to fig. 3, specifically, in step S3, training the low-dose CT image denoising network with the training data set includes the steps of:
s31, combining the first image block x in the plurality of training image block groupsiAnd inputting the low-dose CT image denoising network to obtain a plurality of output images.
And S32, constructing a loss function according to the plurality of output images and the second image blocks in the plurality of training image block groups respectively.
Specifically, in step S32, the formula for constructing the loss function according to the plurality of output images and the second image blocks in the plurality of training image block groups is:
wherein theta represents the network parameter of the low-dose CT image denoising network, loss (theta) represents a loss function, n represents the number of training image block groups in the training data set, and G (X)i(ii) a θ) represents the ith output image, YiRepresenting the second image block in the ith training image block group.
The embodiment takes the absolute value difference as the loss function, so that the difference between each area of the image can be increased, and the boundary between each area in the image is clearer.
And S33, optimizing the minimum value of the loss function to obtain the optimized network parameters.
In step S33, an Adam optimization algorithm is used to optimize the minimum value of the loss function to obtain an optimized network parameter, and an iterative process of the Adam optimization algorithm is as follows:
calculating the gradient:
biased first moment estimation: s (k +1) ═ ρ1s(k)+(1-ρ1)g；
Biased second moment estimation: r (k +1) ═ p2r(k)+(1-ρ2)g⊙g；
Correcting the first moment:
correcting the second moment:
parameter correction value：
Updating the network parameters: θ + Δ θ;
and judging whether the iteration times are equal to the preset termination iteration times, if so, outputting the updated network parameter theta, and if not, continuing to perform the next iteration until the iteration times are equal to the preset termination iteration times. Preferably, the number of termination iterations in this embodiment is 1000, and the number of iterations may be set according to actual needs, which is not limited herein.
In the above optimization algorithm, the initial value conditions of the first iteration are initial network parameters θ, k is 0, s (k) is 0, and r (k) is 0;
representing the gradient operator, p1Is 0.9, p2The default value of (a) is 0.999, k is the iteration number, epsilon represents the learning rate, and the default value of epsilon is 0.0001; delta is a small constant, delta has a default value of 10-8。
And S34, updating the low-dose CT image denoising network by using the optimized network parameters.
The updated low-dose CT image denoising network obtained in the embodiment can avoid information loss caused by continuous multiple convolution or continuous multiple deconvolution operation after continuous multiple convolution, and the input of each convolution network can be fused with the output signal of the third convolution layer in the convolution network, so that the details of the original image can be better extracted, the problem of distortion after multiple cascading is avoided, and the reconstructed image is clearer.
Example two
Referring to fig. 4, the difference between the present embodiment and the first embodiment is that the first deconvolution layer 22 and the fourth convolution layer 21 in the first channel in the present embodiment are alternately connected in sequence, and here, it should be noted that the first channel is the first deconvolution layer 22 and the fourth convolution layer 21 connected in sequence if the number of the first deconvolution layer 22 and the fourth convolution layer 21 can be one, and the first channel is the first deconvolution layer 22, the fourth convolution layer 21, the … …, the first deconvolution layer 22, and the fourth convolution layer 21 connected in sequence if the number of the first deconvolution layer 22 and the fourth convolution layer 21 is multiple. The first deconvolution layer 22 is used for up-sampling, the fourth convolution layer 21 is used for down-sampling, and the first deconvolution layer 22 and the fourth convolution layer 21 are alternately connected in sequence, that is, up-sampling and down-sampling are alternately performed in sequence.
Another difference between this embodiment and the first embodiment is that the fourth convolutional layer 21 in this embodiment does not include an activation function, the first deconvolution layer 22 includes an activation function, and the first deconvolution layer 22 performs nonlinear processing on the data after the deconvolution operation by the activation function.
Compared with the first embodiment, the method and the device can also avoid information loss caused by continuous multiple down-sampling or continuous multiple up-sampling after continuous multiple down-sampling, avoid the problem of distortion after multiple cascading, and enable the reconstructed image to be clearer.
EXAMPLE III
Referring to fig. 5, the present embodiment is different from the first embodiment in that the channel layer in the present embodiment further includes a second channel and a splice layer 25. The second channel is connected in parallel with the first channel, the splicing layer 25 is connected between the channel layer and the third convolutional layer 23, the splicing layer 25 is used for splicing the output signal of the first channel and the output signal of the second channel, and the splicing layer 25 can splice the output signal of the first channel and the output signal of the second channel by adopting a plurality of splicing methods.
Specifically, the second channel includes fifth convolutional layers 26 and second convolutional layers 27, and the second convolutional layers 27 and the fifth convolutional layers 26 are alternately connected in this order. Here, if the number of the second deconvolution layers 27 and the number of the fifth convolution layers 26 are both one, the second channel is the second deconvolution layer 27 and the fifth convolution layer 26 which are connected in sequence, and if the number of the second deconvolution layer 27 and the fifth convolution layer 26 is both plural, the second channel is the second deconvolution layer 27, the fifth convolution layer 26, the second deconvolution layer 27, the fifth convolution layers 2622, … …, the second deconvolution layer 27, and the fifth convolution layer 26 which are connected in sequence. The second deconvolution layer 27 is used for up-sampling, the fifth convolution layer 26 is used for down-sampling, and the second deconvolution layer 27 and the fifth convolution layer 26 are sequentially and alternately connected, i.e. up-sampling and down-sampling are sequentially and alternately performed, so that information loss caused by continuous down-sampling for multiple times or continuous up-sampling for multiple times after continuous down-sampling for multiple times is avoided.
The second deconvolution layer 27 in this embodiment further includes an activation function, and the second deconvolution layer 27 performs nonlinear processing on the data after the deconvolution operation by the activation function. The convolution kernel of the second deconvolution layer 27 has a size of 3 x 64, and the number of channels of the second deconvolution layer 27 may be other values, and the number of channels of the second deconvolution layer 27 is generally selected to be 2mFor example, the number of channels of the second deconvolution layer 27 is 8, 16, 32, etc., and the convolution kernel of the fifth convolution layer 26 has a size of 3 x 64.
Here, when the number of the convolutional networks 20 is plural, the parameters of the fourth convolutional layer 21, the first deconvolution layer 22, and the third convolutional layer 24 in the plural convolutional networks 20 may be the same or different; the parameters of the second deconvolution layer 27 and the fifth convolution layer 26 in the plurality of convolutional networks 20 may be the same or different. Preferably, the weight of the first channel is different from the weight of the second channel.
The improvement of this embodiment over the first embodiment is that the channel layer in this embodiment further includes a second channel, the feature information of the image is extracted through different channels, and then the output information of the two channels is fused through the stitching layer 25, thereby further avoiding the lack of image information and further better extracting the details of the original image.
Example four
Referring to fig. 6, the present embodiment is different from the second embodiment in that the channel layer in the present embodiment further includes a second channel and a splice layer 25. The second channel is connected in parallel with the first channel, the splicing layer 25 is connected between the channel layer and the third convolutional layer 23, the splicing layer 25 is used for splicing the output signal of the first channel and the output signal of the second channel, and the splicing layer 25 can splice the output signal of the first channel and the output signal of the second channel by adopting a plurality of splicing methods.
Specifically, the second channel includes fifth convolutional layers 26 and second deconvolution layers 27, and the fifth convolutional layers 26 and the second deconvolution layers 27 are alternately connected in this order. Here, if the number of the fifth convolutional layer 26 and the second deconvolution layer 27 may be one, the second channel is the fifth convolutional layer 26 and the second deconvolution layer 27 connected in sequence, and if the number of the fifth convolutional layer 26 and the second deconvolution layer 27 is plural, the second channel is the fifth convolutional layer 26, the second deconvolution layer 27, … …, the fifth convolutional layer 26, and the second deconvolution layer 27 connected in sequence. The fifth convolution layer 26 is used for performing downsampling, the second deconvolution layer 27 is used for performing upsampling, and the fifth convolution layer 26 and the second deconvolution layer 27 are sequentially and alternately connected, that is, downsampling and upsampling are sequentially and alternately performed, so that information loss caused by continuous downsampling for multiple times or continuous upsampling for multiple times after continuous downsampling for multiple times is avoided.
The fifth convolutional layer 26 in this embodiment further includes an activation function, and the fifth convolutional layer 26 performs nonlinear processing on the data after the convolution operation by using the activation function. The convolution kernel of the fifth convolutional layer 26 has a size of 3 x 64, where the number of channels of the fifth convolutional layer 26 may also be other values, and the number of channels of the fifth convolutional layer 26 is generally selected to be 2mFor example, the number of channels of the fifth convolutional layer 26 is 8, 16, 32, etc., and the convolution kernel of the second deconvolution layer 27 has a size of 3 x 64.
Here, when the number of the convolutional networks 20 is plural, the parameters of the fourth convolutional layer 21, the first deconvolution layer 22, and the third convolutional layer 24 in the plural convolutional networks 20 may be the same or different; the parameters of the fifth convolutional layer 26 and the second convolutional layer 27 in the plurality of convolutional networks 20 may be the same or different. Preferably, the weight of the first channel is different from the weight of the second channel.
The improvement of this embodiment over the first embodiment is that the channel layer in this embodiment further includes a second channel, the feature information of the image is extracted through different channels, and then the output information of the two channels is fused through the stitching layer 25, thereby further avoiding the lack of image information and further better extracting the details of the original image.
EXAMPLE five
Referring to fig. 7, the embodiment provides a training system for a low-dose CT image denoising network, and the training system includes a training data set acquisition module 100, a network construction module 101, and a training module 102.
The training data set obtaining module 100 is configured to obtain a training data set, where the training data set includes a plurality of training image block groups, each training image block group includes a first image block and a second image block, the first image block and the second image block are respectively a low-dose CT image and a standard-dose CT image block and are located at the same position, the first image block and the second image block in the same training image block group have the same size, and the first image blocks in different training image block groups may have the same size or may not have the same size. For specific setting of the training data set, reference may be made to embodiment one, which is not described herein again.
The network construction module 101 is configured to establish a low-dose CT image denoising network, where the low-dose CT image denoising network includes a first convolution layer, a convolution module, a first fusion layer, and a second convolution layer, which are connected in sequence, the first fusion layer is used to fuse an input signal of the convolution module and an output signal of the convolution module, the convolution module includes at least one convolution network, and the at least one convolution network is connected in sequence, each convolution network includes a channel layer, a third convolution layer, and a second fusion layer, which are connected in sequence, the channel layer includes a first channel, the first channel includes a fourth convolution layer and a first deconvolution layer, the fourth convolution layer and the first deconvolution layer are alternately connected, and the second fusion layer is used to fuse an input signal of the convolution network and an output signal of the third convolution layer. The specific structure of the low-dose CT image denoising network may refer to implementation one to implementation four, which are not described herein again.
The training module 102 is configured to train the low-dose CT image denoising network with a training data set to obtain an updated low-dose CT image denoising network.
Specifically, the training module 102 includes an input unit, a loss function construction unit, an optimization unit, an update unit, and an output unit.
The input unit is used for inputting a first image block x in a plurality of training image block groupsiInputting a low-dose CT image denoising network to obtain a plurality of output images; an output unit for outputting a plurality of output images; the loss function constructing unit is used for constructing a loss function according to the plurality of output images and the second image blocks in the plurality of training image block groups respectively; the optimization unit is used for optimizing the minimum value of the loss function to obtain optimized network parameters; and the updating unit is used for updating the low-dose CT image denoising network by using the optimized network parameters.
Through the training system of the embodiment, information loss caused by continuous multiple convolution or continuous multiple deconvolution operation after continuous multiple convolution can be avoided, details of an original image can be better extracted, the problem of distortion after multiple cascading is avoided, and the reconstructed image is clearer.
EXAMPLE six
The embodiment provides a denoising method of a low-dose CT image, which comprises the following steps: inputting the low-dose CT image to be denoised into an updated low-dose CT image denoising network obtained by using the training method of the low-dose CT image denoising network described in the first to fourth embodiments, and obtaining the denoised low-dose CT image.
It should be noted that the denoising method in this embodiment includes two implementation manners, the first implementation manner is to use the low-dose CT image denoising network trained in the first to fourth embodiments as the denoising network of the low-dose CT image, and the low-dose CT image to be denoised is input to the low-dose CT image denoising network, so as to obtain the denoised low-dose CT image. In the second implementation manner, the low-dose CT image denoising network is trained by using the training method of the low-dose CT image denoising network described in the first to fourth embodiments, and then the low-dose CT image to be denoised is input into the trained low-dose CT image denoising network to obtain the denoised low-dose CT image.
By the denoising method, information loss caused by continuous multiple convolution or continuous multiple deconvolution operation after continuous multiple convolution can be avoided, details of an original image can be better extracted, the problem of distortion after multiple cascading is avoided, and the reconstructed image is clearer.
EXAMPLE seven
Referring to fig. 8, the present embodiment provides a processor 200, the processor 200 is connected to a computer storage medium 201, a computer program is stored in the computer storage medium 201, and the processor 200 is configured to read and execute the computer program stored in the computer storage medium 201, so as to implement the training method for denoising network of low-dose CT images according to the first to fourth embodiments.
In the above embodiments, the implementation may be wholly or partially realized by software, hardware, firmware, or any combination thereof. When implemented in software, may be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When loaded and executed on a computer, cause the processes or functions described in accordance with the embodiments of the invention to occur, in whole or in part. The computer may be a general purpose computer, a special purpose computer, a network of computers, or other programmable device. The computer instructions may be stored on a computer storage medium or transmitted from one computer storage medium to another, for example, from one website, computer, server, or data center to another website, computer, server, or data center via wire (e.g., coaxial cable, fiber optic, Digital Subscriber Line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.). The computer storage media may be any available media that can be accessed by a computer or a data storage device, such as a server, data center, etc., that incorporates one or more available media. The usable medium may be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., Solid State Disk (SSD)), among others.
Embodiments of the present invention are described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
The foregoing is directed to embodiments of the present application and it is noted that numerous modifications and adaptations may be made by those skilled in the art without departing from the principles of the present application and are intended to be within the scope of the present application.
Claims (10)
1. A training method of a low-dose CT image denoising network is characterized by comprising the following steps:
acquiring a training data set, wherein the training data set comprises a plurality of training image block groups, each training image block group comprises a first image block and a second image block, and the first image block and the second image block are respectively image blocks positioned at the same position in a low-dose CT image and a standard-dose CT image block;
establishing a low-dose CT image denoising network, wherein the low-dose CT image denoising network comprises a first convolution layer, a convolution module, a first fusion layer and a second convolution layer which are sequentially connected, the first fusion layer is used for fusing an input signal of the convolution module and an output signal of the convolution module, the convolution module comprises at least one convolution network, the at least one convolution network is sequentially connected, each convolution network comprises a channel layer, a third convolution layer and a second fusion layer which are sequentially connected, the channel layer comprises a first channel, the first channel comprises a fourth convolution layer and a first deconvolution layer, the fourth convolution layer and the first deconvolution layer are alternately connected, and the second fusion layer is used for fusing an input signal of the convolution network and an output signal of the third convolution layer;
and training the low-dose CT image denoising network by using the training data set to obtain an updated low-dose CT image denoising network.
2. The training method according to claim 1, wherein the fourth convolutional layer and the first deconvolution layer are alternately connected in sequence.
3. The training method of claim 1, wherein the first deconvolution layer and the fourth convolution layer are alternately connected in sequence.
4. The training method of claim 2, wherein the channel layer further comprises a second channel and a splice layer, the second channel being connected in parallel with the first channel, the splice layer being connected between the channel layer and the third convolutional layer, the splice layer being configured to splice an output signal of the first channel and an output signal of the second channel; the second channel comprises a fifth convolution layer and a second deconvolution layer, and the second deconvolution layer and the fifth convolution layer are sequentially and alternately connected.
5. The training method of claim 3, wherein the channel layer further comprises a second channel and a splice layer, the second channel being connected in parallel with the first channel, the splice layer being connected between the channel layer and the third convolutional layer, the splice layer being configured to splice an output signal of the first channel and an output signal of the second channel; the second channel comprises a fifth convolution layer and a second deconvolution layer, and the second deconvolution layer and the fifth convolution layer are sequentially and alternately connected.
6. A training method as recited in claim 1, wherein training the low-dose CT image using the training data specifically comprises:
inputting a first image block in the training image block groups into the low-dose CT image denoising network to obtain a plurality of output images;
constructing a loss function according to the plurality of output images and the second image blocks in the plurality of training image block groups respectively;
optimizing the minimum value of the loss function to obtain an optimized network parameter;
and updating the low-dose CT image denoising network by using the optimized network parameters.
7. Training method according to claim 6, characterized in that the loss function is:
where loss (θ) represents a loss function, n represents the number of training image block groups in the training data set, and G (X)i(ii) a θ) represents the ith output image, YiRepresenting the second image block in the ith training image block group.
8. A training system of a low-dose CT image denoising network is characterized by comprising:
the training data set acquisition module is used for acquiring a training data set, the training data set comprises a plurality of training image block groups, each training image block group comprises a first image block and a second image block, and the first image block and the second image block are respectively image blocks positioned at the same position in a low-dose CT image and a standard-dose CT image block;
the network construction module is used for establishing a low-dose CT image denoising network, the low-dose CT image denoising network comprises a first convolution layer, a convolution module, a first fusion layer and a second convolution layer which are connected in sequence, the first fusion layer is used for fusing an input signal of the convolution module and an output signal of the convolution module, the convolution module comprises at least one convolution network, the convolution networks are connected in sequence, each convolution network comprises a channel layer, a third convolution layer and a second fusion layer which are connected in sequence, the channel layer includes a first channel including a fourth convolutional layer and a first deconvolution layer, the fourth convolutional layer and the first deconvolution layer are alternately connected, and the second fusion layer is used for fusing an input signal of the convolutional network and an output signal of the third convolutional layer;
and the training module is used for training the low-dose CT image denoising network by using the training data set to obtain an updated low-dose CT image denoising network.
9. A denoising method of a low-dose CT image is characterized by comprising the following steps: inputting the low-dose CT image to be denoised into an updated low-dose CT image denoising network obtained by using the training method of the low-dose CT image denoising network according to any one of claims 1 to 7, and obtaining the denoised low-dose CT image.
10. A computer storage medium having stored thereon a computer program which, when read and executed by one or more processors, implements a method of training a low-dose CT image denoising network according to any one of claims 1-7.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911249569.0A CN110992290B (en) | 2019-12-09 | 2019-12-09 | Training method and system for low-dose CT image denoising network |
| PCT/CN2019/124377 WO2021114105A1 (en) | 2019-12-09 | 2019-12-10 | Training method and system for low-dose ct image denoising network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911249569.0A CN110992290B (en) | 2019-12-09 | 2019-12-09 | Training method and system for low-dose CT image denoising network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110992290A true CN110992290A (en) | 2020-04-10 |
| CN110992290B CN110992290B (en) | 2023-09-15 |
Family
ID=70091260
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911249569.0A Active CN110992290B (en) | 2019-12-09 | 2019-12-09 | Training method and system for low-dose CT image denoising network |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN110992290B (en) |
| WO (1) | WO2021114105A1 (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111968058A (en) * | 2020-08-25 | 2020-11-20 | 北京交通大学 | Low-dose CT image noise reduction method |
| CN112488951A (en) * | 2020-12-07 | 2021-03-12 | 深圳先进技术研究院 | Training method of low-dose image denoising network and denoising method of low-dose image |
| CN112541871A (en) * | 2020-12-07 | 2021-03-23 | 深圳先进技术研究院 | Training method of low-dose image denoising network and denoising method of low-dose image |
| CN113256752A (en) * | 2021-06-07 | 2021-08-13 | 太原理工大学 | Low-dose CT reconstruction method based on dual-domain interleaved network |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113744149A (en) * | 2021-08-31 | 2021-12-03 | 华中科技大学 | Deep learning post-processing method for solving low-dose CT image over-smoothness |
| CN114757928A (en) * | 2022-04-25 | 2022-07-15 | 东南大学 | One-step dual-energy finite angle CT reconstruction method based on deep training network |
| CN114742917B (en) * | 2022-04-25 | 2024-04-26 | 桂林电子科技大学 | CT image segmentation method based on convolutional neural network |
| CN116703772B (en) * | 2023-06-15 | 2024-03-15 | 山东财经大学 | Image denoising method, system and terminal based on adaptive interpolation algorithm |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108492269A (en) * | 2018-03-23 | 2018-09-04 | 西安电子科技大学 | Low-dose CT image de-noising method based on gradient canonical convolutional neural networks |
| CN108537824A (en) * | 2018-03-15 | 2018-09-14 | 上海交通大学 | Topological expansion method based on the enhancing of the alternately characteristic pattern of deconvolution and convolution |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108520203B (en) * | 2018-03-15 | 2021-08-20 | 上海交通大学 | Multi-target feature extraction method based on fusion of self-adaptive multi-peripheral frame and cross pooling feature |
| CN109685717A (en) * | 2018-12-14 | 2019-04-26 | 厦门理工学院 | Image super-resolution rebuilding method, device and electronic equipment |
| CN109886243B (en) * | 2019-03-01 | 2021-03-26 | 腾讯医疗健康(深圳)有限公司 | Image processing method, device, storage medium, equipment and system |
-
2019
- 2019-12-09 CN CN201911249569.0A patent/CN110992290B/en active Active
- 2019-12-10 WO PCT/CN2019/124377 patent/WO2021114105A1/en active Application Filing
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108537824A (en) * | 2018-03-15 | 2018-09-14 | 上海交通大学 | Topological expansion method based on the enhancing of the alternately characteristic pattern of deconvolution and convolution |
| CN108492269A (en) * | 2018-03-23 | 2018-09-04 | 西安电子科技大学 | Low-dose CT image de-noising method based on gradient canonical convolutional neural networks |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111968058A (en) * | 2020-08-25 | 2020-11-20 | 北京交通大学 | Low-dose CT image noise reduction method |
| CN111968058B (en) * | 2020-08-25 | 2023-08-04 | 北京交通大学 | Low-dose CT image noise reduction method |
| CN112488951A (en) * | 2020-12-07 | 2021-03-12 | 深圳先进技术研究院 | Training method of low-dose image denoising network and denoising method of low-dose image |
| CN112541871A (en) * | 2020-12-07 | 2021-03-23 | 深圳先进技术研究院 | Training method of low-dose image denoising network and denoising method of low-dose image |
| WO2022120883A1 (en) * | 2020-12-07 | 2022-06-16 | 深圳先进技术研究院 | Training method for low-dose image denoising network and denoising method for low-dose image |
| WO2022120694A1 (en) * | 2020-12-07 | 2022-06-16 | 深圳先进技术研究院 | Low-dose image denoising network training method and low-dose image denoising method |
| CN113256752A (en) * | 2021-06-07 | 2021-08-13 | 太原理工大学 | Low-dose CT reconstruction method based on dual-domain interleaved network |
Also Published As
| Publication number | Publication date |
|---|---|
| WO2021114105A1 (en) | 2021-06-17 |
| CN110992290B (en) | 2023-09-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110992290B (en) | Training method and system for low-dose CT image denoising network | |
| US20170372193A1 (en) | Image Correction Using A Deep Generative Machine-Learning Model | |
| US9235889B1 (en) | Systems, apparatus and methods for collecting and storing raw scan data and software for performing data processing, image reconstruction and interpretation | |
| Lee et al. | Machine friendly machine learning: interpretation of computed tomography without image reconstruction | |
| US20200402204A1 (en) | Medical imaging using neural networks | |
| CN112541871A (en) | Training method of low-dose image denoising network and denoising method of low-dose image | |
| JP2003038476A (en) | Computer-aided diagnostic system | |
| CN111709897B (en) | Domain transformation-based positron emission tomography image reconstruction method | |
| CN112489154A (en) | MRI motion artifact correction method for generating countermeasure network based on local optimization | |
| CN107909618A (en) | Image re-construction system and method | |
| Ote et al. | Deep-learning-based fast TOF-PET image reconstruction using direction information | |
| CN111815735A (en) | Human tissue self-adaptive CT reconstruction method and reconstruction system | |
| CN110246200B (en) | Magnetic resonance cardiac cine imaging method and device and magnetic resonance scanner | |
| Wang et al. | Improving generalizability in limited-angle CT reconstruction with sinogram extrapolation | |
| CN110874855B (en) | Collaborative imaging method and device, storage medium and collaborative imaging equipment | |
| Yu et al. | Image reconstruction for few-view computed tomography based on ℓ0 sparse regularization | |
| US9953440B2 (en) | Method for tomographic reconstruction | |
| US10249066B2 (en) | System and method for MPR streak reduction | |
| CN112581554A (en) | CT imaging method, device, storage equipment and medical imaging system | |
| Yoshida et al. | Motion correction in MR image for analysis of VSRAD using generative adversarial network | |
| CN117437152A (en) | PET iterative reconstruction method and system based on diffusion model | |
| Park et al. | Denoising of pediatric low dose abdominal CT using deep learning based algorithm | |
| CN114298920A (en) | Super-visual field CT image reconstruction model training and super-visual field CT image reconstruction method | |
| CN112488951B (en) | Training method of low-dose image denoising network and denoising method of low-dose image | |
| US20060115136A1 (en) | Image processing method and apparatus |
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 |