CN117095073A - Medical image denoising method and device based on deep learning - Google Patents
Medical image denoising method and device based on deep learning Download PDFInfo
- Publication number
- CN117095073A CN117095073A CN202311071728.9A CN202311071728A CN117095073A CN 117095073 A CN117095073 A CN 117095073A CN 202311071728 A CN202311071728 A CN 202311071728A CN 117095073 A CN117095073 A CN 117095073A
- Authority
- CN
- China
- Prior art keywords
- signal
- emi
- layer
- noise
- network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 61
- 238000013135 deep learning Methods 0.000 title claims abstract description 25
- 238000002595 magnetic resonance imaging Methods 0.000 claims description 95
- 238000012549 training Methods 0.000 claims description 68
- 238000013507 mapping Methods 0.000 claims description 65
- 230000006870 function Effects 0.000 claims description 40
- 230000006698 induction Effects 0.000 claims description 37
- 238000012545 processing Methods 0.000 claims description 31
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 claims description 18
- 238000013256 Gubra-Amylin NASH model Methods 0.000 claims description 15
- 238000005481 NMR spectroscopy Methods 0.000 claims description 13
- 238000010606 normalization Methods 0.000 claims description 12
- 238000013421 nuclear magnetic resonance imaging Methods 0.000 claims description 12
- 238000009826 distribution Methods 0.000 claims description 9
- 230000004913 activation Effects 0.000 claims description 7
- 238000012937 correction Methods 0.000 claims description 6
- 238000009827 uniform distribution Methods 0.000 claims description 3
- 238000012360 testing method Methods 0.000 abstract description 6
- 238000004458 analytical method Methods 0.000 abstract 12
- 238000004451 qualitative analysis Methods 0.000 abstract 1
- 230000000694 effects Effects 0.000 description 24
- 238000013527 convolutional neural network Methods 0.000 description 22
- 230000008569 process Effects 0.000 description 19
- 238000013461 design Methods 0.000 description 14
- 238000010586 diagram Methods 0.000 description 12
- 238000004590 computer program Methods 0.000 description 6
- 230000008030 elimination Effects 0.000 description 5
- 238000003379 elimination reaction Methods 0.000 description 5
- 238000004880 explosion Methods 0.000 description 5
- 238000003384 imaging method Methods 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 4
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000002708 enhancing effect Effects 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 238000009877 rendering Methods 0.000 description 2
- 208000028389 Nerve injury Diseases 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000008764 nerve damage Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000012805 post-processing Methods 0.000 description 1
- 230000002035 prolonged effect Effects 0.000 description 1
- 238000004904 shortening Methods 0.000 description 1
- 230000006641 stabilisation Effects 0.000 description 1
- 238000011105 stabilization Methods 0.000 description 1
Classifications
-
- 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
- G06T11/005—Specific pre-processing for tomographic reconstruction, e.g. calibration, source positioning, rebinning, scatter correction, retrospective gating
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0475—Generative networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
- G06V10/765—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects using rules for classification or partitioning the feature space
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Medical Informatics (AREA)
- Databases & Information Systems (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Magnetic Resonance Imaging Apparatus (AREA)
Abstract
The invention discloses a denoising method, a denoising device, electronic equipment and a denoising storage medium for medical images based on deep learning, which can directly utilize answer data of analysis test questions of an evaluator on a target item and standard answer data of the analysis test questions to quantify the weight of the evaluator on one hand; on the other hand, determining an alternative analysis result based on a plurality of initial analysis results, and realizing qualitative analysis of deviation between the initial result and the alternative analysis result based on a decision deviation value between the initial analysis result and the alternative analysis result; finally, determining the optimal analysis result of the target item by utilizing the negative correlation relationship of the weight and the deviation; therefore, the method breaks away from the framework of the average method, abandons the consensus principle, and can conduct decision analysis by taking the optimal decision as the target guide, so that the optimal analysis result can be determined from a plurality of initial analysis results of the project, and the accuracy of the analysis of the project result is improved.
Description
Technical Field
The invention belongs to the technical field of image processing, and particularly relates to a denoising method and device for medical images based on deep learning.
Background
In recent years, portable Magnetic Resonance Imaging (MRI) technology has proven to be capable of bedside imaging beside an intensive care unit or a mobile stroke unit in nerve injury assessment, and thus has found wide application, and at the same time, these scanners are low in cost and high in portability, so that hardware support is provided for the development of point-of-care (POC) services.
However, the amplitude of the nuclear magnetic resonance signal at a low magnetic field is very weak compared to a high magnetic field, and thus, for the convenience of diagnosis by doctors, improving image quality and pursuing a higher signal-to-noise ratio (SNR) have become an unavoidable problem; in the nuclear magnetic resonance scanning, an image with higher signal-to-noise ratio can be obtained by increasing the average frequency of signals, but the scanning time is prolonged, so that a patient feels uncomfortable, and the motion artifact phenomenon is aggravated by the movement of the patient, so that the method is not suitable for the field of nuclear magnetic resonance image denoising.
Meanwhile, the image post-processing technology does not influence the scanning time, so that the method can become a potential competitor for removing irrelevant noise in the ultralow-field MRI image; the deep learning method is used as a tool with great prospect, and is still rarely applied to an ultralow-field MRI scanner; most deep learning EMI removal algorithms applied to ULF MRI scanners so far mainly use CNN networks to learn the mapping relation between the EMI signals detected by an EMI sensing coil and the EMI signals detected by an MRI receiving coil, and then apply a trained model to the scan, so as to finish the denoising processing of the MRI signals scanned at the time, and when the next scan is needed, the data is needed to be acquired again to train the network, so that the training time of the network influences the imaging time; however, as the fitting capacity of the CNN network is poor and the number of cycles required for fitting is large, the CNN network has the problems of poor denoising effect and low efficiency; based on this, how to provide a medical image denoising method with good denoising effect and high efficiency has become a problem to be solved.
Disclosure of Invention
The invention aims to provide a denoising method and device for medical images based on deep learning, which are used for solving the problems of poor effect and low efficiency existing in the prior art of denoising images by adopting a CNN network.
In order to achieve the above purpose, the present invention adopts the following technical scheme:
in a first aspect, a denoising method for medical images based on deep learning is provided, including:
acquiring an MRI signal and a first EMI signal, wherein the first EMI signal is a noise signal detected by an EMI induction coil positioned in front of an MRI scanner;
acquiring a noise mapping model, wherein the noise mapping model is a trained GAN model or a trained residual error network, the noise mapping model is used for determining a mapping relation between a second EMI signal detected by the EMI induction coil and a target noise signal received by an MRI scanner, and the target noise signal is generated by interference of the second EMI signal;
inputting the first EMI signal into a noise mapping model to obtain a target noise signal corresponding to the first EMI signal;
obtaining a denoised nuclear magnetic resonance imaging signal according to the target noise signal corresponding to the MRI signal and the first EMI signal;
And carrying out image reconstruction processing based on the denoised nuclear magnetic resonance imaging signals so as to obtain denoised nuclear magnetic resonance images after the image reconstruction processing.
Based on the above disclosure, the present invention learns the mapping relationship between the EMI signal detected by the EMI induction coil and the EMI signal received by the MRI scanner by using the GAN model or the residual network in advance, so that the noise signal generated in the MRI signal corresponding to the MRI scanner under the interference of the electromagnetic interference signal is obtained; based on the above, in practical application, inputting the first EMI signal obtained in practice into the trained GAN model or the trained residual network, so as to obtain a noise signal in the MRI signal; then, subtracting the noise signal from the MRI signal to finish denoising the MRI signal; finally, performing image reconstruction by using the denoised MRI signals to obtain denoised nuclear magnetic resonance images; therefore, the invention uses the GAN network or the residual error network to replace the CNN network to realize the prediction and removal of the EMI noise signal in the MRI signal, compared with the traditional CNN network, the GAN has stronger fitting effect and shorter fitting period, and the residual error network adopts residual error connection, thereby effectively solving the problems of gradient elimination and explosion in the training process, and also obtaining better fitting capacity while improving the network convergence speed.
In one possible design, the noise mapping model is trained as follows:
acquiring a training data set, wherein the training data set comprises a plurality of second EMI signals and real target noise signals corresponding to the second EMI signals, and the real target noise signals corresponding to any one of the second EMI signals refer to electromagnetic interference signals generated by interference of any one of the second EMI signals received by the MRI scanner;
taking each real target noise signal as tag data;
and training a GAN network or a residual network by taking each second EMI signal and corresponding label data in the training data set as input and the target noise signal corresponding to each second EMI signal as output so as to obtain the noise mapping model after training is completed.
In one possible design, when the noise mapping model is the trained GAN model, the noise mapping model includes a generator and a discriminator, where the generator is configured to extract characteristic information of the input second EMI signal during training, and generate a target noise signal corresponding to the input second EMI signal based on the characteristic information;
the discriminator is used for carrying out two-classification on the output result of the generator during training, and updating the network parameters of the generator according to the two-classification result.
In one possible design, the generator includes four first residual blocks and one first output layer, where the four first residual blocks are sequentially arranged according to a signal processing direction, each first residual block includes a first convolution structure layer, a first batch normalization layer, and a first correction linearity layer, the first output layer includes a first convolution layer, and an activation function used by the first output layer is a hyperbolic tangent function;
the discriminator comprises a second convolution layer, a first LeakyReLu layer, a first Dropout layer, a third convolution layer, a second LeakyReLu layer, a second Dropout layer and a second batch normalization layer which are sequentially arranged according to the signal processing direction, and the discriminator is used for carrying out secondary classification on the output result of the generator based on the second convolution layer, the first LeakyReLu layer, the first Dropout layer, the third convolution layer, the second LeakyReLu layer, the second Dropout layer and the second batch normalization layer so as to output a classification result.
In one possible design, the first convolution structure layer includes nine fourth convolution layers, where the nine fourth convolution layers use convolution kernels having dimensions of 11×11, 9×9, 5×5, 1×1, and 1×1 in order.
In one possible design, the loss function of the generator is:
in the above formula (1), L G Represents the loss function of the generator, x represents the second EMI signal of the input, x z The true target noise signal corresponding to the inputted second EMI signal,representing the distribution of the second EMI signal detected by the EMI induction coil, G () representing the generator, D () representing the discriminator, MSE () representing the mean square error;
correspondingly, the loss function of the arbiter is:
in the above formula (2), L D Representing the loss function of the arbiter,representing the distribution of the second EMI signal received by said MRI scanner corresponding to the real target noise signal,/I>The gradient is represented by a gradient, lambda represents the gradient penalty coefficient, I 2 Represents a two-norm and +.>Representing x and x z Is a first linear combination of->Representation->Distribution of (1), and->Expressed as:
in the above formula (3), ε represents x and x z And U represents a uniform distribution.
In one possible design, when the noise mapping model is the trained residual network, the noise mapping model includes four second residual blocks and a second output layer, where each second residual block includes a second convolution structure layer, a third batch normalization layer, and a second correction linearity layer, the second output layer includes a fifth convolution layer, and an activation function used by the second output layer is a hyperbolic tangent function.
In one possible design, the second convolution structure layer includes nine sixth convolution layers, and the convolution kernels used by the nine sixth convolution layers are 11×11, 9×9, 5×5, 1×1, and 1×1 in order;
correspondingly, the loss function of the noise mapping model is as follows:
L=MSE(x,x z ) (4)
in the above formula (4), L represents a loss function of the noise mapping model, x represents the second EMI signal inputted, x z The MSE () represents the mean square error of the true target noise signal corresponding to the second EMI signal input.
In one possible design, two EMI induction coils are provided in front of the MRI scanner, wherein the two EMI induction coils are placed orthogonally in front of the MRI scanner along the slice selection gradient direction of the MRI scanner.
In a second aspect, there is provided a denoising apparatus for medical image based on deep learning, comprising:
an acquisition unit configured to acquire an MRI signal and a first EMI signal, wherein the first EMI signal is a noise signal detected by an EMI induction coil located in front of an MRI scanner;
the acquisition unit is further used for acquiring a noise mapping model, wherein the noise mapping model is a trained GAN model or a trained residual error network, the noise mapping model is used for determining a mapping relation between a second EMI signal detected by the EMI induction coil and a target noise signal received by an MRI scanner, and the target noise signal is generated by interference of the second EMI signal;
The denoising unit is used for inputting the first EMI signal into a noise mapping model so as to obtain a target noise signal corresponding to the first EMI signal;
the denoising unit is also used for obtaining a denoised nuclear magnetic resonance imaging signal according to the target noise signal corresponding to the MRI signal and the first EMI signal;
and the image reconstruction unit is used for carrying out image reconstruction processing based on the denoised nuclear magnetic resonance imaging signals so as to obtain denoised nuclear magnetic resonance images after the image reconstruction processing.
In a third aspect, another denoising apparatus for medical image based on deep learning is provided, taking an apparatus as an electronic device, and the apparatus includes a memory, a processor and a transceiver, which are sequentially communicatively connected, where the memory is used to store a computer program, the transceiver is used to send and receive a message, and the processor is used to read the computer program, and execute the denoising method for medical image based on deep learning, where the denoising method is as in the first aspect or any one of the possible designs of the first aspect.
In a fourth aspect, a storage medium is provided, on which instructions are stored which, when run on a computer, perform the denoising method of a deep learning-based medical image as in the first aspect or any one of the possible designs of the first aspect.
In a fifth aspect, there is provided a computer program product comprising instructions which, when run on a computer, cause the computer to perform the method of denoising of medical images based on deep learning as may be devised in the first aspect or any one of the first aspects.
The beneficial effects are that:
(1) According to the invention, the GAN network or the residual error network is used for replacing the CNN network to realize the prediction and removal of the EMI noise signal in the MRI signal, compared with the traditional CNN network, the GAN has a stronger fitting effect and a shorter fitting period, the residual error network adopts residual error connection, so that the gradient elimination and explosion problems in the training process can be effectively solved, and the better fitting capacity can be obtained while the network convergence speed is improved, therefore, the invention can obtain a better denoising effect while improving the denoising efficiency no matter whether the GAN network or the residual error network is used for the prediction and removal of the EMI signal, and is suitable for large-scale application and popularization in the field of nuclear magnetic resonance image denoising.
(2) Compared with the existing GAN network, the GAN network provided by the invention adopts a lighter model structure, and the generator only comprises four residual blocks; therefore, the invention reduces the number of layers of the network, and is beneficial to improving training and convergence efficiency; similarly, the residual error network also samples a lightweight structure, and a convolution kernel different from the traditional residual error network is used, so that the model can be more suitable for the training task requirement while the convergence time is reduced during training, and the training precision is improved.
(3) When the GAN network is used as a noise mapping model, the invention combines the loss function of the WGAN and the loss function of the Mean Square Error (MSE), and designs the loss function which can stably train the GAN network, thus the EMI prediction and removal tasks can be effectively realized, and the training stability and the denoising effect are improved.
(4) Compared with the number of 4-10 coils used by the existing algorithm, the invention only uses two EMI sensing coils to collect data, thus reducing hardware cost.
Drawings
Fig. 1 is a schematic flow chart of a step of a denoising method for medical image based on deep learning according to an embodiment of the present invention;
fig. 2 is a schematic layout diagram of an EMI induction coil according to an embodiment of the present invention;
fig. 3 is a schematic structural diagram of an improved GAN network according to an embodiment of the present invention;
fig. 4 is a schematic structural diagram of an improved residual network according to an embodiment of the present invention;
fig. 5 is a schematic diagram of a loss curve of an improved GAN network according to an embodiment of the present invention;
fig. 6 is a schematic diagram of an effect of image denoising using an improved GAN network according to an embodiment of the present invention;
fig. 7 is a schematic diagram of a loss curve of an improved residual network and a conventional CNN network according to an embodiment of the present invention;
Fig. 8 is a comparison diagram of image denoising between an improved residual network and a conventional CNN network according to an embodiment of the present invention;
fig. 9 is a schematic structural diagram of a denoising device based on a medical image for deep learning according to an embodiment of the present invention;
fig. 10 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.
Detailed Description
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the present invention will be briefly described below with reference to the accompanying drawings and the description of the embodiments or the prior art, and it is obvious that the following description of the structure of the drawings is only some embodiments of the present invention, and other drawings can be obtained according to these drawings without inventive effort to a person skilled in the art. It should be noted that the description of these examples is for aiding in understanding the present invention, but is not intended to limit the present invention.
It will be understood that, although the terms first, second, etc. may be used herein to describe various elements, these elements should not be limited by these terms. These terms are only used to distinguish one element from another element. For example, a first element could be termed a second element, and, similarly, a second element could be termed a first element, without departing from the scope of example embodiments of the present invention.
It should be understood that for the term "and/or" that may appear herein, it is merely one association relationship that describes an associated object, meaning that there may be three relationships, e.g., a and/or B, may represent: a alone, B alone, and both a and B; for the term "/and" that may appear herein, which is descriptive of another associative object relationship, it means that there may be two relationships, e.g., a/and B, it may be expressed that: a alone, a alone and B alone; in addition, for the character "/" that may appear herein, it is generally indicated that the context associated object is an "or" relationship.
Examples:
referring to fig. 1, in the denoising method based on deep learning provided by the embodiment, a GAN network or a residual network is used to replace a traditional CNN network to realize the prediction and removal of EMI signals in MRI signals, wherein compared with the traditional CNN network, the GAN network has a stronger fitting effect and a shorter fitting period, and can exert a better effect in the task of EMI prediction and removal, while the residual network adopts residual connection, so that the problem of gradient elimination and explosion in the training process can be effectively solved, and the better fitting capability can be obtained while the network convergence speed is improved; therefore, the method can obtain better denoising effect while improving denoising efficiency without adopting a GAN network or a residual network; in this embodiment, the method may be, for example, but not limited to, running on the side of the medical image denoising end, alternatively, the medical image denoising end may be, but not limited to, a personal computer (personal computer, PC), and it is understood that the foregoing execution subject does not constitute limitation of the embodiment of the present application, and accordingly, the running steps of the method may be, but not limited to, those shown in the following steps S1 to S5.
S1, acquiring an MRI signal and a first EMI signal, wherein the first EMI signal is a noise signal detected by an EMI induction coil positioned in front of an MRI scanner; in this embodiment, the number of EMI induction coils can be, but is not limited to, 2, that is, 2 EMI induction coils are disposed in front of the MRI scanner for detecting electromagnetic interference around the MRI scanner, and meanwhile, referring to fig. 2, two EMI induction coils are disposed in front of the MRI scanner orthogonally along a layer selection gradient direction (that is, a z-axis direction in fig. 2) of the MRI scanner; therefore, by adopting the placement structure, on one hand, the EMI signal detected by the EMI induction coil can be better associated with the noise signal received by the MRI scanner (namely the EMI signal received by the MRI scanner, namely the target noise signal described below), so that the subsequent GAN network can conveniently learn the mapping relation of the EMI signal and the noise signal; on the other hand, compared with the traditional method of using 4-10 coils to induce the EMI signals, the invention only needs 2 induction coils, thereby greatly reducing the denoising cost.
In addition, for example, when in actual use, the MRI scanner is set to be in a radio frequency on state, and in the state, a first EMI signal detected by an EMI induction coil in front of the MRI scanner and an MRI signal generated by the MRI scanner are acquired; then, denoising of the nuclear magnetic resonance image can be completed based on the two signals.
When the MRI scanner is actually used, the MRI signals generated by the MRI scanner work are acquired, and after the first EMI signals detected by the EMI induction coil in front of the MRI scanner are acquired, a pre-trained noise mapping module can be acquired to predict target noise signals (namely, the EMI signals received by the MRI scanner) under the interference of the first EMI signals based on the first EMI signals detected by the EMI induction coil; the specific process of obtaining the noise mapping model may be, but is not limited to, as shown in step S2 below.
S2, acquiring a noise mapping model, wherein the noise mapping model is a trained GAN model or a trained residual error network, the noise mapping model is used for determining a mapping relation between a second EMI signal detected by the EMI induction coil and a target noise signal received by the MRI scanner, and the target noise signal is generated by interference of the second EMI signal; in this embodiment, the noise mapping model is essentially trained by taking the EMI signal detected by the EMI induction coil as input and the EMI signal received by the MRI scanner as output; in popular terms, a GAN model or a residual network is used to learn the mapping relationship between the EMI signals detected by the EMI induction coil and the MRI scanner, so as to predict the EMI signals included in the signals of the MRI scanner when performing MRI based on the EMI signals detected by the EMI induction coil, thereby denoising the MRI signals based on the predicted EMI signals.
Alternatively, the embodiment may use a trained GAN model or a residual network as the noise mapping model; in specific application, the GAN model and the residual network are improved based on the existing network, and the following descriptions are respectively provided:
first, the present embodiment improves the network architecture based on the conventional GAN network, as follows:
in particular applications, when using a trained GAN model as the noise mapping model, an example modified GAN network (i.e., noise mapping model) may include, but is not limited to: a generator and a arbiter; referring to fig. 3, the generator may include, for example, but not limited to, four first residual blocks and one first output layer sequentially arranged according to a signal processing direction (i.e., a direction indicated by left-to-right arrow in fig. 3), wherein each first residual block includes a first convolution structure layer, a first batch normalization layer, and a first correction linearity layer (two blocks connected by each jump arrow in fig. 3 represent one first residual block), the first output layer includes the first convolution layer, and an activation function used by the first output layer is a hyperbolic tangent function; in this embodiment, the generator is mainly configured to perform feature extraction on an input signal, and perform feature learning based on the extracted feature information, so as to obtain an output signal (i.e., a mapping relationship between an EMI signal detected by the EMI induction coil and an EMI signal detected by the MRI scan signal) corresponding to the input signal.
Meanwhile, the discriminator includes a second convolution layer, a first LeakyReLu layer, a first Dropout layer, a third convolution layer, a second LeakyReLu layer, a second Dropout layer and a second batch normalization layer which are sequentially arranged according to the signal processing direction, wherein, referring to fig. 3, a first arrow in fig. 3 represents the second convolution layer, the first LeakyReLu layer and the first Dropout layer, each arrow in the future represents four structural layers in the future, and a last arrow represents the output; in addition, in the embodiment, the discriminator is mainly used in training, and is used for constraining the output of the generator to approach to real data, so that training of a model is more reliable, and the output data is more real.
Further, in this embodiment, the first convolution structure layer includes nine fourth convolution layers, where the sizes of convolution kernels used by the nine fourth convolution layers are 11×11, 9×9, 5×5, 1×1, and 1×1 in order; therefore, through the design, on one hand, the GAN network provided by the embodiment reduces the number of layers of the network, so that the convergence of the model can be facilitated during training, and the training efficiency is improved; on the other hand, when the convolution processing in the generator is performed, different convolution kernels are used, and based on the convolution kernels, model learning tasks (namely, the mapping relation between the EMI signals detected by the EMI induction coil and the EMI signals received by the MRI scanner) can be better met, so that accuracy of predicting the EMI signals mingled in the MRI signals is improved.
After completing the foregoing description of the specific architecture of the GAN network, the following disclosure discloses one of the training manners of the GAN network, which may be, but not limited to, the following first, second, and third steps:
the first step: acquiring a training data set, wherein the training data set comprises a plurality of second EMI signals and real target noise signals corresponding to the second EMI signals, and the real target noise signals corresponding to any one of the second EMI signals refer to electromagnetic interference signals generated by interference of any one of the second EMI signals received by the MRI scanner; in this embodiment, the training data may be acquired by using a 3D gradient echo (GRE) sequence with a null-picking EMI noise function, for example, to obtain a more accurate data set, and provide a more reliable data base for training of the model.
Further, for example, data of the MRI scanner when the RF (radio frequency) is turned on and off may be collected during a TR (repet it ion t ime) time period, wherein each second EMI signal of the EMI induction coil collected when the radio frequency is turned off, and a corresponding EMI signal (i.e., a real target noise signal) received by the MRI scanner is used as the training data, and the collected data is used as the test data when the radio frequency is turned on.
In this embodiment, since the second EMI signals detected by the EMI induction coil in front of the MRI scanner are acquired at one TR time when the MRI scanner is turned on and off, and the EMI signals received by the MRI scanner itself interfered by the second EMI signals (i.e., real target noise signals); thus, the second EMI signal in the radio frequency off state can be used to train the GAN model, so that after model training is completed, when the radio frequency is turned on, the target noise signal corresponding to the second EMI signal is tested (i.e., the second EMI signal in the radio frequency on state is used as input to obtain the EMI signal received by the MRI scanner due to interference thereof).
Optionally, the training process is as shown in the second and third steps below.
And a second step of: taking each real target noise signal as tag data; in this embodiment, the MRI scanner receives the electromagnetic interference signal (i.e., the received EMI signal itself, that is, the real target noise signal) generated by the second EMI signal whenever the EMI induction coil detects the second EMI signal, so that the real noise target signal is a real value to perform model training; wherein the specific training process is shown in the following third step.
And a third step of: taking each second EMI signal and corresponding label data in the training data set as input, taking a target noise signal corresponding to each second EMI signal as output, training a GAN (gas insulated network) to obtain the noise mapping model after training is completed; in this embodiment, the FE (frequency encod ing) lines input and l abel are used for example training and testing to obtain better training effect.
Meanwhile, the training process is described by combining the specific architecture of the GAN network; the generator is used for extracting characteristic information of the input second EMI signal during training, generating a target noise signal corresponding to the input second EMI signal based on the characteristic information, namely continuously extracting the characteristic through four first residual blocks (wherein the first residual block convolves an image from a single channel to 128 channels, the remaining first residual blocks gradually restore the image to the single channel), and finally outputting a network through a first convolution layer and an activation function in a first output layer to obtain the target noise signal corresponding to the input second EMI signal, wherein the target noise signal refers to an EMI signal which is predicted by a model and is received at an MRI scanner end, and the signal is electromagnetic interference noise in the MRI signal; meanwhile, the discriminator is used for carrying out two-class on the output result of the generator during training, updating the network parameters of the generator according to the two-class result (namely, the forward propagation process convolves the image from a single channel to 128 channels, then maps the image into one dimension, and finally outputs the two-class result value); specifically, based on the second convolution layer, the first LeakyReLu layer, the first Dropout layer, the third convolution layer, the second LeakyReLu layer, the second Dropout layer and the second batch normalization layer, the output result of the generator is subjected to two-classification so as to output a classification result; furthermore, in the GAN network, the results generated by the generator and the arbiter are in opposition relation, and the optimal parameters are required to be reduced through the gradient, namely, when the arbiter cannot restrict updating the generator based on the two classification results, the output result of the generator can reach the required precision; thus, model training can be assisted based on the discriminators, so that training is more stable.
In addition, for the stabilization of training and to accommodate EMI prediction and removal tasks, the present embodiment combines the WGAN loss function and the MSE loss function, specifically, the generator loss function is shown in the following equation (1):
in the above formula (1), L G Represents the loss function of the generator, x represents the second EMI signal of the input, x z The true target noise signal corresponding to the inputted second EMI signal,representing the distribution of this second EMI signal detected by the EMI induction coil, G () represents the generator, D () represents the discriminator, and MSE () represents the mean square error.
Correspondingly, the loss function of the arbiter is:
in the above formula (2), L D Representation discriminatorIs used for the loss function of (a),representing the distribution of the second EMI signal received by said MRI scanner corresponding to the real target noise signal,/I>The gradient is represented by a gradient, lambda represents the gradient penalty coefficient, I 2 Represents a two-norm and +.>Representing x and x z Is a first linear combination of->Representation->Distribution of (1), and->Expressed as:
in the above formula (3), ε represents x and x z And U represents a uniform distribution.
Secondly, after the description of the specific architecture and training process of the GAN network is completed, the following describes the specific network structure when the noise mapping model adopts the trained residual network:
In this embodiment, the improved residual network is consistent with the structure of the generator in the GAN network, as shown in fig. 4, and each includes four second residual blocks and one second output layer (for distinguishing from the generator, naming by different names) sequentially set according to the signal processing direction, where each second residual block includes a second convolution structure layer, a third batch normalization layer and a second correction linearity layer, the second output layer includes a fifth convolution layer, and an activation function used by the second output layer is also a hyperbolic tangent function; in a specific application, the improved residual network plays the same role as the generator in training, and is used for extracting features of the input second EMI signal, and performing feature learning based on the extracted feature information, so as to obtain an output signal (i.e. a target noise signal) corresponding to the input signal.
Meanwhile, the second convolution structure layer has the same structure as the first convolution structure layer and also comprises nine sixth convolution layers, and the sizes of convolution kernels used by the nine sixth convolution layers are 11×11, 9×9, 5×5, 1×1 and 1×1 in sequence; furthermore, the training process of the improved residual network is the same as the training process of the GAN network, and will not be described again.
Finally, the loss function of the residual network after the improvement is exemplified as follows:
L=MSE(x,x z ) (4)
in the above formula (4), L represents a loss function of the noise mapping model (i.e., the modified residual network), x represents the input second EMI signal, x z The MSE () represents the mean square error of the true target noise signal corresponding to the second EMI signal input.
Based on the foregoing explanation, the noise mapping model provided in this embodiment can achieve the capability of immediate imaging by reducing the training time of the network, so as to solve the problem of slow efficiency of the conventional technology due to long training period of the model when each time is used; meanwhile, by adopting the model, a better fitting effect can be obtained during training, and based on the model, the denoising effect can be improved during actual use.
Therefore, through the detailed description of the GAN network structure, the network structure of the residual error network, the training process and the loss function thereof, a noise mapping model which can be suitable for the prediction and removal of the EMI signal can be constructed from the network architecture and the loss function; then, in actual use, the model can be used to predict the received EMI signal (i.e., the target noise signal corresponding to the first EMI signal) of the MRI scanner under the interference of the first EMI signal; finally, denoising the signal based on the original MRI signal and the predicted target noise signal; the denoising process is as follows in steps S3 to S5.
S3, inputting the first EMI signal into a noise mapping model to obtain a target noise signal corresponding to the first EMI signal; in this embodiment, the noise mapping model is equivalent to predicting an electromagnetic interference signal in an MRI signal generated by the MRI scanner, which is affected by the first EMI signal; the target noise signal output by the model is the electromagnetic interference signal in the MRI signal; after the electromagnetic interference signal in the MRI signal is predicted, denoising processing may be performed as shown in step S4.
S4, obtaining a denoised nuclear magnetic resonance imaging signal according to the target noise signal corresponding to the MRI signal and the first EMI signal; in this embodiment, the target noise signal corresponding to the first EMI signal is subtracted from the MRI signal (certainly, subtraction is performed in the K space, that is, subtraction is performed in the fourier space), so as to obtain the denoised MRI signal.
After the denoising of the MRI signals is completed, image reconstruction can be carried out, so that a denoised nuclear magnetic resonance image is obtained; wherein the reconstruction process is shown in step S5 below.
S5, performing image reconstruction processing based on the denoised nuclear magnetic resonance imaging signals to obtain denoised nuclear magnetic resonance images after the image reconstruction processing; in this embodiment, for example, but not limited to, based on the denoised nmr imaging signal, an inverse fourier transform algorithm is adopted to implement image reconstruction, so as to obtain a denoised nmr image; of course, the foregoing inverse fourier transform algorithm is a common method for image reconstruction, and the principle thereof is not described in detail.
According to the deep learning-based medical image denoising method described in detail in the steps S1-S5, the GAN network or the secondary residual error network is used for replacing the traditional CNN network to realize the prediction and removal of the EMI signals in the MRI signals, wherein compared with the traditional CNN network, the GAN network has a stronger fitting effect and a shorter fitting period, and can play a better role in the EMI prediction and removal task, and the residual error network adopts residual error connection, so that the gradient elimination and explosion problems in the training process can be effectively solved, and the network convergence speed can be improved, and meanwhile, the better fitting capability can be obtained; therefore, the method can reduce the training time of the model in each use no matter the GAN network or the residual network is adopted, so that the denoising efficiency is improved, and meanwhile, a better denoising effect is obtained, and the method is more suitable for large-scale application and popularization in the field of medical image denoising.
In one possible design, the second aspect of the present embodiment provides an example graph using the denoising method described in the first aspect of the present embodiment, so as to illustrate the effectiveness of the GAN network provided in the present invention in enhancing the network fitting capability and the denoising effect in place of the conventional CNN network.
In this embodiment Adam is selected as an optimizer, and the super parameters of the network training are set as follows: learning rate α=3.0×e-4 and two exponential decay factors β 1 =0.5 and β 2 Batch sizes for training and testing were set to 8 and 1, respectively, with training of the network being performed on a single NVIDIA GeForce RTX 3050 notebook GPU, fig. 5 shows a graph of Loss (i.e. showing the relationship between Loss and batch number), and "Convergence Loss" in fig. 5 corresponds to the average after curve Convergence; meanwhile, fig. 6 shows the results of imaging after removing EMI signals in MRI signals using the GAN network provided in the present embodiment, the reconstructed image of the original data is shown above each result, and the reconstructed image after removing EMI is shown below each result; it can be seen that the model converges around batch 100 and the GAN network works well in removing EMI noise from the raw data and reconstructed image after EMI removal.
In one possible design, the third aspect of the present embodiment provides an example graph using the denoising method according to the first aspect of the present embodiment, so as to illustrate the effectiveness of the residual network provided by the present invention in replacing the conventional CNN network in enhancing the network fitting capability, reducing the training time, and denoising effect.
To illustrate the effectiveness of replacing CNN with a residual network in accelerating network convergence and shortening EMI prediction and cancellation time, we calculated various loss functions with CNN network and residual network on 2048 pairs of FE lines, respectively.
We selected Adam as the optimizer, both networks trained using the same hyper parameters: learning rate α=3.0×e-4 and two exponential decay factors β 1 =0.5 and β 2 Batch sizes for training and testing were set to 8 and 1, respectively, =0.999, training for both networks was performed on a single NVIDIA GeForce RTX 3050 notebook GPU.
To better visualize the process of the Loss curves of the two networks from the beginning of training to Convergence, the training period settings of the two models are different, fig. 7 shows the MSE Loss curve of the two networks (i.e. the relationship between Loss and batch number is shown), fig. 7 shows the time spent by the curve from beginning to end, so that the Convergence speed of the two networks is intuitively felt, which does not precisely correspond to the Convergence time, the specific Convergence point (batch number) should be determined according to the trend of the curve decrease, and the "Convergence Loss" corresponds to the average value after the curve Convergence; fig. 7A is a graph of loss of a conventional CNN network, and fig. 7B is a graph of loss of a residual network provided in this embodiment, as can be seen from fig. 7, the CNN network converges around batch 2500, and the residual network provided in this embodiment converges around batch 110; in addition, gradient explosion occurs in the loss curve of the CNN network, and the loss curve of the residual error network remains relatively stable, which indicates that the residual error network provided by the embodiment not only accelerates the prediction and elimination of the EMI, but also enhances the stability of network training.
Meanwhile, referring to fig. 8, fig. 8A shows a denoising effect diagram of a conventional CNN model, and fig. 8B shows an effect diagram of the residual network after denoising provided in the present embodiment (the upper part of the two diagrams are both original images, and the lower part is an image from which EMI signals are removed); as shown on the right side of fig. 8A, there is a hot spot in the center of the fourth image, resulting in the other part of the image being too dark, supposing that some parts of the network are not fitted correctly and are trapped in local optimum, in many training attempts, it is found that the brightness of the hot spot existing on the specific image is random, and the hot spot phenomenon occurs in both networks; however, when the residual network is output, the frequency of occurrence of the hot spot phenomenon is relatively low, so that the problem of local optimum can be effectively reduced, and the denoising effect can be improved.
As shown in fig. 9, a fourth aspect of the present embodiment provides a hardware apparatus for implementing the denoising method based on a medical image for deep learning according to the first aspect of the present embodiment, including:
and an acquisition unit for acquiring the MRI signal and a first EMI signal, wherein the first EMI signal is a noise signal detected by an EMI induction coil located in front of the MRI scanner.
The acquisition unit is further configured to acquire a noise mapping model, where the noise mapping model is a trained GAN model or a trained residual network, the noise mapping model is configured to determine a mapping relationship between a second EMI signal detected by the EMI induction coil and a target noise signal received by the MRI scanner, and the target noise signal is generated by interference of the second EMI signal.
And the denoising unit is used for inputting the first EMI signal into the noise mapping model so as to obtain a target noise signal corresponding to the first EMI signal.
And the denoising unit is also used for obtaining a denoised nuclear magnetic resonance imaging signal according to the target noise signal corresponding to the MRI signal and the first EMI signal.
And the image reconstruction unit is used for carrying out image reconstruction processing based on the denoised nuclear magnetic resonance imaging signals so as to obtain denoised nuclear magnetic resonance images after the image reconstruction processing.
The working process, working details and technical effects of the device provided in this embodiment may refer to the first aspect of the embodiment, and are not described herein again.
As shown in fig. 10, a fifth aspect of the present embodiment provides another denoising apparatus for medical image based on deep learning, taking the apparatus as an electronic device, including: the system comprises a memory, a processor and a transceiver which are connected in sequence in communication, wherein the memory is used for storing a computer program, the transceiver is used for receiving and transmitting messages, and the processor is used for reading the computer program and executing the denoising method of the medical image based on the deep learning according to the first aspect of the embodiment.
By way of specific example, the Memory may include, but is not limited to, random access Memory (random access Memory, RAM), read Only Memory (ROM), flash Memory (Flash Memory), first-in-first-out Memory (First Input First Output, FIFO) and/or first-in-last-out Memory (First In Last Out, FILO), etc.; in particular, the processor may include one or more processing cores, such as a 4-core processor, an 8-core processor, or the like. The processor may be implemented in at least one hardware form of DSP (Digital Signal Processing ), FPGA (Field-Programmable Gate Array, field programmable gate array), PLA (Programmable Logic Array ), and may also include a main processor and a coprocessor, where the main processor is a processor for processing data in an awake state, and is also called CPU (Central Processing Unit ); a coprocessor is a low-power processor for processing data in a standby state.
In some embodiments, the processor may be integrated with a GPU (Graphics Processing Unit, image processor) for taking charge of rendering and rendering of content required to be displayed by the display screen, for example, the processor may not be limited to a microprocessor employing a model number of STM32F105 family, a reduced instruction set computer (reduced instruction set computer, RISC) microprocessor, an X86 or other architecture processor, or a processor integrating an embedded neural network processor (neural-network processing units, NPU); the transceiver may be, but is not limited to, a wireless fidelity (WIFI) wireless transceiver, a bluetooth wireless transceiver, a general packet radio service technology (General Packet Radio Service, GPRS) wireless transceiver, a ZigBee protocol (low power local area network protocol based on the ieee802.15.4 standard), a 3G transceiver, a 4G transceiver, and/or a 5G transceiver, etc. In addition, the device may include, but is not limited to, a power module, a display screen, and other necessary components.
The working process, working details and technical effects of the electronic device provided in this embodiment may refer to the first aspect of the embodiment, and are not described herein again.
A sixth aspect of the present embodiment provides a storage medium storing instructions containing the denoising method for a medical image based on deep learning according to the first aspect of the present embodiment, that is, the storage medium storing instructions thereon, which when executed on a computer, perform the denoising method for a medical image based on deep learning according to the first aspect of the present embodiment.
The storage medium refers to a carrier for storing data, and may include, but is not limited to, a floppy disk, an optical disk, a hard disk, a flash Memory, a flash disk, and/or a Memory Stick (Memory Stick), where the computer may be a general purpose computer, a special purpose computer, a computer network, or other programmable devices.
The working process, working details and technical effects of the storage medium provided in this embodiment may refer to the first aspect of the embodiment, and are not described herein again.
A seventh aspect of the present embodiment provides a computer program product comprising instructions which, when run on a computer, cause the computer to perform the deep learning based medical image denoising method according to the first aspect of the embodiment, wherein the computer may be a general purpose computer, a special purpose computer, a computer network, or other programmable apparatus.
Finally, it should be noted that: the foregoing description is only of the preferred embodiments of the invention and is not intended to limit the scope of the invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (10)
1. A method for denoising medical images based on deep learning, comprising:
acquiring an MRI signal and a first EMI signal, wherein the first EMI signal is a noise signal detected by an EMI induction coil positioned in front of an MRI scanner;
acquiring a noise mapping model, wherein the noise mapping model is a trained GAN model or a trained residual error network, the noise mapping model is used for determining a mapping relation between a second EMI signal detected by the EMI induction coil and a target noise signal received by an MRI scanner, and the target noise signal is generated by interference of the second EMI signal;
inputting the first EMI signal into a noise mapping model to obtain a target noise signal corresponding to the first EMI signal;
obtaining a denoised nuclear magnetic resonance imaging signal according to the target noise signal corresponding to the MRI signal and the first EMI signal;
And carrying out image reconstruction processing based on the denoised nuclear magnetic resonance imaging signals so as to obtain denoised nuclear magnetic resonance images after the image reconstruction processing.
2. The method of claim 1, wherein the noise mapping model is trained by:
acquiring a training data set, wherein the training data set comprises a plurality of second EMI signals and real target noise signals corresponding to the second EMI signals, and the real target noise signals corresponding to any one of the second EMI signals refer to electromagnetic interference signals generated by interference of any one of the second EMI signals received by the MRI scanner;
taking each real target noise signal as tag data;
and training a GAN network or a residual network by taking each second EMI signal and corresponding label data in the training data set as input and the target noise signal corresponding to each second EMI signal as output so as to obtain the noise mapping model after training is completed.
3. The method of claim 2, wherein when the noise mapping model is the trained GAN model, the noise mapping model comprises a generator and a arbiter, wherein the generator is configured to extract characteristic information of the input second EMI signal and generate a target noise signal corresponding to the input second EMI signal based on the characteristic information when training;
The discriminator is used for carrying out two-classification on the output result of the generator during training, and updating the network parameters of the generator according to the two-classification result.
4. A method according to claim 3, wherein the generator comprises four first residual blocks and one first output layer arranged in sequence according to a signal processing direction, wherein each first residual block comprises a first convolution structure layer, a first batch normalization layer and a first correction linearity layer, the first output layer comprises a first convolution layer, and an activation function used by the first output layer is a hyperbolic tangent function;
the discriminator comprises a second convolution layer, a first LeakyReLu layer, a first Dropout layer, a third convolution layer, a second LeakyReLu layer, a second Dropout layer and a second batch normalization layer which are sequentially arranged according to the signal processing direction, and the discriminator is used for carrying out secondary classification on the output result of the generator based on the second convolution layer, the first LeakyReLu layer, the first Dropout layer, the third convolution layer, the second LeakyReLu layer, the second Dropout layer and the second batch normalization layer so as to output a classification result.
5. The method of claim 4, wherein the first convolutional structure layer comprises nine fourth convolutional layers, wherein the nine fourth convolutional layers use convolutional kernels of sizes 11 x 11, 9 x 9, 5 x 5, 1 x 1, and 1 x 1 in order.
6. A method according to claim 3, wherein the generator has a loss function of:
in the above formula (1), L G Represents the loss function of the generator, x represents the second EMI signal of the input, x z The true target noise signal corresponding to the inputted second EMI signal,representing the distribution of the second EMI signal detected by the EMI induction coil, G () representing the generator, D () representing the discriminator, MSE () representing the mean square error;
correspondingly, the loss function of the arbiter is:
in the above formula (2), L D Representing the loss function of the arbiter,representing the distribution of the second EMI signal received by said MRI scanner corresponding to the real target noise signal,/I>The gradient is represented by a gradient, lambda represents the gradient penalty coefficient, I 2 Represents a two-norm and +.>Representing x and x z Is a first linear combination of->Representation->Distribution of (1), and->Expressed as:
in the above formula (3), ε represents x and x z And U represents a uniform distribution.
7. The method of claim 2, wherein when the noise mapping model is the trained residual network, the noise mapping model includes four second residual blocks and one second output layer sequentially arranged according to a signal processing direction, wherein each second residual block includes a second convolution structure layer, a third batch normalization layer, and a second correction linearity layer, the second output layer includes a fifth convolution layer, and an activation function used by the second output layer is a hyperbolic tangent function.
8. The method of claim 7, wherein the second convolutional structure layer comprises nine sixth convolutional layers, and the convolutional kernels used by the nine sixth convolutional layers have sizes of 11 x 11, 9 x 9, 5 x 5, 1 x 1, and 1 x 1 in order;
correspondingly, the loss function of the noise mapping model is as follows:
L=MSE(x,x z ) (4)
in the above formula (4), L represents a loss function of the noise mapping model, x represents the second EMI signal inputted, x z The MSE () represents the mean square error of the true target noise signal corresponding to the second EMI signal input.
9. The method of claim 1, wherein two EMI induction coils are disposed in front of the MRI scanner, wherein the two EMI induction coils are disposed orthogonally in front of the MRI scanner along a slice selection gradient direction of the MRI scanner.
10. A denoising apparatus for medical images based on deep learning, comprising:
an acquisition unit configured to acquire an MRI signal and a first EMI signal, wherein the first EMI signal is a noise signal detected by an EMI induction coil located in front of an MRI scanner;
the acquisition unit is further used for acquiring a noise mapping model, wherein the noise mapping model is a trained GAN model or a trained residual error network, the noise mapping model is used for determining a mapping relation between a second EMI signal detected by the EMI induction coil and a target noise signal received by an MRI scanner, and the target noise signal is generated by interference of the second EMI signal;
The denoising unit is used for inputting the first EMI signal into a noise mapping model so as to obtain a target noise signal corresponding to the first EMI signal;
the denoising unit is also used for obtaining a denoised nuclear magnetic resonance imaging signal according to the target noise signal corresponding to the MRI signal and the first EMI signal;
and the image reconstruction unit is used for carrying out image reconstruction processing based on the denoised nuclear magnetic resonance imaging signals so as to obtain denoised nuclear magnetic resonance images after the image reconstruction processing.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202311071728.9A CN117095073A (en) | 2023-08-23 | 2023-08-23 | Medical image denoising method and device based on deep learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202311071728.9A CN117095073A (en) | 2023-08-23 | 2023-08-23 | Medical image denoising method and device based on deep learning |
Publications (1)
Publication Number | Publication Date |
---|---|
CN117095073A true CN117095073A (en) | 2023-11-21 |
Family
ID=88769426
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202311071728.9A Pending CN117095073A (en) | 2023-08-23 | 2023-08-23 | Medical image denoising method and device based on deep learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN117095073A (en) |
Citations (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109658344A (en) * | 2018-11-12 | 2019-04-19 | 哈尔滨工业大学(深圳) | Image de-noising method, device, equipment and storage medium based on deep learning |
CN110097512A (en) * | 2019-04-16 | 2019-08-06 | 四川大学 | Construction method and the application of the three-dimensional MRI image denoising model of confrontation network are generated based on Wasserstein |
CN110473154A (en) * | 2019-07-31 | 2019-11-19 | 西安理工大学 | A kind of image de-noising method based on generation confrontation network |
CN112509074A (en) * | 2020-11-09 | 2021-03-16 | 成都易检医疗科技有限公司 | Artifact eliminating method, artifact eliminating system, terminal and storage medium |
CN113176528A (en) * | 2021-04-29 | 2021-07-27 | 杭州微影医疗科技有限公司 | Interference cancellation method, medium, and apparatus |
CN113180636A (en) * | 2021-04-29 | 2021-07-30 | 杭州微影医疗科技有限公司 | Interference cancellation method, medium, and apparatus |
CN113203969A (en) * | 2021-04-29 | 2021-08-03 | 杭州微影医疗科技有限公司 | Interference cancellation method, medium, and apparatus |
CN113557526A (en) * | 2018-08-15 | 2021-10-26 | 海珀菲纳股份有限公司 | Deep learning technique for suppressing artifacts in magnetic resonance images |
CN114189695A (en) * | 2020-09-14 | 2022-03-15 | 四川大学 | HEVC compressed video visual perception improving method based on GAN |
WO2022183988A1 (en) * | 2021-03-03 | 2022-09-09 | The University Of Hong Kong | Systems and methods for magnetic resonance image reconstruction with denoising |
CN115598573A (en) * | 2022-10-28 | 2023-01-13 | 安徽华米健康科技有限公司(Cn) | Signal processing method for magnetic resonance system, electronic device, and storage medium |
CN116205807A (en) * | 2023-01-13 | 2023-06-02 | 广东工业大学 | Deep learning-based frozen electron microscope image denoising method and system |
CN116405158A (en) * | 2023-04-04 | 2023-07-07 | 电子科技大学 | End-to-end communication system based on deep learning under non-Gaussian noise |
-
2023
- 2023-08-23 CN CN202311071728.9A patent/CN117095073A/en active Pending
Patent Citations (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113557526A (en) * | 2018-08-15 | 2021-10-26 | 海珀菲纳股份有限公司 | Deep learning technique for suppressing artifacts in magnetic resonance images |
CN109658344A (en) * | 2018-11-12 | 2019-04-19 | 哈尔滨工业大学(深圳) | Image de-noising method, device, equipment and storage medium based on deep learning |
CN110097512A (en) * | 2019-04-16 | 2019-08-06 | 四川大学 | Construction method and the application of the three-dimensional MRI image denoising model of confrontation network are generated based on Wasserstein |
CN110473154A (en) * | 2019-07-31 | 2019-11-19 | 西安理工大学 | A kind of image de-noising method based on generation confrontation network |
CN114189695A (en) * | 2020-09-14 | 2022-03-15 | 四川大学 | HEVC compressed video visual perception improving method based on GAN |
CN112509074A (en) * | 2020-11-09 | 2021-03-16 | 成都易检医疗科技有限公司 | Artifact eliminating method, artifact eliminating system, terminal and storage medium |
WO2022183988A1 (en) * | 2021-03-03 | 2022-09-09 | The University Of Hong Kong | Systems and methods for magnetic resonance image reconstruction with denoising |
CN113203969A (en) * | 2021-04-29 | 2021-08-03 | 杭州微影医疗科技有限公司 | Interference cancellation method, medium, and apparatus |
CN113180636A (en) * | 2021-04-29 | 2021-07-30 | 杭州微影医疗科技有限公司 | Interference cancellation method, medium, and apparatus |
CN113176528A (en) * | 2021-04-29 | 2021-07-27 | 杭州微影医疗科技有限公司 | Interference cancellation method, medium, and apparatus |
CN115598573A (en) * | 2022-10-28 | 2023-01-13 | 安徽华米健康科技有限公司(Cn) | Signal processing method for magnetic resonance system, electronic device, and storage medium |
CN116205807A (en) * | 2023-01-13 | 2023-06-02 | 广东工业大学 | Deep learning-based frozen electron microscope image denoising method and system |
CN116405158A (en) * | 2023-04-04 | 2023-07-07 | 电子科技大学 | End-to-end communication system based on deep learning under non-Gaussian noise |
Non-Patent Citations (1)
Title |
---|
吴茂贵 等: "《智能***与技术丛书 Python深度学习 基于TensorFlow》", vol. 2, 31 October 2022, 北京:机械工业出版社, pages: 266 * |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Chung et al. | Score-based diffusion models for accelerated MRI | |
Tellez et al. | Whole-slide mitosis detection in H&E breast histology using PHH3 as a reference to train distilled stain-invariant convolutional networks | |
CN111784652B (en) | MRI (magnetic resonance imaging) segmentation method based on reinforcement learning multi-scale neural network | |
CN113012246A (en) | Method and processing system for generating synthetic image data and providing trained functions | |
Usman et al. | Gaussian smoothing and modified histogram normalization methods to improve neural-biomarker interpretations for dyslexia classification mechanism | |
CN113592972B (en) | Magnetic resonance image reconstruction method and device based on multi-mode aggregation | |
Wang et al. | A two-step automated quality assessment for liver MR images based on convolutional neural network | |
CN117095073A (en) | Medical image denoising method and device based on deep learning | |
CN112101396A (en) | Classification method, classification device, classification equipment and storage medium | |
CN117197521A (en) | Simulation image preprocessing method, local specific absorption rate estimation method and device | |
CN117095074B (en) | Medical image double denoising method and device based on deep learning | |
CN111383217A (en) | Visualization method, device and medium for evaluation of brain addiction traits | |
CN105654527A (en) | Magnetic resonance imaging reconstruction method and device based on structural dictionary learning | |
CN110289861A (en) | The half precision compressed sensing method of sampling | |
CN115659144A (en) | Sub-health personnel classification method, system and equipment based on fMRI multi-band analysis | |
CN113269711B (en) | Brain image processing method and device, electronic equipment and storage medium | |
CN117094906A (en) | Electromagnetic imaging detection double denoising method and device based on N2V and EMI removal algorithm | |
Zheng et al. | A novel approach to activation detection in fMRI based on empirical mode decomposition | |
CN110889836A (en) | Image data analysis method and device, terminal equipment and storage medium | |
CN111709946A (en) | Image data analysis method and device, terminal equipment and storage medium | |
CN110517239A (en) | A kind of medical image detection method and device | |
Lou et al. | Predicting radiologist attention during mammogram reading with deep and shallow high-resolution encoding | |
CN110288565A (en) | Defect identification method, device and the storage medium of carbon fiber composite core wire | |
CN116934597B (en) | FDNN model-based magnetic particle imaging spatial resolution improvement method | |
CN116883396B (en) | Resting state magnetic resonance image analysis method and system based on artificial intelligence |
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 |