CN113240078B - Magnetic resonance R2 based on deep learning network * Parameter quantization method, medium and device - Google Patents

Magnetic resonance R2 based on deep learning network * Parameter quantization method, medium and device Download PDF

Info

Publication number
CN113240078B
CN113240078B CN202110469551.2A CN202110469551A CN113240078B CN 113240078 B CN113240078 B CN 113240078B CN 202110469551 A CN202110469551 A CN 202110469551A CN 113240078 B CN113240078 B CN 113240078B
Authority
CN
China
Prior art keywords
deep learning
magnetic resonance
learning network
noise
weighted image
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202110469551.2A
Other languages
Chinese (zh)
Other versions
CN113240078A (en
Inventor
冯衍秋
陆琪琪
张鑫媛
连梓锋
王华峰
陈凌剑
龚剑
陈武凡
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Southern Medical University
Original Assignee
Southern Medical University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Southern Medical University filed Critical Southern Medical University
Priority to CN202110469551.2A priority Critical patent/CN113240078B/en
Publication of CN113240078A publication Critical patent/CN113240078A/en
Application granted granted Critical
Publication of CN113240078B publication Critical patent/CN113240078B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G06T7/0012—Biomedical image inspection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/70—Denoising; Smoothing
    • 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
    • G06T2207/10088—Magnetic resonance imaging [MRI]
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
    • Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
    • Y02A90/30—Assessment of water resources

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Biophysics (AREA)
  • Molecular Biology (AREA)
  • Biomedical Technology (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Computational Linguistics (AREA)
  • Software Systems (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Medical Informatics (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Radiology & Medical Imaging (AREA)
  • Quality & Reliability (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Magnetic Resonance Imaging Apparatus (AREA)

Abstract

The invention discloses a magnetic resonance R2 parameter quantization method based on a deep learning network, which comprises the following steps: denoising the input magnetic resonance T2 weighted image by using a first deep learning network; simultaneously, performing magnetic resonance R2 quantification prediction on the magnetic resonance T2 weighted image subjected to noise treatment by using a second deep learning network; the first deep learning network and the second deep learning network constitute a cascading network. And the R2 parameter is quantized by using a deep learning algorithm, so that the complexity of the traditional algorithm is avoided, the quantization error is reduced, the calculation speed is improved, and the time required by calculation is shortened.

Description

Magnetic resonance R2 parameter quantization method, medium and equipment based on deep learning network
Technical Field
The invention belongs to the technical field of medical image processing, and particularly relates to a magnetic resonance R2 parameter quantization method, medium and equipment based on a deep learning network.
Background
Magnetic resonance imaging (Magnetic resonance imaging, MRI) is one of the important examination means of clinical medical imaging. However, due to limitations of the imaging mechanism, noise is inevitably introduced during imaging, mainly including random background noise and structural noise. Background noise is mainly composed of the resistance of a radio frequency coil, the electronic noise of a preamplifier, dielectric loss and induction loss (molecular motion, electromagnetic noise generated by charged particles in a human body) in an imaging object, and the like. Noise in the image can reduce the quality of the image, so that the displayed signal values of the image deviate, thereby affecting quantization of other parameter values and further affecting clinical diagnosis. Therefore, it is important for clinical diagnosis and image analysis that the parameter quantization method is insensitive to noise or can eliminate the influence of noise. Other researchers have proposed various methods for solving the noise impact quantization results in the R2 parameter quantization process, including denoising before voxel-by-voxel parameter fitting quantization, pixel-by-pixel quantization in combination with noise statistical models (M 1 NCM), voxel-by-voxel quantization (PCANR) with adaptive regularization using information of neighboring voxels, and so forth. However, most methods have problems of high computational complexity, long computation time, and reduced quantization accuracy when noise is severe.
Disclosure of Invention
To overcome the above technical drawbacks, a first aspect of the present invention provides a method for quantifying a magnetic resonance R2 parameter based on a deep learning network, including the steps of:
denoising the input magnetic resonance T2 weighted image by using a first deep learning network;
simultaneously, a second deep learning network is used for predicting the magnetic resonance R2 quantization parameter of the denoised magnetic resonance T2 weighted image;
the first deep learning network and the second deep learning network constitute a cascading network.
As a further improvement of the present invention, before the first deep learning network and the second deep learning network are used, the method further comprises the steps of:
and training the first deep learning network and the second deep learning network until the network converges and the verification set loss is stabilized.
As a further improvement of the present invention, the step of training the first deep learning network and the second deep learning network simultaneously specifically includes the steps of:
adopting a loss function, and simultaneously carrying out joint training on the first deep learning network and the second deep learning network, wherein the loss function is as follows:
wherein Φ represents a mapping relationship of the first deep learning network; u represents the mapping relation of the second deep learning network; y represents the input magnetic resonance T2 weighted noise images at different TE times; x represents an image without noise or an image with good quality; lambda (lambda) 1 And lambda (lambda) 2 And respectively corresponding constraint coefficients, wherein M is an analytical model of T2 attenuation.
As a further improvement of the present invention, before the step of simultaneously training the first deep learning network and the second deep learning network, the method further comprises the steps of:
according to the acquired multi-echo magnetic resonance T2 weighted image, a simulation method is used for calculating to obtain a noise-free multi-echo magnetic resonance T2 weighted image;
adding noise to the denoised magnetic resonance T2 weighted image to obtain a magnetic resonance T2 weighted image polluted by the noise;
the noise-free magnetic resonance T2 weighted image and the noise-polluted magnetic resonance T2 weighted image jointly construct a training set which is used for training a cascade network formed by the first deep learning network and the second deep learning network;
training parameters of the first deep learning network and the second deep learning network are respectively set.
As a further improvement of the present invention, in the step of adding noise to the denoised image to obtain a noise-contaminated magnetic resonance T2 x weighted image, at least one of white noise, gaussian noise, rice noise, and non-center chi-square distributed noise is added to the denoised magnetic resonance T2 x weighted image.
As a further improvement of the present invention, the step of calculating a noise-free magnetic resonance T2 weighted image according to the acquired multi-echo magnetic resonance T2 weighted image by using the simulation method specifically includes the steps of:
denoising the acquired magnetic resonance T2 weighted image by using a traditional denoising method;
calculating parameters by using a traditional least square fitting parameter quantization method to obtain a parameter map;
and obtaining a noise-free magnetic resonance T2 weighted image according to a specific model corresponding to the image acquisition protocol.
As a further improvement of the invention, the first deep learning network and the second deep learning network are jointly trained by adopting a Patch method.
As a further development of the invention, the magnetic resonance T2-weighted images are magnetic resonance T2-weighted images acquired at different echo times.
Compared with the prior art, the invention has the following beneficial effects:
and the R2 parameter is quantized by using a deep learning algorithm, so that the complexity of the traditional algorithm is avoided, the quantization error is reduced, the calculation speed is improved, and the time required by calculation is shortened.
In a second aspect of the present invention, a computer readable storage medium is provided, where a computer program is stored, where the computer program when executed by a processor causes the processor to implement the method for quantifying a magnetic resonance R2 parameter as described above.
In a third aspect of the invention, a computer device is provided, comprising a processor and a memory; the memory is used for storing a computer program; the processor is configured to execute the computer program and implement the magnetic resonance R2 parameter quantization method when executing the computer program.
Drawings
The invention is described in further detail below with reference to the attached drawing figures, wherein:
fig. 1 is an overall frame diagram of a method for quantifying the R2 parameter of the magnetic resonance in embodiment 1;
fig. 2 is a schematic diagram of the layout of a denoise deep learning network (first deep learning network for Denoising) and a Mapping deep learning network (second deep learning network for parameter quantization) in embodiment 1;
fig. 3 is a technical effect diagram obtained in embodiment 1.
Detailed Description
The preferred embodiments of the present invention will be described below with reference to the accompanying drawings, it being understood that the preferred embodiments described herein are for illustration and explanation of the present invention only, and are not intended to limit the present invention.
Example 1
The embodiment discloses a magnetic resonance R2 parameter quantization method based on a deep learning network, as shown in fig. 1 and fig. 2, comprising the steps of:
s1, denoising an input magnetic resonance T2-weighted image by using a first deep learning network, wherein T2 is an effective transverse relaxation time.
Specifically, the input magnetic resonance T2 x weighted image is acquired by magnetic resonance acquisition. The input magnetic resonance T2-weighted image may be a magnitude image, a "real-valued" image or a "complex-valued" image, each pixel of which reflects the magnitude, real-valued or complex-valued of the MR signal at the corresponding voxel location.
In some embodiments, the input magnetic resonance T2-weighted image is obtained by fourier transforming the acquired k-space data.
In some embodiments, the acquired k-space data is preprocessed prior to fourier transformation.
In some embodiments, the fourier transformed data is processed to obtain an input magnetic resonance T2-x weighted image, which may include any suitable processing during reconstruction, or a combination thereof. The input magnetic resonance T2 weighted image may have any size, for example, 64×128 pixels, 128×128 pixels, 256×256 pixels, 512×512 pixels, etc.
Different levels, sources, and distributions of noise may be generated during the acquisition of an input image due to a variety of factors. Noise is mainly derived from random background noise (e.g., white noise, gaussian noise, rice noise, or non-center chi-square distributed noise, etc.), and may be due to the resistance of the radio frequency coil, the electronic noise of the preamplifier, or the dielectric effects of the imaged object itself, so that denoising is required.
The first deep learning network may employ a denoise deep learning network comprising a plurality of convolutional layers, each layer having one or more input channels and one or more output channels. The parameters of the convolution kernel of each convolution layer are learnable, and the values obtained after convolution are transferred to the activation function in response to the output of the activation function.
In some embodiments, the activation function may be a ReLU function, comprising: common ReLU functions, leakage ReLU functions, parameter ReLU functions, etc. The input image is accepted by the first convolution layer.
In some embodiments, there are one or more convolution layers between the first convolution layer and the last convolution layer. Note that in the example embodiment, only 8 convolutional layers are shown in fig. 2 for illustration. In an example embodiment, the denoise deep learning network is a residual network. It should be noted that the layout of the deep learning network shown in fig. 2 is for illustration and not limitation, and any other suitable configuration may be used herein.
S2, simultaneously using a second deep learning network to conduct magnetic resonance R2 quantization prediction on the magnetic resonance T2 weighted image subjected to denoising, wherein R2 is the inverse number of T2, in other words, parameter quantization is achieved through the fact that the first deep learning network and the second deep learning network are connected in series, and the first deep learning network and the second deep learning network are used cooperatively.
The second deep learning network may employ a Mapping deep learning network that includes a plurality of convolutional layers, each layer having one or more input channels and one or more output channels. The parameters of the convolution kernel of each convolution layer are learnable, and the values obtained after convolution are transferred to the activation function to output the activation function as corresponding.
In some embodiments, the activation function may be a ReLU function, comprising: common ReLU functions, leakage ReLU functions, parameter ReLU functions, etc. The input image is accepted by the first convolution layer. In an example embodiment, the Mapping deep learning network is a 6-layer UNet convolutional neural network, including an encoder portion and a decoder portion, each having three convolutional layers. The decoder section converts the image features extracted by the encoder into a parametric image of pixels.
In some embodiments, the number of layers of the convolutional layer is not limited to the number of layers shown in the example embodiment, and it should be noted that the deep-learning network layout shown in FIG. 2 is for illustration and not limitation, and any other suitable configuration may be used herein.
S3, the first deep learning network and the second deep learning network form a cascade network.
In the above embodiment, before the first deep learning network and the second deep learning network are used, the method further includes the steps of:
the simulation data is used to perform iterative training on the first deep network learning network and the second deep network learning network at the same time, for example, different distribution types and different levels of noise can be added to a noise-free simulation image or a magnetic resonance T2-x weighted image with a plurality of TE times of good quality, so as to obtain a synthesized noise-contaminated image. The type of noise distribution depends on the hardware and software parameters of the imaged magnetic resonance. The combined noisy image is taken as an input image and a magnetic resonance image without noise or with good quality is taken as an output to jointly train the first and second deep learning networks. Training the first deep learning network and the second deep learning network until the network converges, verifying that the set loss reaches stability, and stopping iteration.
In the above embodiment, the step of training the first deep learning network and the second deep learning network simultaneously specifically includes the steps of:
adopting a loss function, and simultaneously carrying out joint training on a first deep learning network and a second deep learning network, wherein the loss function is as follows:
wherein Φ represents the mapping relation of the Denoising deep learning network (namely the first deep learning network); u represents the Mapping relation of the Mapping deep learning network (namely a second deep learning network); y represents the input magnetic resonance T2 weighted noise images at different TE times; x represents an image without noise or an image with good quality; lambda (lambda) 1 And lambda (lambda) 2 And respectively corresponding constraint coefficients, wherein M is an analytical model of T2 attenuation.
In the above embodiment, before the step of training the first deep learning network and the second deep learning network simultaneously, the method further includes the steps of:
according to the acquired multi-echo magnetic resonance T2 weighted image, a simulation method is used for calculating to obtain a noise-free multi-echo magnetic resonance T2 weighted image;
adding noise to the denoised magnetic resonance T2 weighted image to obtain a noise-contaminated magnetic resonance T2 weighted image, wherein the method specifically comprises the following steps of: adding at least one of white noise, gaussian noise, rice noise and non-center chi-square distributed noise into the denoised magnetic resonance T2 weighted image;
the noise-free magnetic resonance T2 weighted image and the noise-polluted magnetic resonance T2 weighted image jointly construct a training set which is used for training a cascade network formed by a first deep learning network and a second deep learning network;
training parameters of the first deep learning network and the second deep learning network are respectively set.
In the above embodiment, the step of calculating the noise-free magnetic resonance T2 x-weighted image according to the acquired multi-echo magnetic resonance T2 x-weighted image by using the simulation method specifically includes the steps of:
denoising the acquired magnetic resonance T2 weighted image by using a traditional denoising method;
calculating parameters by using a traditional least square fitting parameter quantization method to obtain a parameter map;
and obtaining a noise-free magnetic resonance T2 weighted image according to a specific model corresponding to the image acquisition protocol.
In the above embodiment, the Patch method is used to train the first deep learning network and the second deep learning network.
In the above embodiment, the magnetic resonance T2-weighted images are magnetic resonance T2-weighted images acquired at different echo times.
The effect obtained by the magnetic resonance R2 parameter quantification method of this example is shown in fig. 3, wherein the 1 st column image is a clinical magnetic resonance T2 weighted image at different liver iron deposition levels, including normal (normal), mild (mid), moderate (moderate) and severe (grade) iron deposition levels. And 2-4 columns are R2 parameter images calculated by different methods respectively. M is M 1 NCM is a traditional least square fitting algorithm based on a noise model; PCANR is a traditional self-adaptive neighborhood regularized voxel-by-voxel quantization algorithm; cadamNet (Cascade of denoising and mapping networks) is a method for quantizing the R2 parameter of the magnetic resonance method according to this embodiment, so that it is seen that the result is less affected by noise, and the accuracy of the quantized result is improved. From the figure3, the R2 image obtained by the method is similar to the traditional method on the whole, but the framework is smooth and uniform, and the error is smaller; and the R2 parameter is quantized by using a deep learning algorithm, so that the complexity of the traditional algorithm is avoided, the calculation speed is improved, and the time required by calculation is shortened.
Example 2
The embodiment discloses a computer device, which comprises a processor and a memory; the memory is used for storing a computer program; the processor is configured to execute the computer program and implement the method for quantifying the magnetic resonance R2 parameter in embodiment 1 when executing the computer program.
Those of ordinary skill in the art will appreciate that all or some of the steps, systems, functional modules/units in the apparatus, and methods disclosed above may be implemented as software, firmware, hardware, and suitable combinations thereof. In a hardware implementation, the division between the functional modules/units mentioned in the above description does not necessarily correspond to the division of physical components; for example, one physical component may have multiple functions, or one function or step may be performed cooperatively by several physical components. Some or all of the physical components may be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application specific integrated circuit. Such software may be distributed on computer-readable storage media, which may include computer-readable storage media (or non-transitory media) and communication media (or transitory media).
The term computer-readable storage medium includes volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storage of information, such as computer-readable instructions, data structures, program modules or other data, as known to those skilled in the art. Computer-readable storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital Versatile Disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other medium which can be used to store the desired information and which can be accessed by a computer. Furthermore, as is well known to those of ordinary skill in the art, communication media typically embodies computer readable instructions, data structures, program modules or other data in a modulated data signal such as a carrier wave or other transport mechanism and includes any information delivery media.
The computer readable storage medium may be an internal storage unit of the network management device according to the foregoing embodiment, for example, a hard disk or a memory of the network management device. The computer readable storage medium may also be an external storage device of the network management device, for example, a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) Card, a Flash memory Card (Flash Card) or the like, which are provided on the network management device.
Example 3
The embodiment discloses a computer device, which comprises a processor and a memory; the memory is used for storing a computer program; the processor is configured to execute the computer program and implement the method for quantifying the magnetic resonance R2 parameter in embodiment 1 when executing the computer program.
It should be appreciated that the processor may be a central processing unit (Central Processing Unit, CPU), but may also be other general purpose processors, digital signal processors (Digital Signal Processor, DSP), application specific integrated circuits (Application Specific Integrated Circuit, ASIC), field-programmable gate arrays (Field-Programmable Gate Array, FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or the like. Wherein the general purpose processor may be a microprocessor or the processor may be any conventional processor or the like.
The present invention is not limited to the preferred embodiments, and any modifications, equivalent variations and modifications made to the above embodiments according to the technical principles of the present invention are within the scope of the technical proposal of the present invention.

Claims (8)

1. The magnetic resonance R2 parameter quantization method based on the deep learning network is characterized by comprising the following steps of:
denoising the input magnetic resonance T2 weighted image by using a first deep learning network;
simultaneously, a second deep learning network is used for predicting the magnetic resonance R2 quantization parameter of the denoised magnetic resonance T2 weighted image;
the first deep learning network and the second deep learning network form a cascade network;
before using the first deep learning network and the second deep learning network, the method further comprises the steps of:
training the first deep learning network and the second deep learning network until the network converges and the verification set loss is stable;
the step of training the first deep learning network and the second deep learning network simultaneously specifically includes the steps of:
adopting a loss function, and simultaneously carrying out joint training on the first deep learning network and the second deep learning network, wherein the loss function is as follows:
L＝λ 1 ||Φ(y)-x|| 2 2 +λ 2 ||M(U(Φ(y)))-x|| 2 2
wherein Φ represents a mapping relationship of the first deep learning network; u represents the mapping relation of the second deep learning network; y represents the input magnetic resonance T2 weighted noise images at different TE times; x represents an image without noise or an image with good quality; lambda (lambda) 1 And lambda (lambda) 2 And respectively corresponding constraint coefficients, wherein M is an analytical model of T2 attenuation.
2. The method of claim 1, further comprising, prior to the step of simultaneously training the first deep learning network and the second deep learning network, the step of:
according to the acquired multi-echo magnetic resonance T2 weighted image, a simulation method is used for calculating to obtain a noise-free multi-echo magnetic resonance T2 weighted image;
adding noise to the denoised magnetic resonance T2 weighted image to obtain a magnetic resonance T2 weighted image polluted by the noise;
the noise-free magnetic resonance T2 weighted image and the noise-polluted magnetic resonance T2 weighted image jointly construct a training set which is used for training a cascade network formed by the first deep learning network and the second deep learning network;
training parameters of the first deep learning network and the second deep learning network are respectively set.
3. The method for quantizing R2 parameters of claim 2, wherein in the step of adding noise to the denoised image to obtain a noise-contaminated magnetic resonance T2-weighted image, at least one of white noise, gaussian noise, rice noise, and non-center chi-square distributed noise is added to the denoised magnetic resonance T2-weighted image.
4. The method for quantifying the R2 parameter of the magnetic resonance according to claim 2, wherein the step of calculating a noise-free magnetic resonance T2 weighted image according to the acquired multi-echo magnetic resonance T2 weighted image by using the simulation method specifically comprises the steps of:
denoising the acquired magnetic resonance T2 weighted image by using a traditional denoising method;
calculating parameters by using a traditional least square fitting parameter quantization method to obtain a parameter map;
and obtaining a noise-free magnetic resonance T2 weighted image according to a specific model corresponding to the image acquisition protocol.
5. The method of claim 1, wherein the first deep learning network and the second deep learning network are jointly trained using a Patch method.
6. The method of claim 1, wherein the magnetic resonance T2 x-weighted images are magnetic resonance T2 x-weighted images acquired at different echo times.
7. A computer readable storage medium, characterized in that the computer readable storage medium stores a computer program, which when executed by a processor causes the processor to implement the magnetic resonance R2 parameter quantification method according to any of the claims 1 to 6.
8. A computer device comprising a processor and a memory; the memory is used for storing a computer program; the processor for executing the computer program and for implementing the magnetic resonance R2 parameter quantification method according to any of the claims 1 to 6 when the computer program is executed.
CN202110469551.2A 2021-04-26 2021-04-26 Magnetic resonance R2 based on deep learning network * Parameter quantization method, medium and device Active CN113240078B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110469551.2A CN113240078B (en) 2021-04-26 2021-04-26 Magnetic resonance R2 based on deep learning network * Parameter quantization method, medium and device

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110469551.2A CN113240078B (en) 2021-04-26 2021-04-26 Magnetic resonance R2 based on deep learning network * Parameter quantization method, medium and device

Publications (2)

Publication Number Publication Date
CN113240078A CN113240078A (en) 2021-08-10
CN113240078B true CN113240078B (en) 2024-03-19

Family

ID=77131286

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110469551.2A Active CN113240078B (en) 2021-04-26 2021-04-26 Magnetic resonance R2 based on deep learning network * Parameter quantization method, medium and device

Country Status (1)

Country Link
CN (1) CN113240078B (en)

Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6605943B1 (en) * 1999-08-24 2003-08-12 Inner Vision Biometrics Pty Ltd Method of mapping proton transverse relaxation time constants or functions thereof in a target subject to localised movement using nuclear magnetic resonance imaging
CN103218788A (en) * 2013-04-24 2013-07-24 南方医科大学 Method for measuring liver magnetic resonance crosswise relaxation rate R2* parameter
CN103714521A (en) * 2013-12-30 2014-04-09 南方医科大学 Liver R2* graph measuring method based on inquiry table
CN105608681A (en) * 2016-02-14 2016-05-25 电子科技大学 Liver magnetic resonance R2* parameter graph drafting method
CN108663644A (en) * 2018-03-14 2018-10-16 厦门大学 Single sweep Quantitative MRI Measurement T based on residual error network reconnection2* imaging method
CN110163812A (en) * 2019-04-10 2019-08-23 深圳大学 Muscle ultrasonic image denoising method, apparatus, storage medium and computer product
CN110782452A (en) * 2019-11-05 2020-02-11 厦门大学 T2 quantitative image imaging method and system
CN111915007A (en) * 2020-07-29 2020-11-10 厦门大学 Magnetic resonance spectrum noise reduction method based on neural network

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8854038B2 (en) * 2011-04-19 2014-10-07 Wisconsin Alumni Research Foundation Method for R2* quantification with magnetic resonance imaging
KR20200126825A (en) * 2019-04-30 2020-11-09 삼성전자주식회사 Magnetic resonance imaging apparatus for generating a parameter map from a magnetic resonance signal data and method for operating the same

Patent Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6605943B1 (en) * 1999-08-24 2003-08-12 Inner Vision Biometrics Pty Ltd Method of mapping proton transverse relaxation time constants or functions thereof in a target subject to localised movement using nuclear magnetic resonance imaging
CN103218788A (en) * 2013-04-24 2013-07-24 南方医科大学 Method for measuring liver magnetic resonance crosswise relaxation rate R2* parameter
CN103714521A (en) * 2013-12-30 2014-04-09 南方医科大学 Liver R2* graph measuring method based on inquiry table
CN105608681A (en) * 2016-02-14 2016-05-25 电子科技大学 Liver magnetic resonance R2* parameter graph drafting method
CN108663644A (en) * 2018-03-14 2018-10-16 厦门大学 Single sweep Quantitative MRI Measurement T based on residual error network reconnection2* imaging method
CN110163812A (en) * 2019-04-10 2019-08-23 深圳大学 Muscle ultrasonic image denoising method, apparatus, storage medium and computer product
CN110782452A (en) * 2019-11-05 2020-02-11 厦门大学 T2 quantitative image imaging method and system
CN111915007A (en) * 2020-07-29 2020-11-10 厦门大学 Magnetic resonance spectrum noise reduction method based on neural network

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
Cascade of Denoising and Mapping Neural Networks for MRI R2* Relaxometry of Iron-Loaded Liver;Qiqi Lu et al;Bioengineering;第10卷(第2期);DOI10.3390/bioengineering10020209 *
Improved Liver R*2 Mapping by Pixel-Wise Curve Fitting With Adaptive Neighborhood Regularization;Changqing Wang et al;Magnetic Resonance in Medicine;第80卷;792-801 *
Optimal region-of-interest MRI R2* measurements for the assessment of hepatic iron content in thalassaemia major;Meiyan Feng et al;Magnetic Resonance Imaging;647-653 *
基于级联卷积神经网络的前列腺磁共振图像分类;刘可文 等;波谱学杂志;第37卷(第2期);152-161 *

Also Published As

Publication number Publication date
CN113240078A (en) 2021-08-10

Similar Documents

Publication Publication Date Title
Jiang et al. Denoising of 3D magnetic resonance images with multi-channel residual learning of convolutional neural network
Pierre et al. Multiscale reconstruction for MR fingerprinting
US11080897B2 (en) Systems and methods for a PET image reconstruction device
Coupé et al. 3D wavelet subbands mixing for image denoising
Eun et al. Deep-learning-based image quality enhancement of compressed sensing magnetic resonance imaging of vessel wall: comparison of self-supervised and unsupervised approaches
You et al. Denoising of MR images with Rician noise using a wider neural network and noise range division
US11035919B2 (en) Image reconstruction using a colored noise model with magnetic resonance compressed sensing
Pal et al. Rician noise removal in magnitude MRI images using efficient anisotropic diffusion filtering
Baselice et al. A 3D MRI denoising algorithm based on Bayesian theory
Wang et al. High‐field mr diffusion‐weighted image denoising using a joint denoising convolutional neural network
Pawar et al. A deep learning framework for transforming image reconstruction into pixel classification
Khaleel et al. Denoising of Rician corrupted 3D magnetic resonance images using tensor-SVD
Manjón et al. MRI denoising using Deep Learning and Non-local averaging
US11222406B2 (en) Method and system for training a machine learning-based image denoising system
Tan et al. Pixelwise Estimation of Signal‐Dependent Image Noise Using Deep Residual Learning
KR20220082302A (en) MAGNETIC RESONANCE IMAGE PROCESSING APPARATUS AND METHOD USING ARTIFICIAL NEURAL NETWORK AND RESCAlING
Yeganli et al. Noise removal from MR images via iterative regularization based on higher-order singular value decomposition
CN113240078B (en) Magnetic resonance R2 based on deep learning network * Parameter quantization method, medium and device
Islam et al. Compressed sensing regularized calibrationless parallel magnetic resonance imaging via deep learning
US11941732B2 (en) Multi-slice MRI data processing using deep learning techniques
Aetesam et al. Perceptually motivated generative model for magnetic resonance image denoising
Ferrari Off-line determination of the optimal number of iterations of the robust anisotropic diffusion filter applied to denoising of brain MR images
US20220180574A1 (en) Reconstruction with magnetic resonance compressed sensing
Korti Regularization in parallel magnetic resonance imaging
Dhengre et al. An edge guided cascaded U‐net approach for accelerated magnetic resonance imaging reconstruction

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