CN109272486A - Training method, device, equipment and the storage medium of MR image prediction model - Google Patents
Training method, device, equipment and the storage medium of MR image prediction model Download PDFInfo
- Publication number
- CN109272486A CN109272486A CN201810924191.9A CN201810924191A CN109272486A CN 109272486 A CN109272486 A CN 109272486A CN 201810924191 A CN201810924191 A CN 201810924191A CN 109272486 A CN109272486 A CN 109272486A
- Authority
- CN
- China
- Prior art keywords
- image
- prediction model
- image prediction
- training
- training sample
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
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
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10072—Tomographic images
- G06T2207/10081—Computed x-ray tomography [CT]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Radiology & Medical Imaging (AREA)
- Quality & Reliability (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Apparatus For Radiation Diagnosis (AREA)
Abstract
The present invention is applicable in technical field of medical image processing, provide a kind of training method of MR image prediction model, device, equipment and storage medium, this method comprises: obtaining CT image and the corresponding MR image of CT image from sample data gathered in advance, to generate training sample set, building MR image prediction model simultaneously initializes the model, the MR image prediction model after initialization is trained according to training sample set, obtain trained MR image prediction model, MR image prediction model is the convolutional neural networks of U-shaped network structure, to obtain corresponding MR image by CT image prediction by trained MR image prediction model realization, forecasting efficiency is higher and predicts that obtained MR picture quality is preferable, and then the MR image by combining CT image and prediction to obtain, it reduces and suffers from Person's sees a doctor cost, improves the therapeutic effect of doctor.
Description
Technical field
The invention belongs to technical field of medical image processing more particularly to a kind of training method of MR image prediction model,
Device, equipment and storage medium.
Background technique
CT image can in the radiation scope of safety to human body high density tissue have preferable imaging effect, and its at
As fast speed, it is commonly used for clinical diagnosis, but ineffective for the soft-tissue imaging of low-density.MR image is with higher
Soft tissue contrast, but its imaging time is long, expensive, and for claustrophobia or with bodies such as pacemakers
The patient of interior ferromagnetic foreign matter cannot acquire MR image.In order to diagnose the state of an illness in clinic, doctor generally requires to understand surroundings thereof
The case where soft tissue, therefore the additional MR image for acquiring patient is needed when acquiring CT image, but this also increases patient simultaneously
The economic cost and time cost seen a doctor.
It needs to carry out pendulum position to patient before the treatment in radiotherapy, correctly putting position can ensure that ray more accurately irradiates
Target area avoids or reduces the irradiation to normal surrounding tissue.It is often done by CT image image-guided in clinic, patient is put
Position, but CT image is not high for the resolution ratio of soft tissue, is difficult with CT image for the soft tissue neoplasms patient such as head and defines
The exact position of tumour.If not can be carried out accurate pendulum position, ray may jeopardize surrounding tissue in Patients During Radiotherapy, after causing seriously
Fruit.MR image soft tissue contrast with higher, can specify brain tumor position, more accurately carry out pendulum position to patient,
Since the high-intensity magnetic field and radiotherapy apparatus of MR equipment have complicated compatibility issue, the radiotherapy apparatus of mainstream and it is not introduced into MR at present
Image is as image-guided means.
Summary of the invention
The purpose of the present invention is to provide training method, device, equipment and the storage medium of a kind of MR image prediction model,
It aims to solve the problem that since the imaging of MR image is at high cost in the prior art and can not be compatible with radiotherapy apparatus, and CT image is to soft tissue
The problem of imaging effect is bad, increases the therapeutic effect of patient seen a doctor cost and influence doctor.
On the one hand, the present invention provides a kind of training method of MR image prediction model, the method includes the following steps:
From sample data gathered in advance, CT image and the corresponding MR image of the CT image are obtained, according to the CT
Image and the corresponding MR image of the CT image generate training sample set；
MR image prediction model is constructed, and the MR image prediction model is initialized, the MR image prediction mould
Type is the convolutional neural networks of U-shaped network structure；
According to the training sample set, Training is carried out to the MR image prediction model after initialization, is obtained
Trained MR image prediction model.
On the other hand, the present invention provides a kind of training device of MR image prediction model, described device includes:
Training sample generation unit, for obtaining CT image and the CT image pair from sample data gathered in advance
The MR image answered generates training sample set according to the CT image and the corresponding MR image of the CT image；
Prediction model construction unit carries out just for constructing MR image prediction model, and to the MR image prediction model
Beginningization, the MR image prediction model are the convolutional neural networks of U-shaped network structure；And
Prediction model training unit is used for according to the training sample set, to the MR image prediction mould after initialization
Type carries out Training, obtains trained MR image prediction model.
On the other hand, the present invention also provides a kind of Medical Image Processing equipment, including memory, processor and storage
In the memory and the computer program that can run on the processor, the processor execute the computer program
Step of the Shi Shixian as described in the training method of above-mentioned MR image prediction model.
On the other hand, the present invention also provides a kind of computer readable storage medium, the computer readable storage mediums
It is stored with computer program, the training side such as above-mentioned MR image prediction model is realized when the computer program is executed by processor
Step described in method.
The present invention provides a kind of method of the corresponding MR image of the training using known MR image prediction model.The present invention
Using the method for deep learning, a convolutional neural networks have been built, a large amount of CT image and MR image have been trained, finally
Realize the corresponding MR image of training by MR image prediction model.This method can be used for auxiliary diagnosis based on ct images in clinic
With the accurate pendulum position in radiotherapy.
The present invention is constructed by CT image and the corresponding MR image construction training sample set of CT image and to initialize MR image pre-
Model is surveyed, is trained according to MR image prediction model of the training sample set to initialization, obtains trained MR image prediction
Model, MR image prediction model is the convolutional neural networks of U-shaped network structure, to pass through trained MR image prediction mould
Type predicts corresponding MR image in the case where only scanning computed tomography image, and forecasting efficiency is higher and predicts obtained MR picture quality
Preferably, and then doctor is helped to understand the soft tissue situation of region of interest, assists diagnosis, it can also be in Patients During Radiotherapy using pre-
The MR visual aids of survey put position, while also reduce patient sees a doctor cost.
Detailed description of the invention
Fig. 1 is the implementation flow chart of the training method for the MR image prediction model that the embodiment of the present invention one provides；
Fig. 2 is the knot of MR image prediction model in the training method for the MR image prediction model that the embodiment of the present invention one provides
Structure exemplary diagram；
Fig. 3 is experimental result example in the training method for the MR image prediction model that the embodiment of the present invention one provides；
Fig. 4 is the structural schematic diagram of the training device of MR image prediction model provided by Embodiment 2 of the present invention；
Fig. 5 is the preferred structure schematic diagram of the training device of MR image prediction model provided by Embodiment 2 of the present invention；With
And
Fig. 6 is the structural schematic diagram for the image processing equipment that the embodiment of the present invention three provides.
Specific embodiment
In order to make the objectives, technical solutions, and advantages of the present invention clearer, with reference to the accompanying drawings and embodiments, right
The present invention is further elaborated.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, and
It is not used in the restriction present invention.
Specific implementation of the invention is described in detail below in conjunction with specific embodiment:
Embodiment one:
Fig. 1 shows the implementation process of the training method of the MR image prediction model of the offer of the embodiment of the present invention one, in order to
Convenient for explanation, only parts related to embodiments of the present invention are shown, and details are as follows:
In step s101, from sample data gathered in advance, CT image and the corresponding MR image of CT image, root are obtained
According to CT image and the corresponding MR image of CT image, training sample set is generated.
The present invention is suitable for Medical Image Processing platform, system or equipment.It can be adopted from existing medical image library in advance
Collect the preferable sample data of image quality, includes CT in sample data to improve the training effect of subsequent MR image prediction model
Image and the corresponding MR image of CT image, by these CT images MR image construction training sample set corresponding with these CT images.
Preferably, when by CT image and CT image corresponding MR image construction training sample set, CT image and CT are schemed
As the progress image registration of corresponding MR image, the CT image after registration is set to the training sample of training sample concentration, by CT
Image is corresponding, the MR image after registration is set as the corresponding label of training sample, thus for subsequent MR image prediction model
Training is prepared, and the training effect of subsequent MR image prediction model is improved.
In step s 102, MR image prediction model is constructed, and MR image prediction model is initialized, MR image is pre-
Survey the convolutional neural networks that model is U-shaped network structure.
In embodiments of the present invention, MR image prediction model is constructed, wherein MR image prediction model is U-shaped network structure
Convolutional neural networks (such as 23 layers of convolutional neural networks).After building MR image prediction model, to MR image prediction mould
Type is initialized.Preferably, MR image prediction model is initialized using preset normal distribution, to improve MR image
The initialization effect of prediction model.It is further preferred that the mean value and standard deviation of the used normal distribution of initialization are respectively 0 HeWherein, n indicates the number of input unit in MR image prediction model, to improve initialization effect.
When constructing MR image prediction model, it is preferable that be connected to batch processing specification layer (Batch behind each convolutional layer
Normalization, abbreviation BN) and amendment linear unit (Rectified Linear Units, abbreviation ReLU), ReLU is mono-
Activation unit of the member as MR image prediction model, to improve the training effect of MR image prediction model.
When constructing MR image prediction model, and preferably, the left side of U-shaped network structure is criticized per continuous two convolutional layers-
Process layer-ReLU unit is connected to a maximum pond layer (MaxPooling), so that every by primary maximum pond layer CT figure
The size reduction half of picture, subsequent extracted to characteristics of image one times more than upper one layer of characteristics of image extracted, to improve MR
The image characteristics extraction effect of image prediction model.A up-sampling layer is first passed through on the right side of U-shaped network structure
(UpSampling), convolutional layer, BN layers and the ReLU unit for being 2*2 using convolution kernel size, so that input up-sampling layer
One times of image expanded in size, two continuous-BN layers of-ReLU units of convolutional layer are reconnected, are so recycled, output layer is convolution
Core is the convolutional layer of 1*1 size, to improve the MR image prediction effect of MR image prediction model.
It is further preferred that in MR image prediction model, between the network layer of U-shaped network structure left and right side same level
There are connections, are attached for example, by using cascaded structure (Concatenat), so that the last one ReLU unit of left side be exported
Characteristics of image be added right side image processing process in, to reduce the loss of characteristics of image.
As illustratively, the U-shaped network structure in Fig. 2 is the topology example of MR image prediction model, the left side network left side
256*256,128*128,64*64,32*32 mark and that network the right in right side marks indicate that the image of respective wire network layers is big
Small, each rectangle indicates corresponding image set in U-shaped network structure, and number (such as 1, the 64,128 etc.) expression above rectangle mentions
The characteristics of image quantity taken.
In step s 103, according to training sample set, Training is carried out to the MR image prediction model after initialization,
Obtain trained MR image prediction model.
In embodiments of the present invention, the training sample (i.e. CT image) training sample concentrated inputs MR image prediction mould
Type obtains the output of MR image prediction model, the corresponding MR forecast image of training sample, is predicted according to the corresponding MR of training sample
Image and the corresponding label of training sample (i.e. the corresponding MR image of CT image), pass through preset loss function and preset optimization
Algorithm is adjusted the parameter of MR image prediction model, i.e., is trained to MR image prediction model.
Preferably, optimization algorithm is adaptive moments estimation (adaptive moment estimation, abbreviation Adam) calculation
Method, loss function are mean absolute error (Mean Absolute Error, abbreviation MAE) function, to effectively improve MR figure
As the training effect of prediction model, quality and accuracy that MR image prediction model generates MR forecast image are improved.Wherein, average
Absolute error function representation are as follows:
Wherein, nsamplesIndicate that training sample concentrates the number of training sample
Amount, y indicate the MR image as label in training sample set,Indicate the MR forecast image of MR image prediction model output, yi
Indicate i-th MR image as label in training sample set,Indicate the MR forecast image of MR image prediction model output.
It is further preferred that the learning rate of adaptive moments estimation algorithm is 1*10-2, to improve the instruction of MR image prediction model
Practice effect.
In embodiments of the present invention, by trained MR image prediction model, the CT image that user inputs can be carried out
Processing, prediction obtain the corresponding MR image of the CT image, to greatly reduce MR image prediction by MR image prediction model
Step complexity, improve the effect of MR image prediction.
As illustratively, during the experiment, using it is 4036 big it is small be 256*256 brain CT image and size be
The brain T2 weighted MR image of 256*256, is trained MR image prediction model, can obtain preferable MR forecast image,
For experimental result as shown in figure 3, in Fig. 3, the original CT of the first row is the brain CT image of acquisition, and the original MR of the second row is to adopt
The brain T2 weighted MR image of collection, the prediction MR of the third line be obtained by MR image prediction model prediction, corresponding MR it is pre-
Altimetric image.Compare the image of the second row and the third line, it can be seen that the prediction effect of MR image prediction model is preferable.
In embodiments of the present invention, it by CT image and the corresponding MR image construction training sample set of CT image, constructs and first
Beginningization MR image prediction model is trained according to MR image prediction model of the training sample set to initialization, is trained
MR image prediction model, MR image prediction model be U-shaped network structure convolutional neural networks, thus pass through trained MR
Image prediction model, predicts corresponding MR image in the case where only scanning computed tomography image, and forecasting efficiency is higher and prediction obtains
MR picture quality is preferable, and then doctor is helped to understand the soft tissue situation of region of interest, assists diagnosis, can also be in radiotherapy
Position is put using the MR visual aids of prediction in journey, while also reduce patient sees a doctor cost.
Embodiment two:
Fig. 4 shows the structure of the training device of MR image prediction model provided by Embodiment 2 of the present invention, for the ease of
Illustrate, only parts related to embodiments of the present invention are shown, including:
Training sample generation unit 41, it is corresponding for from sample data gathered in advance, obtaining CT image and CT image
MR image training sample set is generated according to CT image and the corresponding MR image of CT image.
In embodiments of the present invention, the preferable sample number of image quality can be acquired from existing medical image library in advance
According to including CT image and the corresponding MR figure of CT image to improve the training effect of subsequent MR image prediction model, in sample data
Picture, by these CT images MR image construction training sample set corresponding with these CT images.
Preferably, when by CT image and CT image corresponding MR image construction training sample set, CT image and CT are schemed
As the progress image registration of corresponding MR image, the CT image after registration is set to the training sample of training sample concentration, by CT
Image is corresponding, the MR image after registration is set as the corresponding label of training sample, thus for subsequent MR image prediction model
Training is prepared, and the training effect of subsequent MR image prediction model is improved.
Prediction model construction unit 42 for constructing MR image prediction model, and carries out MR image prediction model initial
Change, MR image prediction model is the convolutional neural networks of U-shaped network structure.
In embodiments of the present invention, MR image prediction model is constructed, wherein MR image prediction model is U-shaped network structure
Convolutional neural networks.After building MR image prediction model, MR image prediction model is initialized.
When constructing MR image prediction model, it is preferable that be connected to batch processing specification layer and amendment behind each convolutional layer
Linear unit, using ReLU unit as the activation unit of MR image prediction model, to improve the training effect of MR image prediction model
Fruit.
When constructing MR image prediction model, and preferably, the left side of U-shaped network structure is criticized per continuous two convolutional layers-
Process layer-ReLU unit is connected to a maximum pond layer, so that every size reduction by primary maximum pond layer CT image
Half, subsequent extracted to characteristics of image one times more than upper one layer of characteristics of image extracted, to improve MR image prediction model
Image characteristics extraction effect.A up-sampling layer is first passed through on the right side of U-shaped network structure, is 2*2 using convolution kernel size
Convolutional layer, BN layers and ReLU unit so that input up-sampling one times of image expanded in size of layer, reconnect two it is continuous
- BN layers of-ReLU unit of convolutional layer so recycle, and output layer is the convolutional layer that convolution kernel is 1*1 size, pre- to improve MR image
Survey the MR image prediction effect of model.
It is further preferred that in MR image prediction model, between the network layer of U-shaped network structure left and right side same level
There are connections, are attached for example, by using cascaded structure, so that the characteristics of image that the last one ReLU unit of left side exports be added
In the image processing process for entering right side, to reduce the loss of characteristics of image.
Prediction model training unit 43, for being carried out to the MR image prediction model after initialization according to training sample set
Training obtains trained MR image prediction model.
In embodiments of the present invention, training sample training sample concentrated inputs MR image prediction model, obtains MR figure
Exported as prediction model, the corresponding MR forecast image of training sample, according to the corresponding MR forecast image of training sample and training
The corresponding label of sample adjusts the parameter of MR image prediction model by preset loss function and preset optimization algorithm
It is whole, i.e., MR image prediction model is trained.
Preferably, optimization algorithm is adaptive moments estimation algorithm, and loss function is mean absolute error function, thus effectively
Ground improves the training effect of MR image prediction model, improves MR image prediction model and generates the quality of MR forecast image and accurate
Degree.Wherein, mean absolute error function representation are as follows:
Wherein, nsamplesIndicate that training sample concentrates the number of training sample
Amount, y indicate the MR image as label in training sample set,Indicate the MR forecast image of MR image prediction model output, yiTable
It is shown as i-th MR image of label in training sample set,Indicate the MR forecast image of MR image prediction model output.
It is further preferred that the learning rate of adaptive moments estimation algorithm is 1*10-2, to improve the instruction of MR image prediction model
Practice effect.
Preferably, prediction model construction unit 42 includes:
Normal distribution initialization unit 521 is used for according to preset normal distribution, to the convolution kernel of MR image prediction model
It is initialized.
In embodiments of the present invention, MR image prediction model is initialized using preset normal distribution, to improve
The initialization effect of MR image prediction model.It is further preferred that the mean value and standard difference of the used normal distribution of initialization
It Wei not 0 HeWherein, n indicates the number of input unit in MR image prediction model, to improve initialization effect.
Preferably, prediction model training unit 43 includes:
Training unit 531, for according to pretreated training sample set, preset optimization algorithm and preset
Loss function carries out Training to MR image prediction model, and optimization algorithm is adaptive moments estimation algorithm, and loss function is
Mean absolute error function.
In embodiments of the present invention, it by CT image and the corresponding MR image construction training sample set of CT image, constructs and first
Beginningization MR image prediction model is trained according to MR image prediction model of the training sample set to initialization, is trained
MR image prediction model, MR image prediction model be U-shaped network structure convolutional neural networks, thus pass through trained MR
Image prediction model, predicts corresponding MR image in the case where only scanning computed tomography image, and forecasting efficiency is higher and prediction obtains
MR picture quality is preferable, and then doctor is helped to understand the soft tissue situation of region of interest, assists diagnosis, can also be in radiotherapy
Position is put using the MR visual aids of prediction in journey, while also reduce patient sees a doctor cost.
In embodiments of the present invention, each unit of the training device of MR image prediction model can be by corresponding hardware or software
Unit realizes that each unit can be independent soft and hardware unit, also can integrate as a soft and hardware unit, herein not to
The limitation present invention.
Embodiment three:
Fig. 6 shows the structure of the device of the offer of the embodiment of the present invention three, for ease of description, illustrates only and the present invention
The relevant part of embodiment.
The device 6 of the embodiment of the present invention includes processor 60, memory 61 and is stored in memory 61 and can locate
The computer program 62 run on reason device 60.The processor 60 is realized in above method embodiment when executing computer program 62
Step, such as step S101 to S104 shown in FIG. 1.Alternatively, processor 60 realizes above-mentioned apparatus when executing computer program 62
The function of each unit in embodiment, such as the function of unit 41 to 44 shown in Fig. 4.
In embodiments of the present invention, it by CT image and the corresponding MR image construction training sample set of CT image, constructs and first
Beginningization MR image prediction model is trained according to MR image prediction model of the training sample set to initialization, is trained
MR image prediction model, MR image prediction model be U-shaped network structure convolutional neural networks, thus pass through trained MR
Image prediction model, predicts corresponding MR image in the case where only scanning computed tomography image, and forecasting efficiency is higher and prediction obtains
MR picture quality is preferable, and then doctor is helped to understand the soft tissue situation of region of interest, assists diagnosis, can also be in radiotherapy
Position is put using the MR visual aids of prediction in journey, while also reduce patient sees a doctor cost.
Example IV:
In embodiments of the present invention, a kind of computer readable storage medium is provided, which deposits
Computer program is contained, the step in above method embodiment is realized when which is executed by processor, for example, Fig. 1
Shown step S101 to S104.Alternatively, realizing each list in above-mentioned apparatus embodiment when the computer program is executed by processor
The function of member, such as the function of unit 41 to 44 shown in Fig. 4.
In embodiments of the present invention, it by CT image and the corresponding MR image construction training sample set of CT image, constructs and first
Beginningization MR image prediction model is trained according to MR image prediction model of the training sample set to initialization, is trained
MR image prediction model, MR image prediction model be U-shaped network structure convolutional neural networks, thus pass through trained MR
Image prediction model, predicts corresponding MR image in the case where only scanning computed tomography image, and forecasting efficiency is higher and prediction obtains
MR picture quality is preferable, and then doctor is helped to understand the soft tissue situation of region of interest, assists diagnosis, can also be in radiotherapy
Position is put using the MR visual aids of prediction in journey, while also reduce patient sees a doctor cost.
The computer readable storage medium of the embodiment of the present invention may include can carry computer program code any
Entity or device, recording medium, for example, the memories such as ROM/RAM, disk, CD, flash memory.
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the invention, all in essence of the invention
Made any modifications, equivalent replacements, and improvements etc., should all be included in the protection scope of the present invention within mind and principle.
Claims (10)
1. a kind of training method of MR image prediction model, which is characterized in that the method includes the following steps:
From sample data gathered in advance, CT image and the corresponding MR image of the CT image are obtained, according to the CT image
MR image corresponding with the CT image generates training sample set；
MR image prediction model is constructed, and the MR image prediction model is initialized, the MR image prediction model is U
The convolutional neural networks of type network structure；
According to the training sample set, Training is carried out to the MR image prediction model after initialization, is trained
Good MR image prediction model.
2. the method as described in claim 1, which is characterized in that schemed according to the CT image and the corresponding MR of the CT image
The step of picture, generation training sample set, comprising:
CT image MR image corresponding with the CT image is registrated；
The training sample that the training sample is concentrated is set by the CT image after registration, by the CT image after registration
Corresponding MR image is set as the training sample and concentrates the corresponding label of training sample.
3. the method as described in claim 1, which is characterized in that the step of MR image prediction model is initialized,
Include:
According to preset normal distribution, the convolution kernel of the MR image prediction model is initialized.
4. method as claimed in claim 3, which is characterized in that the mean value and standard deviation of the normal distribution are respectively 0 HeWherein, the n indicates the number of input unit in the MR image prediction model.
5. the method as described in claim 1, which is characterized in that carried out prison to the MR image prediction model after initialization
Supervise and instruct experienced step, comprising:
According to the pretreated training sample set, preset optimization algorithm and preset loss function, to the MR image
Prediction model carries out Training, and the optimization algorithm is adaptive moments estimation algorithm, and the loss function is average absolute
Error function.
6. a kind of training device of MR image prediction model, which is characterized in that described device includes:
Training sample generation unit, for from sample data gathered in advance, acquisition CT image and the CT image to be corresponding
MR image generates training sample set according to the CT image and the corresponding MR image of the CT image；
Prediction model construction unit is initialized for constructing MR image prediction model, and to the MR image prediction model,
The MR image prediction model is the convolutional neural networks of U-shaped network structure；And
Prediction model training unit, for according to the training sample set, to the MR image prediction model after initialization into
Row Training obtains trained MR image prediction model.
7. device as claimed in claim 6, which is characterized in that the prediction model construction unit includes:
Normal distribution initialization unit, for according to preset normal distribution, to the convolution kernel of the MR image prediction model into
Row initialization.
8. device as claimed in claim 6, which is characterized in that prediction model training unit includes:
Training unit, for according to the pretreated training sample set, preset optimization algorithm and preset damage
Function is lost, Training is carried out to the MR image prediction model, the optimization algorithm is adaptive moments estimation algorithm, described
Loss function is mean absolute error function.
9. a kind of image processing equipment, including memory, processor and storage are in the memory and can be in the processing
The computer program run on device, which is characterized in that the processor realizes such as claim 1 when executing the computer program
The step of to any one of 5 the method.
10. a kind of computer readable storage medium, the computer-readable recording medium storage has computer program, and feature exists
In when the computer program is executed by processor the step of any one of such as claim 1 to 5 of realization the method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810924191.9A CN109272486B (en) | 2018-08-14 | 2018-08-14 | Training method, device and equipment of MR image prediction model and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810924191.9A CN109272486B (en) | 2018-08-14 | 2018-08-14 | Training method, device and equipment of MR image prediction model and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109272486A true CN109272486A (en) | 2019-01-25 |
| CN109272486B CN109272486B (en) | 2022-07-08 |
Family
ID=65153407
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810924191.9A Active CN109272486B (en) | 2018-08-14 | 2018-08-14 | Training method, device and equipment of MR image prediction model and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109272486B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109785405A (en) * | 2019-01-29 | 2019-05-21 | 子威·林 | The method for generating quasi- CT image using the multivariate regression of multiple groups magnetic resonance image |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106373109A (en) * | 2016-08-31 | 2017-02-01 | 南方医科大学 | Medical image modal synthesis method |
| CN107481185A (en) * | 2017-08-24 | 2017-12-15 | 深圳市唯特视科技有限公司 | A kind of style conversion method based on video image optimization |
| CN108171320A (en) * | 2017-12-06 | 2018-06-15 | 西安工业大学 | A kind of image area switching network and conversion method based on production confrontation network |
| US20180197317A1 (en) * | 2017-01-06 | 2018-07-12 | General Electric Company | Deep learning based acceleration for iterative tomographic reconstruction |
| CN108304755A (en) * | 2017-03-08 | 2018-07-20 | 腾讯科技（深圳）有限公司 | The training method and device of neural network model for image procossing |
-
2018
- 2018-08-14 CN CN201810924191.9A patent/CN109272486B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106373109A (en) * | 2016-08-31 | 2017-02-01 | 南方医科大学 | Medical image modal synthesis method |
| US20180197317A1 (en) * | 2017-01-06 | 2018-07-12 | General Electric Company | Deep learning based acceleration for iterative tomographic reconstruction |
| CN108304755A (en) * | 2017-03-08 | 2018-07-20 | 腾讯科技（深圳）有限公司 | The training method and device of neural network model for image procossing |
| CN107481185A (en) * | 2017-08-24 | 2017-12-15 | 深圳市唯特视科技有限公司 | A kind of style conversion method based on video image optimization |
| CN108171320A (en) * | 2017-12-06 | 2018-06-15 | 西安工业大学 | A kind of image area switching network and conversion method based on production confrontation network |
Non-Patent Citations (1)
| Title |
|---|
| CHENG-BIN JIN 等: "Deep CT to MR Synthesis using Paired and Unpaired Data", 《HTTPS://ARXIV.ORG/ABS/1805.10790V1》 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109785405A (en) * | 2019-01-29 | 2019-05-21 | 子威·林 | The method for generating quasi- CT image using the multivariate regression of multiple groups magnetic resonance image |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109272486B (en) | 2022-07-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| JP6960509B2 (en) | Neural network for generating synthetic medical images | |
| Dinkla et al. | Dosimetric evaluation of synthetic CT for head and neck radiotherapy generated by a patch‐based three‐dimensional convolutional neural network | |
| Kida et al. | Cone beam computed tomography image quality improvement using a deep convolutional neural network | |
| US10709394B2 (en) | Method and system for 3D reconstruction of X-ray CT volume and segmentation mask from a few X-ray radiographs | |
| US11744465B2 (en) | Method and program for generating three-dimensional brain map | |
| WO2018120644A1 (en) | Blood vessel extraction method and system | |
| US10149987B2 (en) | Method and system for generating synthetic electron density information for dose calculations based on MRI | |
| WO2019211307A1 (en) | Modality-agnostic method for medical image representation | |
| US11080895B2 (en) | Generating simulated body parts for images | |
| CN109215014B (en) | Training method, device and equipment of CT image prediction model and storage medium | |
| WO2020214911A1 (en) | Method and system for generating attenuation map from spect emission data based upon deep learning | |
| EP2886057B1 (en) | Medical imaging apparatus and method of reconstructing medical image | |
| CN114881848A (en) | Method for converting multi-sequence MR into CT | |
| JP2022530108A (en) | Generation of synthetic electron density images from magnetic resonance images | |
| US11534623B2 (en) | Determining at least one final two-dimensional image for visualizing an object of interest in a three dimensional ultrasound volume | |
| Xie et al. | Generation of contrast-enhanced CT with residual cycle-consistent generative adversarial network (Res-CycleGAN) | |
| JP2018535746A (en) | Tissue classification method, computer program, and magnetic resonance imaging system | |
| CN110852993B (en) | Imaging method and device under action of contrast agent | |
| WO2023160720A1 (en) | Methods, systems, and storage mediums for image generation | |
| CN109272486A (en) | Training method, device, equipment and the storage medium of MR image prediction model | |
| CN116168097A (en) | Method, device, equipment and medium for constructing CBCT sketching model and sketching CBCT image | |
| CN115861464A (en) | Pseudo CT (computed tomography) synthesis method based on multimode MRI (magnetic resonance imaging) synchronous generation | |
| CN113052840B (en) | Processing method based on low signal-to-noise ratio PET image | |
| Largent et al. | Pseudo-CT generation for Mri-only radiotherapy: Comparative study between a generative adversarial network, a U-net network, a patch-based, and an atlas based methods | |
| KR102658413B1 (en) | Apparatus and method of extracting biliary tree image using CT image 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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |