CN109242849A - Medical image processing method, device, system and storage medium - Google Patents
Medical image processing method, device, system and storage medium Download PDFInfo
- Publication number
- CN109242849A CN109242849A CN201811126424.7A CN201811126424A CN109242849A CN 109242849 A CN109242849 A CN 109242849A CN 201811126424 A CN201811126424 A CN 201811126424A CN 109242849 A CN109242849 A CN 109242849A
- Authority
- CN
- China
- Prior art keywords
- neural network
- network model
- image
- detection
- segmentation
- 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
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
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- 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]
Abstract
The present invention relates to a kind of processing method of medical image, device, system and storage mediums.This method comprises: obtaining detection image, the detection image includes the image in black substance region and/or rubrum region；Segmentation neural network model is obtained, operation is split to the detection image using the segmentation neural network model, obtains target mask image；Detection neural network model is obtained, the target mask image is detected using the detection neural network model, obtains testing result.This method can make more accurate detection to detection image as a result, further increasing the accuracy rate of testing result according to trained detection neural network model.
Description
Technical field
The present invention relates to nerual network technique field, more particularly to a kind of medical image processing method, device, system and
Storage medium.
Background technique
Parkinson's disease (Parkinson ' s disease, PD) is a kind of common nervous system degeneration disease, for making a definite diagnosis
For the patient of Parkinson's disease, it is contemplated that the service life is about 7 to 14 years, therefore the treatment in later period is closed in the diagnosis of Parkinson's disease very much
Key.Patient history, clinical symptoms and drug response clinically are relied primarily on to the diagnosis of Parkinson disease, and by brain image
Auxiliary diagnosis is carried out to Parkinson's disease.
In traditional technology, magnetic resonance (Magnetic Resonance Imaging, MRI) T1 weighted imaging skill is mainly used
Art obtains the MRI T1 image of diagnosis subject brain, and MRI T1 image mainly reflects brain structure, by comparing the brain structure and is good for
Difference between the brain structure of health people, to diagnose whether patient is PD.
But the difference that the brain overall structure and Healthy People of patient Parkinson are considerable there is no naked eyes, therefore using biography
The image processing method of system can not provide good diagnosis basis to the diagnosis of Parkinson's disease, and auxiliaring effect is poor.
Summary of the invention
Based on this, it is necessary to diagnosis well can not be provided the diagnosis of Parkinson's disease for traditional image processing method
Foundation, the problem of auxiliaring effect difference provide processing method, device, system and the storage medium of a kind of medical image.
In a first aspect, the present invention provides a kind of medical image processing method, this method comprises:
Detection image is obtained, the detection image includes the image in black substance region and/or rubrum region；
Segmentation neural network model is obtained, behaviour is split to the detection image using the segmentation neural network model
Make, obtains target mask image；
Detection neural network model is obtained, the target mask image is examined using the detection neural network model
It surveys, obtains testing result.
Medical image processing method provided in this embodiment, computer equipment are available including black substance region and/or red
The detection image of core region, and segmentation neural network model is used to be split operation to detection image to obtain target mask figure
Then picture detects target mask image using detection neural network model, to obtain testing result, passes through the detection
As a result doctor can be assisted to judge whether test object suffers from Parkinsonian symptoms.On the one hand, what the present embodiment utilized be and pa gold
The maximally related detection image comprising black substance and/or rubrum of gloomy illness, therefore the accuracy rate of testing result can be improved；Another party
Face, computer equipment detect the detection image of test object using trained detection neural network model,
That is computer equipment is using inspection of the neural network model to test object for having been able to correct output test result
What altimetric image was detected, thus, computer equipment can be according to trained detection neural network model to test object
Detection image makes more accurate detection as a result, further improving the accuracy rate of testing result.
In one of the embodiments, before the acquisition detects neural network model, the method also includes:
Training dataset is obtained, the training dataset includes multipair training data, and every a pair of training data includes one
Training mask images and the corresponding true tag of the trained mask images；
Preset initial detecting neural network model is trained according to the training dataset, obtains the detection mind
Through network model.
Medical image processing method provided in this embodiment, the available training dataset of computer equipment, and according to instruction
Practice data set to be trained preset initial detecting neural network model, and then obtains detection neural network model.Such
To detection neural network model be the trained neural network model that can export correct result, therefore can be with utility
Above-mentioned trained detection neural network model is that accurate testing result can be exported when detecting to detection image, is mentioned
The accuracy rate of high detection result.
In one of the embodiments, it is described according to the training dataset to preset initial detecting neural network model
It is trained, obtains the detection neural network model, comprising:
The trained mask images are input to the initial detecting neural network model, obtain initial labels；
Determine whether the initial labels and the error of the true tag are less than preset threshold；And in response to the mistake
Difference is less than the judgement of preset threshold, and the initial detecting neural network model is appointed as the detection neural network model.
Training mask images can be input to initially by medical image processing method provided in this embodiment, computer equipment
Neural network model is detected, initial labels are obtained；Determine whether the error of initial labels and true tag is less than preset threshold；With
And it is less than the judgement of preset threshold in response to error, initial detecting neural network model is appointed as detection neural network model.
In the present embodiment, computer equipment can be according to initial labels and the error of the corresponding true tag of training mask images and default
It is that condition is compared as a result, adjustment initial detecting neural network model network parameter, until initial detecting neural network mould
Until the error of the initial labels of type output true tag corresponding with training mask images meets preset condition, i.e., initially through examining
It, in this way can be with the above-mentioned trained detection nerve of utility until survey neural network model can export correct testing result
Network model can export accurate testing result when detecting to detection image, improve the accuracy rate of testing result.
The initial detecting neural network model includes feature extraction group and sorting group in one of the embodiments,；Institute
Stating feature extraction group includes the first convolutional layer, and the sorting group includes the second convolutional layer, pond layer and full articulamentum；It is described by institute
It states trained mask images and is input to the initial detecting neural network model, obtain initial labels, comprising:
Based on the feature extraction group, the corresponding detection characteristic pattern of the trained mask images is obtained；
Based on the sorting group and the detection characteristic pattern, the corresponding initial mark of the trained mask images is obtained
Label.
Medical image processing method provided in this embodiment, computer equipment can be according to initial detecting neural network models
Feature extraction group and sorting group training mask images convolution, pond and full connection etc. are handled, to obtain initial labels.
On the one hand, in above-mentioned multiple first convolutional layers the 2nd the first convolutional layer to each of the last one the first convolutional layer first
It may include one batch of normalization layer before convolutional layer, this batch of normalization layer can accelerate the study of initial detecting neural network model
Efficiency, to improve the speed that the training of initial detecting neural network model is become to detection neural network model.On the other hand, convenient
Computer equipment calculates the error between its true tag corresponding with training mask images using the initial labels, thus to first
Beginning, it is preferably trained to detect neural network model progress.
It is described in one of the embodiments, that the detection image is divided using preset segmentation neural network model
Operation is cut, target mask image is obtained, comprising:
The detection image is input to the segmentation neural network model, obtains Target Segmentation image；
According to the product of the Target Segmentation image and the detection image, the target mask image is obtained.
Medical image processing method provided in this embodiment, computer equipment can will test image and be input to segmentation nerve
Network model obtains Target Segmentation image, and according to the product of Target Segmentation image and detection image, obtains target mask figure
Picture, since computer equipment can carry out feature extraction to detection image using segmentation neural network model, to obtain target mask
Image, the target mask image with higher segmentation precision available in this way, so that utilizing the segmentation nerve net
When the target mask image that network model obtains carries out image detection, the accuracy rate of detection is improved.
The segmentation neural network model is V-Net model, U-Net model, FCN mould in one of the embodiments,
Any one in type, GAN model and the dense piece of neural network model combined, the multiple operational groups of the dense piece of correspondence,
Each operational group includes the process of convolution carried out according to sequencing, activation handles and process of convolution again.
The segmentation neural network model includes downsampling unit and up-sampling unit, institute in one of the embodiments,
Stating downsampling unit includes multiple dense piece for feature extraction, and the up-sampling unit includes multiple for feature reconstruction
It is dense piece, described to carry out skip floor connection for dense piece of feature extraction and between dense piece described for feature reconstruction.
Second aspect, the present invention provide a kind of medical image processing devices, and described device includes:
First obtains module, and for obtaining detection image, the detection image includes black substance region and/or rubrum region
Image；
First determining module, for obtaining segmentation neural network model, using the segmentation neural network model to described
Detection image is split operation, obtains target mask image；
Second determining module, for obtaining detection neural network model, using the detection neural network model to described
Target mask image is detected, and testing result is obtained.
The third aspect, the present invention provide a kind of magic magiscan, including image scanner and computer equipment,
In, the computer equipment includes memory and processor, and the memory is stored with computer program, which is characterized in that institute
It states when processor executes the computer program and performs the steps of
Detection image is obtained, the detection image includes the image in black substance region and/or rubrum region；
Segmentation neural network model is obtained, behaviour is split to the detection image using the segmentation neural network model
Make, obtains target mask image；
Detection neural network model is obtained, the target mask image is examined using the detection neural network model
It surveys, obtains testing result.
Fourth aspect, the present invention provide a kind of computer readable storage medium, are stored thereon with computer program, the meter
Calculation machine program performs the steps of when being executed by processor
Detection image is obtained, the detection image includes the image in black substance region and/or rubrum region；
Segmentation neural network model is obtained, behaviour is split to the detection image using the segmentation neural network model
Make, obtains target mask image；
Detection neural network model is obtained, the target mask image is examined using the detection neural network model
It surveys, obtains testing result.
Image processing apparatus, system and storage medium provided in this embodiment, enable to computer equipment available
Detection image including black substance region and/or rubrum region, and detection image is split using segmentation neural network model
Then operation detects target mask image using detection neural network model with obtaining target mask image, thus
To testing result, doctor can be assisted to judge whether test object suffers from Parkinsonian symptoms by the testing result.On the one hand, originally
What embodiment utilized be with the maximally related detection image comprising black substance and/or rubrum of Parkinsonian symptoms, therefore inspection can be improved
Survey the accuracy rate of result；On the other hand, computer equipment is using trained detection neural network model to test object
Detection image detected, that is to say, that computer equipment is using the nerve for having been able to correct output test result
Network model detects the detection image of test object, thus, computer equipment can be according to trained detection mind
More accurate detection is made through detection image of the network model to test object as a result, further improving the standard of testing result
True rate.
Detailed description of the invention
Fig. 1 is the magic magiscan structural schematic diagram that one embodiment provides；
Fig. 2 is the medical image processing method flow diagram that one embodiment provides；
Fig. 3 is the medical image processing method flow diagram that another embodiment provides；
Fig. 4 is the medical image processing method flow diagram that another embodiment provides；
Fig. 5 is the medical image processing method flow diagram that another embodiment provides；
Fig. 6 is the medical image processing method flow diagram that another embodiment provides；
Fig. 6 a is that the segmentation Artificial Neural Network Structures that the U-Net model that one embodiment provides is combined with dense piece show
It is intended to；
Fig. 6 b is dense piece of the structural schematic diagram that one embodiment provides；
Fig. 7 is the medical image processing method flow diagram that another embodiment provides；
Fig. 8 is the medical image processing devices structural schematic diagram that one embodiment provides；
Fig. 9 is the medical image processing devices structural schematic diagram that another embodiment provides；
Figure 10 is the medical image processing devices structural schematic diagram that another embodiment provides；
Figure 11 is the medical image processing devices structural schematic diagram that another embodiment provides；
Figure 12 is the medical image processing devices structural schematic diagram that another embodiment provides；
Figure 13 is the schematic diagram of internal structure for the computer equipment that one embodiment provides.
Specific embodiment
In order to make the objectives, technical solutions, and advantages of the present invention clearer, pass through following embodiments and combine attached
Figure, the further description of technical solution in the embodiment of the present invention.It should be appreciated that specific embodiment described herein
Only to explain the present invention, it is not intended to limit the present invention.
Medical image processing method provided by the invention can be applied to magic magiscan as shown in Figure 1.It should
System includes image scanner 102 and computer equipment 104.The image scanner 102 can be magnetic resonance scanner, can also be with
It is CT scanner etc., for image scanner 102 for obtaining detection image, which can be terminal, such as personal meter
Calculation machine, notebook and tablet computer etc..The computer equipment 104 include by system bus connect processor, memory,
Network interface, display screen and input unit.Wherein, the processor of the computer equipment 104 is for providing calculating and control ability.
The memory of the computer equipment 104 includes non-volatile memory medium, built-in storage.The non-volatile memory medium is stored with
Operating system and computer program.The built-in storage is the fortune of the operating system and computer program in non-volatile memory medium
Row provides environment.The network interface of the computer equipment 104 is used to communicate with external terminal by network connection.The computer
The medical image processing method that following examples of the present invention provide may be implemented when program is executed by processor.
It is understood that above-mentioned magic magiscan can be used for the medicine figure to the organ or tissue of person under test
As being split, to assist doctor to diagnose, but the system itself can not directly export diagnostic result.
Currently, the cause of disease of Parkinson's disease is difficult to determine so far, but by the brain image of patient, doctor can be helped to combine disease
People's medical history, clinical symptoms and drug response etc. carry out auxiliary diagnosis.In traditional technology, generally obtained using MRI T1 imaging technique
The brain structural images of patient are taken, and according to the brain structural images and the difference between the brain structural images of Healthy People to Parkinson
Patient carries out auxiliary diagnosis, and still, the brain overall structure and Healthy People of the patient Parkinson difference considerable there is no naked eyes is led
By MRI T1 image progress auxiliary diagnosis, the effect is relatively poor for cause.Medical image processing method provided by the invention device, is
System and readable storage medium storing program for executing aim to solve the problem that the technical problem as above of traditional technology.
Fig. 2 is the medical image processing method flow diagram that one embodiment provides.What is involved is calculating for the present embodiment
Machine equipment is using segmentation neural network model and detects neural network model to the detection image progress image segmentation of acquisition and inspection
The specific implementation process of survey.As shown in Fig. 2, this method comprises:
S201 obtains detection image, and the detection image includes the image in black substance region and/or rubrum region.
In the present embodiment, when whether the acquisition test object of above-mentioned detection image is to need to diagnose with Parkinsonian symptoms
When patient, above-mentioned detection image can be quantitative magnetic susceptibility figure (Quantitative susceptibility mapping,
QSM), it is also possible to MRI T2 image, can also is other images including black substance region and/or rubrum region, the present embodiment
Without limitation to the classification of detection image.Optionally, above-mentioned detection image can be gray level image, be also possible to color image,
It can also be bianry image.Optionally, above-mentioned detection image can be two dimensional image, be also possible to 3-D image.It needs to illustrate
, when the acquisition object of detection image without limitation when, above-mentioned detection image for animal, plant or other can need to detect
Image.
S202 obtains segmentation neural network model, is carried out using the segmentation neural network model to the detection image
Cutting operation obtains target mask image.
Specifically, above-mentioned segmentation neural network model includes multiple segmentation task process layers, which can
To be convolutional layer, active coating, pond layer, down-sampling layer or up-sampling layer etc., the type for dividing task process layer is different, computer
Equipment is also different to the processing operation of the detection image of acquisition, for example, detection image enters at the corresponding segmentation task of convolutional layer
When managing layer, computer equipment can carry out process of convolution operation to the detection image.Computer equipment is according to segmentation neural network mould
Type after the processing of multiple segmentation task process layers, can export the corresponding target of the detection image and cover to above-mentioned detection image
Mould image, the target mask image include the image in black substance region and/or rubrum region.
S203 obtains detection neural network model, using the detection neural network model to the target mask image
It is detected, obtains testing result.
Specifically, above-mentioned detection neural network model is to be carried out according to training dataset and initial detecting neural network model
Neural network model after training, the detection neural network model include multiple Detection task process layers, and computer equipment obtains
After above-mentioned target mask image, the target is covered using the Detection task process layer that trained detection neural network model includes
Mould image carries out detection processing, can export corresponding testing result automatically.Optionally, above-mentioned testing result can be above-mentioned inspection
Altimetric image is the image of patient Parkinson or above-mentioned detection image is the image for not suffering from the Healthy People of Parkinsonian symptoms, above-mentioned inspection
Survey result can also be that the corresponding test object of above-mentioned detection image is probability or above-mentioned detection image with Parkinsonian symptoms
Corresponding test object is the probability of Healthy People, and above-mentioned testing result can also suffer from pa gold with preset for above-mentioned detection image
The likelihood of the image of the patient of gloomy illness.
Medical image processing method provided in this embodiment, computer equipment are available including black substance region and/or red
The detection image of core region, and segmentation neural network model is used to be split operation to detection image to obtain target mask figure
Then picture detects target mask image using detection neural network model, to obtain testing result, passes through the detection
As a result whether Parkinsonian symptoms can be suffered from auxiliary judgment test object.On the one hand, what the present embodiment utilized is and Parkinson's disease
The maximally related detection image comprising black substance and/or rubrum of disease, therefore the accuracy rate of testing result can be improved；On the other hand,
Computer equipment detects the detection image of test object using trained detection neural network model, that is,
It says, computer equipment is using detection figure of the neural network model to test object for having been able to correct output test result
As being detected, thus, computer equipment can be according to the trained detection for detecting neural network model to test object
Image makes more accurate detection as a result, further improving the accuracy rate of testing result.
Fig. 3 is the medical image processing method flow diagram that another embodiment provides.What is involved is meters for the present embodiment
It calculates machine equipment and obtains the detailed process of detection neural network model using training dataset.On the basis of the above embodiments, may be used
Choosing, before " obtaining detection neural network model " in above-mentioned S203, the above method further include:
S301 obtains training dataset, and the training dataset includes multipair training data, and every a pair of training data includes
One trained mask images and the corresponding true tag of the trained mask images.
Specifically, above-mentioned training dataset may include multipair training data, wherein it is every a pair of training data may include
One training mask images containing black substance region and/or rubrum region true tag corresponding with the training mask images.It can
Choosing, the acquisition modes of above-mentioned trained mask images can be obtained according to the acquisition modes of above-mentioned target mask image, can also be with
It is obtained, can also be hooked by way of handmarking according to other image partition methods such as Threshold segmentation or semi-automatic partition method
The mode for drawing above-mentioned trained black substance region and/or training rubrum region obtains, and the present embodiment is to above-mentioned trained mask images
Acquisition modes are without limitation.
Optionally, the corresponding true tag of above-mentioned trained mask images is for characterizing whether the training mask images are pa gold
The image of gloomy patient, for example, the image that this is the patient with parkinsonism can be indicated with true tag 1, with true tag 0
Indicate that the training mask images are the image of Healthy People.
S302 is trained preset initial detecting neural network model according to the training dataset, obtains described
Detect neural network model.
Specifically, the training mask images that training data is concentrated can be input to preset initial detecting by computer equipment
Neural network model.Optionally, computer equipment can be the acquisition initial network parameter by way of random initializtion, can also
To receive specified initial network parameter.Computer equipment is by constantly adjusting network parameter such as weight, to preset initial inspection
It surveys neural network model to be trained, the correct initial detecting neural network model of obtained training is above-mentioned detection nerve net
Network model.
Medical image processing method provided in this embodiment, the available training dataset of computer equipment, and according to instruction
Practice data set to be trained preset initial detecting neural network model, and then obtains detection neural network model.Such
To detection neural network model be the trained neural network model that can export correct result, therefore can be with utility
Above-mentioned trained detection neural network model can export accurate testing result when detecting to detection image, improve
The accuracy rate of testing result.
Fig. 4 is the medical image processing method flow diagram that another embodiment provides.What is involved is meters for the present embodiment
Machine equipment is calculated using the detailed process of training mask images and initial detecting neural network model building detection neural network model.
On the basis of the above embodiments, optionally, above-mentioned S302 may include:
The trained mask images are input to the initial detecting neural network model, obtain initial labels by S401.
Specifically, above-mentioned trained mask images can be input to initial detecting neural network model by computer equipment, benefit
The training mask images are performed corresponding processing with each Detection task process layer in initial detecting neural network model, example
Such as process of convolution, activation processing or down-sampling processing.Using each task process layer pair of initial detecting neural network model
After the training mask images of acquisition are handled, the corresponding initial mark of the available above-mentioned trained mask images of computer equipment
Label.
S402, determines whether the initial labels and the error of the true tag are less than preset threshold；And in response to
The error is less than the judgement of preset threshold, and the initial detecting neural network model is appointed as the detection neural network mould
Type.
Specifically, the error of above-mentioned initial labels true tag corresponding with training mask images can be damaged by cross entropy
It loses function to obtain, can also be obtained by maximum distance loss function, the present embodiment is to initial labels and training mask images pair
The acquisition pattern of the error for the true tag answered is without limitation.Computer equipment may determine that above-mentioned initial labels and training mask
Whether the error of the corresponding true tag of image is less than preset threshold, and the initial labels and above-mentioned trained mask images are corresponding
True tag input preset loss function, obtain the error of initial labels true tag corresponding with trained mask images,
And judge whether the error is less than preset threshold.
If it is not, the network parameter of adjustment initial detecting neural network model, the detection neural network model after being adjusted,
And using detection neural network model adjusted as new initial detecting neural network model, returns and execute detection operation, directly
Until the error of obtained new initial labels and true tag meets preset condition, and by new initial detecting neural network
Model is as detection neural network model.Above-mentioned network parameter can be the weight of preset convolution kernel.
When above-mentioned error is not less than preset threshold, illustrate that computer equipment cannot utilize above-mentioned initial detecting neural network
Model exports correct testing result, needs to be adjusted the initial detecting neural network model, and adjustment process includes: to calculate
Machine equipment adjusts network parameter using stochastic gradient descent method, new network parameter is obtained, by new network parameter reverse transfer
To first task process layer of initial detecting neural network model, detection neural network model after being adjusted, and will adjust
Detection neural network model after whole continues to execute above-mentioned detection operation as new initial detecting neural network model, until
Until the error of obtained new initial labels true tag corresponding with training mask images is less than preset threshold, and this is new
Initial detecting neural network model as the detection neural network model.Optionally, computer equipment, which also can use, criticizes
Measure the network parameter of the adjustment initial detecting neural network model such as gradient descent method or small lot gradient descent method, the present embodiment pair
Adjust the mode of initial detecting neural network model without limitation.Optionally, it includes that test is black that computer equipment, which can also utilize,
Matter region and/or test rubrum region known test result test image, to the detection neural network model after training into
Row test, the detection nerve to verify the accuracy of trained detection neural network model, after being further ensured that above-mentioned training
The Detection accuracy of network model.
Training mask images can be input to initially by medical image processing method provided in this embodiment, computer equipment
Neural network model is detected, initial labels are obtained；Determine whether the error of initial labels and true tag is less than preset threshold；With
And it is less than the judgement of preset threshold in response to error, initial detecting neural network model is appointed as detection neural network model.
In the present embodiment, computer equipment can be according to initial labels and the error of the corresponding true tag of training mask images and default
It is that condition is compared as a result, adjustment initial detecting neural network model network parameter, until initial detecting neural network mould
Until the error of the initial labels of type output true tag corresponding with training mask images meets preset condition, i.e., initially through examining
It, in this way can be with the above-mentioned trained detection nerve of utility until survey neural network model can export correct testing result
Network model can export accurate testing result when detecting to detection image, improve the accuracy rate of testing result.
Fig. 5 is the medical image processing method flow diagram that another embodiment provides.It is above-mentioned initial in the present embodiment
Detecting neural network model includes feature extraction group and sorting group；The feature extraction group includes the first convolutional layer, the classification
Group includes the second convolutional layer, pond layer and full articulamentum；On the basis of the above embodiments, optionally, above-mentioned S301 " by institute
State trained mask images and be input to the initial detecting neural network model, obtain initial labels " may include:
S501 is based on the feature extraction group, obtains the corresponding detection characteristic pattern of the trained mask images.
Specifically, the 2nd the first convolutional layer is every into the last one first convolutional layer in above-mentioned multiple first convolutional layers
It may include one batch normalization layer (Batch Normalization, BN) that this batch of normalization layer can be with before a first convolutional layer
Accelerate the learning efficiency of initial detecting neural network model, the training of initial detecting neural network model is become into detection mind to improve
Speed through network model；The 2nd the first convolutional layer is into the last one first convolutional layer in above-mentioned multiple first convolutional layers
It may include an active coating after each first convolutional layer, the corresponding activation primitive of the active coating can be modified line unit
(Rectified linear unit, ReLU), is also possible to Sigmoid function, can also be tahh activation primitive, this implementation
Example to the corresponding activation primitive of active coating without limitation.By taking the number of the first convolutional layer is 5 as an example, the 2nd to the 5th first volume
It may include BN layers before each first convolutional layer of lamination, each first convolution of the 2nd to the 5th the first convolutional layer
It may include active coating after layer, optionally, the 4th the first convolutional layer can receive the 2nd the first convolutional layer and the 3rd first volume
Characteristic pattern after lamination progress process of convolution is as input.Computer equipment according to above-mentioned multiple first convolutional layers processing sequence
Process of convolution carried out to training mask images, the corresponding detection characteristic pattern of available trained mask images.
S502 is based on the sorting group and the detection characteristic pattern, and it is corresponding described first to obtain the trained mask images
Beginning label.
Specifically, carrying out the to the detection characteristic pattern in the detection characteristic pattern input sorting group that computer equipment will acquire
The process of convolution of two convolutional layers, the detection characteristic pattern that obtains that treated, and detection characteristic pattern carries out pond processing to treated,
Obtain multiple characterization factors.Optionally, the features described above factor can be one of the above-mentioned detection characteristic pattern that an expression is extracted
The numerical value of feature.Computer equipment is using multiple characterization factors as input data, the square obtained after the processing using full articulamentum
Battle array and the multiplied by weight in initial network parameter, obtain initial labels.
Medical image processing method provided in this embodiment, computer equipment can be according to initial detecting neural network models
Feature extraction group and sorting group training mask images convolution, pond and full connection etc. are handled, to obtain initial labels.
On the one hand, in above-mentioned multiple first convolutional layers the 2nd the first convolutional layer to each of the last one the first convolutional layer first
It may include one batch of normalization layer before convolutional layer, this batch of normalization layer can accelerate the study of initial detecting neural network model
Efficiency, to improve the speed that the training of initial detecting neural network model is become to detection neural network model.On the other hand, convenient
Computer equipment calculates the error between its true tag corresponding with training mask images using the initial labels, thus to first
Beginning, it is preferably trained to detect neural network model progress.
Fig. 6 is the medical image processing method flow diagram that another embodiment provides.What is involved is meters for the present embodiment
Calculating machine equipment uses segmentation neural network model to be split operation to detection image to obtain the realization of target mask image
Journey.On the basis of the above embodiments, optionally, above-mentioned S202 may include:
The detection image is input to the segmentation neural network model, obtains Target Segmentation image by S601.
Specifically, computer equipment, which can will test image, is input to segmentation neural network model, segmentation nerve is utilized
Each task process layer is split operation, available Target Segmentation image to the detection image in network model.Optionally,
In above-mentioned Target Segmentation image, the pixel value in black substance region and/or rubrum region can be with 1, and Target Segmentation image removes Substantia Nigra
The pixel value in the region other than domain and/or rubrum region can be 0.
Optionally, above-mentioned segmentation neural network model is V-Net model, U-Net model, FCN (Fully
Convolutional Network, FCN) model, GAN (Generative Adversarial Networks, GAN) model
In any one and the dense piece of neural network model combined, this dense piece multiple operational groups of correspondence, each operational group packet
Include the process of convolution carried out according to sequencing, activation is handled and process of convolution again.Any one of the above network mould
Type and dense piece of combination are the same, are this reality as shown in Figure 6 a by U-Net model for dense piece combines
A kind of segmentation neural network model for combining with dense piece " " of U-Net model of example offer is provided, including input mould group 601,
First dense piece 602, second dense piece 603, dense piece 604 of third, the 4th dense piece 605, the 5th dense piece it is the 606, the 6th thick
Close piece 607 and output mould group 608.Dense piece of each of Fig. 6 b can correspond to multiple operational groups, also, next operational group exists
When carrying out process of convolution, the characteristic pattern of any one operational group of front output can receive, it is assumed that dense piece of correspondence 5 operations
Group, operational group 5 can receive the characteristic pattern of each operational group such as operational group 1, operational group 2, operational group 3 and operational group 4 output
Input feature vector figure as operational group 5.Can so it understand, for each operational group, all operational groups before it
Characteristic pattern all can be as the input of the operational group, and the characteristic pattern of own is then used as the input of all operational groups of postorder.It utilizes
Dense piece can effectively alleviate gradient disappearance problem, reinforce feature propagation, encourage feature multiplexing and greatly reduce participation
The number of parameters of the operations such as convolution, activation, down-sampling and up-sampling could preferably handle input picture in this way, mention
More useful features are taken, and then obtain the Target Segmentation image of segmentation precision more, so that computer equipment will divide nerve net
When the target mask image input detection neural network model that network model obtains, more accurate testing result can be obtained, is improved
The accuracy rate of detection.
Optionally, segmentation neural network model includes downsampling unit and up-sampling unit, and downsampling unit includes multiple
For dense piece of feature extraction, the up-sampling unit includes multiple dense piece for feature reconstruction, is used for feature extraction
Dense piece and between dense piece of feature reconstruction carry out skip floor connection.Optionally, in downsampling unit down-sampling time
The number up-sampled in several and up-sampling unit can be 2 times.As shown in Figure 6 b, in the down-sampling of segmentation neural network model
Unit part, input mould group 601 be used for detection image input, first dense piece 602, second dense piece 603 and third it is dense
Block 604 is used to carry out the characteristic pattern that input mould group 601 exports feature extraction, and computer equipment can be to thick by first
Close piece 602 obtained characteristic pattern and second dense piece 603 obtained characteristic pattern carry out a down-sampling processing respectively；Dividing
The up-sampling unit part of neural network model, the 4th dense piece 605, the 5th dense piece 606 and the 6th dense piece 607 for pair
The characteristic pattern that downsampling unit obtains carries out feature reconstruction, and to by dense piece of the 4th 605 obtained characteristic pattern and the 5th
Dense piece 606 obtained characteristic pattern has carried out a up-sampling treatment respectively, and output mould group 608 is for exporting Target Segmentation figure
Picture.Organizational form of the above-mentioned segmentation neural network model using encoder and decoder, above-mentioned input mould group 601, first
Dense piece 602, second dense piece 603 and dense piece 604 of third corresponds to the encoder operation rank for dividing neural network model
Section, by input dense piece of 604 corresponding convolution of 601, first dense piece 602, second dense piece 603 of mould group and third, activation,
After the iterative operation of down-sampling etc., encoder operation stages terminate, at this point, the feature after the available down-sampling of computer equipment
Figure；Above-mentioned 4th dense piece 605, the 5th dense piece 606, the 6th dense piece 607 and output mould group 608 correspond to segmentation nerve net
Network solution to model code device operation stages, by the 4th dense piece 605, the 5th dense piece 606, the 6th dense piece 607 and output mould
After the iterative operations such as 608 corresponding convolution of group, activation and up-sampling, the decoder operation stage terminates, at this point, computer equipment can
To obtain Target Segmentation image.
Optionally, for being used for the thick of feature reconstruction in dense piece of feature extraction and up-sampling unit in downsampling unit
Skip floor connection can be carried out between close piece, specific implementation can be such that when computer equipment utilizes the 5th dense piece 606 pairs
4th dense piece 605 output characteristic pattern carry out convolution operation when, can directly to the 4th dense piece 605 export characteristic pattern into
Row convolution operation can also spell the characteristic pattern that the characteristic pattern that second dense piece 603 exports is exported with the 5th dense piece 605
It connects, obtains spliced splicing characteristic pattern, and process of convolution is carried out to the splicing characteristic pattern, detection image can be extracted in this way
More features, to obtain the Target Segmentation image with preferable segmentation precision.Likewise, the 6th dense piece 607 to the 5th
When the characteristic pattern of dense piece 606 output is handled, it can take and the 4th dense piece 605 same operation.
S602 obtains the target mask image according to the product of the Target Segmentation image and the detection image.
Specifically, computer equipment can obtain mesh according to the corresponding formula of product of Target Segmentation image and detection image
Mask images are marked, target mask figure can also be obtained according to other formula of the product comprising Target Segmentation image and detection image
Picture.Due to black substance region and/or rubrum region in above-mentioned Target Segmentation image pixel value can with 1, Target Segmentation image
The pixel value in the region in addition to black substance region and/or rubrum region can be 0, and therefore, computer equipment can be according to target
In the mask images that the product of segmented image and detection image obtains, region in addition to black substance region and/or rubrum region
Pixel value is also 0, and the pixel value in black substance region and/or rubrum region is not 0, in this way, computer equipment can be only to including
The area-of-interest in black substance region and/or rubrum region carries out feature extraction, to preferably carry out feature extraction, and then improves
The accuracy rate of detection, this is also the embodiment of the present application using target mask image rather than Target Segmentation image is as detection nerve net
The main reason for test object of network model.
Medical image processing method provided in this embodiment, computer equipment can will test image and be input to segmentation nerve
Network model obtains Target Segmentation image, and according to the product of Target Segmentation image and detection image, obtains target mask figure
Picture, since computer equipment can carry out feature extraction to detection image using segmentation neural network model, to obtain target mask
Image, the target mask image with higher segmentation precision available in this way, so that utilizing the segmentation nerve net
When the target mask image that network model obtains carries out image detection, the accuracy rate of detection is improved.
It is following by a simply example, to introduce the process of medical image reason method provided in an embodiment of the present invention.
Specifically it may refer to shown in Fig. 7:
S701, computer equipment obtain detection image, and the detection image includes the figure in black substance region and/or rubrum region
Picture.
The detection image is input to the segmentation neural network model, obtains Target Segmentation by S702, computer equipment
Image.
S703, computer equipment obtain the target according to the product of the Target Segmentation image and the detection image
Mask images.
S704, feature extraction group of the computer equipment based on initial detecting neural network model, obtains the trained mask
The corresponding detection characteristic pattern of image.
S705, sorting group and the detection characteristic pattern of the computer equipment based on initial detecting neural network model, obtains
The corresponding initial labels of the trained mask images.
S706, computer equipment determine whether the initial labels and the error of the true tag are less than preset threshold；
And it is less than the judgement of preset threshold in response to the error, the initial detecting neural network model is appointed as the detection
Neural network model.
S707, computer equipment detect the target mask image using the detection neural network model, obtain
To testing result.
The working principle and technical effect of medical image processing method provided in this embodiment as described in above-described embodiment,
This is repeated no more.
It should be understood that although each step in flow chart shown in Fig. 2-7 is successively shown according to the instruction of arrow,
But these steps are not that the inevitable sequence according to arrow instruction successively executes.Unless expressly state otherwise herein, these
There is no stringent sequences to limit for the execution of step, these steps can execute in other order.Moreover, in Fig. 2-7 extremely
Few a part of step may include that perhaps these sub-steps of multiple stages or stage are not necessarily same to multiple sub-steps
Moment executes completion, but can execute at different times, and the execution sequence in these sub-steps or stage is also not necessarily
It successively carries out, but in turn or can be handed over at least part of the sub-step or stage of other steps or other steps
Alternately execute.
Fig. 8 is the structural schematic diagram for the medical image processing devices that one embodiment provides.As shown in figure 8, described device
It include: the first acquisition module 10, the first determining module 11 and the second determining module 12.
Specifically, first obtains module 10, for obtaining detection image, the detection image include black substance region and/or
The image in rubrum region.
First determining module 11, for obtaining segmentation neural network model, using the segmentation neural network model to institute
It states detection image and is split operation, obtain target mask image.
Second determining module 12, for obtaining detection neural network model, using the detection neural network model to institute
It states target mask image to be detected, obtains testing result.
Medical image processing devices provided in this embodiment can execute above method embodiment, realization principle and skill
Art effect is similar, and details are not described herein.
Fig. 9 is the image processing apparatus that another embodiment provides.It is optional on the basis of above-mentioned embodiment illustrated in fig. 8
, above-mentioned apparatus further includes the second acquisition module 13 and third determining module 14,
Specifically, second obtains module 13, for obtaining training dataset, the training dataset includes multipair trained number
According to every a pair of training data includes a trained mask images and the corresponding true tag of the trained mask images.
Third determining module 14, for being carried out according to the training dataset to preset initial detecting neural network model
Training, obtains the detection neural network model.
Medical image processing devices provided in this embodiment can execute above method embodiment, realization principle and skill
Art effect is similar, and details are not described herein.
Figure 10 is the medical image processing devices that another embodiment provides.On the basis of the above embodiments, optionally,
Above-mentioned third determining module 14 includes detection unit 141 and adjustment unit 142.
Specifically, detection unit 141, for the trained mask images to be input to the initial detecting neural network mould
Type obtains initial labels.
Adjustment unit 142, for determining whether the initial labels and the error of the true tag are less than preset threshold；
And it is less than the judgement of preset threshold in response to the error, the initial detecting neural network model is appointed as the detection
Neural network model.
Medical image processing devices provided in this embodiment can execute above method embodiment, realization principle and skill
Art effect is similar, and details are not described herein.
Figure 11 is the medical image processing devices that another embodiment provides.On the basis of the above embodiments, described first
Begin to detect neural network model to include feature extraction group and sorting group；The feature extraction group include the first convolutional layer, described point
Class group includes the second convolutional layer, pond layer and full articulamentum；Optionally, above-mentioned detection unit 141 includes the first determining subelement
1411 and second determine subelement 1412.
Specifically, first determines subelement 1411, for being based on the feature extraction group, the trained mask images are obtained
Corresponding detection characteristic pattern.
Second determination unit 1412 obtains the trained mask for being based on the sorting group and the detection characteristic pattern
The corresponding initial labels of image.
Medical image processing devices provided in this embodiment can execute above method embodiment, realization principle and skill
Art effect is similar, and details are not described herein.
Figure 12 is the medical image processing devices that another embodiment provides.On the basis of the above embodiments, above-mentioned
One determining module 11 includes the first determination unit 111 and the second determination unit 112.
First determination unit 111 obtains target for the detection image to be input to the segmentation neural network model
Segmented image；
Second determination unit 112 obtains described for the product according to the Target Segmentation image and the detection image
Target mask image.
Medical image processing devices provided in this embodiment can execute above method embodiment, realization principle and skill
Art effect is similar, and details are not described herein.
Optionally, the segmentation neural network model is V-Net model, in U-Net model, FCN model, GAN model
Any one and the dense piece of neural network model combined, the multiple operational groups of the dense piece of correspondence, each operational group include
According to the process of convolution of sequencing progress, activate the process of convolution handled and again.
Optionally, the segmentation neural network model includes downsampling unit and up-sampling unit, the downsampling unit
Including multiple dense pieces for feature extraction, the up-sampling unit includes multiple dense piece for feature reconstruction, described
Skip floor connection is carried out for dense piece of feature extraction and between dense piece described for feature reconstruction.
A kind of magic magiscan is provided in one embodiment, as shown in Figure 1, the system includes image scanning
The schematic diagram of internal structure of instrument 102 and computer equipment 104, computer equipment 104 can be as shown in figure 13.As shown in figure 13,
The computer equipment includes processor, memory, network interface, display screen and the input unit connected by system bus.Its
In, the processor of the computer equipment is for providing calculating and control ability.The memory of the computer equipment includes non-volatile
Property storage medium, built-in storage.The non-volatile memory medium is stored with operating system and computer program.The built-in storage is
The operation of operating system and computer program in non-volatile memory medium provides environment.The network interface of the computer equipment
For being communicated with external terminal by network connection.A kind of medicine figure may be implemented in the computer program when being executed by processor
As processing method.The display screen of the computer equipment can be liquid crystal display or electric ink display screen, which sets
Standby input unit can be the touch layer covered on display screen, be also possible to the key being arranged on computer equipment shell, rail
Mark ball or Trackpad can also be external keyboard, Trackpad or mouse etc..It is real when above-mentioned computer program is executed by processor
Existing following steps:
Detection image is obtained, the detection image includes the image in black substance region and/or rubrum region；
Segmentation neural network model is obtained, behaviour is split to the detection image using the segmentation neural network model
Make, obtains target mask image；
Detection neural network model is obtained, the target mask image is examined using the detection neural network model
It surveys, obtains testing result.
In one embodiment, a kind of computer readable storage medium is provided, computer program is stored thereon with, is calculated
Machine program performs the steps of when being executed by processor
Detection image is obtained, the detection image includes the image in black substance region and/or rubrum region；
Segmentation neural network model is obtained, behaviour is split to the detection image using the segmentation neural network model
Make, obtains target mask image；
Detection neural network model is obtained, the target mask image is examined using the detection neural network model
It surveys, obtains testing result.
Those of ordinary skill in the art will appreciate that realizing all or part of the process in above-described embodiment method, being can be with
Relevant hardware is instructed to complete by computer program, the computer program can be stored in a non-volatile computer
In read/write memory medium, the computer program is when being executed, it may include such as the process of the embodiment of above-mentioned each method.Wherein,
To any reference of memory, storage, database or other media used in each embodiment provided herein,
Including non-volatile and/or volatile memory.Nonvolatile memory may include read-only memory (ROM), programming ROM
(PROM), electrically programmable ROM (EPROM), electrically erasable ROM (EEPROM) or flash memory.Volatile memory may include
Random access memory (RAM) or external cache.By way of illustration and not limitation, RAM is available in many forms,
Such as static state RAM (SRAM), dynamic ram (DRAM), synchronous dram (SDRAM), double data rate sdram (DDRSDRAM), enhancing
Type SDRAM (ESDRAM), synchronization link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM
(RDRAM), direct memory bus dynamic ram (DRDRAM) and memory bus dynamic ram (RDRAM) etc..
Each technical characteristic of embodiment described above can be combined arbitrarily, for simplicity of description, not to above-mentioned reality
It applies all possible combination of each technical characteristic in example to be all described, as long as however, the combination of these technical characteristics is not deposited
In contradiction, all should be considered as described in this specification.
The embodiments described above only express several embodiments of the present invention, and the description thereof is more specific and detailed, but simultaneously
It cannot therefore be construed as limiting the scope of the patent.It should be pointed out that coming for those of ordinary skill in the art
It says, without departing from the inventive concept of the premise, various modifications and improvements can be made, these belong to protection of the invention
Range.Therefore, the scope of protection of the patent of the invention shall be subject to the appended claims.
Claims (10)
1. a kind of medical image processing method, which is characterized in that the described method includes:
Detection image is obtained, the detection image includes the image in black substance region and/or rubrum region；
Segmentation neural network model is obtained, operation is split to the detection image using the segmentation neural network model,
Obtain target mask image；
Detection neural network model is obtained, the target mask image is detected using the detection neural network model,
Obtain testing result.
2. the method according to claim 1, wherein the acquisition detect neural network model before, it is described
Method further include:
Training dataset is obtained, the training dataset includes multipair training data, and every a pair of training data includes a training
Mask images and the corresponding true tag of the trained mask images；
Preset initial detecting neural network model is trained according to the training dataset, obtains the detection nerve net
Network model.
3. according to the method described in claim 2, it is characterized in that, it is described according to the training dataset to preset initial inspection
It surveys neural network model to be trained, obtains the detection neural network model, comprising:
The trained mask images are input to the initial detecting neural network model, obtain initial labels；
Determine whether the initial labels and the error of the true tag are less than preset threshold；And it is small in response to the error
In the judgement of preset threshold, the initial detecting neural network model is appointed as the detection neural network model.
4. according to the method described in claim 3, it is characterized in that, the initial detecting neural network model includes feature extraction
Group and sorting group；
The feature extraction group includes the first convolutional layer, and the sorting group includes the second convolutional layer, pond layer and full articulamentum；
It is described that the trained mask images are input to the initial detecting neural network model, obtain initial labels, comprising:
Based on the feature extraction group, the corresponding detection characteristic pattern of the trained mask images is obtained；
Based on the sorting group and the detection characteristic pattern, the corresponding initial labels of the trained mask images are obtained.
5. the method according to claim 1, wherein described use preset segmentation neural network model to described
Detection image is split operation, obtains target mask image, comprising:
The detection image is input to the segmentation neural network model, obtains Target Segmentation image；
According to the product of the Target Segmentation image and the detection image, the target mask image is obtained.
6. according to the method described in claim 5, it is characterized in that, the segmentation neural network model is V-Net model, U-
Net model, FCN model, any one and the dense piece of neural network model combined in GAN model, described dense piece right
Multiple operational groups are answered, each operational group includes the process of convolution carried out according to sequencing, activation handles and volume again
Product processing.
7. according to the method described in claim 6, it is characterized in that, the segmentation neural network model include downsampling unit and
Up-sampling unit, the downsampling unit include multiple dense piece for feature extraction, and the up-sampling unit includes multiple
For dense piece of feature reconstruction, it is described for dense piece of feature extraction and it is dense piece described for feature reconstruction between into
The connection of row skip floor.
8. a kind of medical image processing devices, which is characterized in that described device includes:
First obtains module, and for obtaining detection image, the detection image includes the figure in black substance region and/or rubrum region
Picture；
First determining module, for obtaining segmentation neural network model, using the segmentation neural network model to the detection
Image is split operation, obtains target mask image；
Second determining module, for obtaining detection neural network model, using the detection neural network model to the target
Mask images are detected, and testing result is obtained.
9. a kind of magic magiscan, including image scanner and computer equipment, wherein the computer equipment includes
Memory and processor, the memory are stored with computer program, which is characterized in that the processor executes the computer
Following steps are realized when program:
Detection image is obtained, the detection image includes the image in black substance region and/or rubrum region；
Segmentation neural network model is obtained, operation is split to the detection image using the segmentation neural network model,
Obtain target mask image；
Detection neural network model is obtained, the target mask image is detected using the detection neural network model,
Obtain testing result.
10. a kind of computer readable storage medium, is stored thereon with computer program, which is characterized in that the computer program
Following steps are realized when being executed by processor:
Detection image is obtained, the detection image includes the image in black substance region and/or rubrum region；
Segmentation neural network model is obtained, operation is split to the detection image using the segmentation neural network model,
Obtain target mask image；
Detection neural network model is obtained, the target mask image is detected using the detection neural network model,
Obtain testing result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811126424.7A CN109242849A (en) | 2018-09-26 | 2018-09-26 | Medical image processing method, device, system and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811126424.7A CN109242849A (en) | 2018-09-26 | 2018-09-26 | Medical image processing method, device, system and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109242849A true CN109242849A (en) | 2019-01-18 |
Family
ID=65056185
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811126424.7A Pending CN109242849A (en) | 2018-09-26 | 2018-09-26 | Medical image processing method, device, system and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109242849A (en) |
Cited By (26)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109859217A (en) * | 2019-02-20 | 2019-06-07 | 厦门美图之家科技有限公司 | The dividing method in pore region and calculating equipment in facial image |
| CN110033019A (en) * | 2019-03-06 | 2019-07-19 | 腾讯科技（深圳）有限公司 | Method for detecting abnormality, device and the storage medium of human body |
| CN110060251A (en) * | 2019-04-26 | 2019-07-26 | 福州大学 | A kind of building surface crack detecting method based on U-Net |
| CN110084745A (en) * | 2019-03-12 | 2019-08-02 | 天津大学 | Image super-resolution rebuilding method based on dense convolutional neural networks in parallel |
| CN110188813A (en) * | 2019-05-24 | 2019-08-30 | 上海联影智能医疗科技有限公司 | Characteristics of image classification method, computer equipment and storage medium |
| CN110197729A (en) * | 2019-05-20 | 2019-09-03 | 华南理工大学 | Tranquillization state fMRI data classification method and device based on deep learning |
| CN110210544A (en) * | 2019-05-24 | 2019-09-06 | 上海联影智能医疗科技有限公司 | Image classification method, computer equipment and storage medium |
| CN110309855A (en) * | 2019-05-30 | 2019-10-08 | 上海联影智能医疗科技有限公司 | Training method, computer equipment and the storage medium of image segmentation |
| CN110310257A (en) * | 2019-05-31 | 2019-10-08 | 上海联影医疗科技有限公司 | Medical image processing method, device, computer equipment and storage medium |
| CN110322432A (en) * | 2019-05-27 | 2019-10-11 | 上海联影医疗科技有限公司 | Medical image processing method, device, computer equipment and readable storage medium storing program for executing |
| CN110443254A (en) * | 2019-08-02 | 2019-11-12 | 上海联影医疗科技有限公司 | The detection method of metallic region, device, equipment and storage medium in image |
| CN110675385A (en) * | 2019-09-25 | 2020-01-10 | 腾讯科技（深圳）有限公司 | Image processing method and device, computer equipment and storage medium |
| CN110796636A (en) * | 2019-09-25 | 2020-02-14 | 中国人民解放军战略支援部队信息工程大学 | CT image bone condition detection method and device based on convolutional neural network |
| CN111178428A (en) * | 2019-12-27 | 2020-05-19 | 上海联影智能医疗科技有限公司 | Cartilage damage classification method and device, computer equipment and storage medium |
| CN111325729A (en) * | 2020-02-19 | 2020-06-23 | 青岛海信医疗设备股份有限公司 | Biological tissue segmentation method based on biomedical images and communication terminal |
| CN111681205A (en) * | 2020-05-08 | 2020-09-18 | 上海联影智能医疗科技有限公司 | Image analysis method, computer device, and storage medium |
| CN111951270A (en) * | 2020-06-29 | 2020-11-17 | 珠海格力电器股份有限公司 | Single-target object segmentation method and detection device based on deep learning |
| CN112184635A (en) * | 2020-09-10 | 2021-01-05 | 上海商汤智能科技有限公司 | Target detection method, device, storage medium and equipment |
| WO2021017372A1 (en) * | 2019-08-01 | 2021-02-04 | 中国科学院深圳先进技术研究院 | Medical image segmentation method and system based on generative adversarial network, and electronic equipment |
| CN112505598A (en) * | 2020-12-10 | 2021-03-16 | 上海交通大学 | Quantitative magnetic susceptibility imaging reconstruction method and system, storage medium and terminal |
| US11013452B1 (en) | 2020-03-03 | 2021-05-25 | Heuron Co., Ltd. | Artificial intelligence (AI) based parkinson's disease diagnosing apparatus and method |
| CN113424223A (en) * | 2019-02-21 | 2021-09-21 | 索尼集团公司 | Multiple neural network based object segmentation in a sequence of color image frames |
| CN114273240A (en) * | 2020-09-27 | 2022-04-05 | 深圳顺丰泰森控股(集团)有限公司 | Express delivery single piece separation method, device and system and storage medium |
| WO2022121531A1 (en) * | 2020-12-09 | 2022-06-16 | 歌尔股份有限公司 | Product defect detection method and apparatus |
| CN116389647A (en) * | 2023-06-02 | 2023-07-04 | 深圳市尚哲医健科技有限责任公司 | Emergency first-aid integrated platform |
| CN110675385B (en) * | 2019-09-25 | 2024-05-14 | 腾讯科技（深圳）有限公司 | Image processing method, device, computer equipment and storage medium |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105938617A (en) * | 2015-03-06 | 2016-09-14 | 法玛科技顾问股份有限公司 | System and method for quantitatively analyzing nuclear medicine brain image |
| CN106898000A (en) * | 2017-02-15 | 2017-06-27 | 华东师范大学 | A kind of automatic division method of magnetic resonance imaging cerebral gray matter core group |
| CN106934816A (en) * | 2017-03-23 | 2017-07-07 | 中南大学 | A kind of eye fundus image Segmentation Method of Retinal Blood Vessels based on ELM |
| CN107220965A (en) * | 2017-05-05 | 2017-09-29 | 上海联影医疗科技有限公司 | A kind of image partition method and system |
| CN107437096A (en) * | 2017-07-28 | 2017-12-05 | 北京大学 | Image classification method based on the efficient depth residual error network model of parameter |
| CN107909588A (en) * | 2017-07-26 | 2018-04-13 | 广州慧扬健康科技有限公司 | Partition system under MRI cortex based on three-dimensional full convolutional neural networks |
| CN108122236A (en) * | 2017-12-18 | 2018-06-05 | 上海交通大学 | Iterative eye fundus image blood vessel segmentation method based on distance modulated loss |
| CN108346145A (en) * | 2018-01-31 | 2018-07-31 | 浙江大学 | The recognition methods of unconventional cell in a kind of pathological section |
| CN108427963A (en) * | 2018-03-02 | 2018-08-21 | 浙江工业大学 | A kind of dermopathic classifying identification method of melanoma based on deep learning |
-
2018
- 2018-09-26 CN CN201811126424.7A patent/CN109242849A/en active Pending
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105938617A (en) * | 2015-03-06 | 2016-09-14 | 法玛科技顾问股份有限公司 | System and method for quantitatively analyzing nuclear medicine brain image |
| CN106898000A (en) * | 2017-02-15 | 2017-06-27 | 华东师范大学 | A kind of automatic division method of magnetic resonance imaging cerebral gray matter core group |
| CN106934816A (en) * | 2017-03-23 | 2017-07-07 | 中南大学 | A kind of eye fundus image Segmentation Method of Retinal Blood Vessels based on ELM |
| CN107220965A (en) * | 2017-05-05 | 2017-09-29 | 上海联影医疗科技有限公司 | A kind of image partition method and system |
| CN107909588A (en) * | 2017-07-26 | 2018-04-13 | 广州慧扬健康科技有限公司 | Partition system under MRI cortex based on three-dimensional full convolutional neural networks |
| CN107437096A (en) * | 2017-07-28 | 2017-12-05 | 北京大学 | Image classification method based on the efficient depth residual error network model of parameter |
| CN108122236A (en) * | 2017-12-18 | 2018-06-05 | 上海交通大学 | Iterative eye fundus image blood vessel segmentation method based on distance modulated loss |
| CN108346145A (en) * | 2018-01-31 | 2018-07-31 | 浙江大学 | The recognition methods of unconventional cell in a kind of pathological section |
| CN108427963A (en) * | 2018-03-02 | 2018-08-21 | 浙江工业大学 | A kind of dermopathic classifying identification method of melanoma based on deep learning |
Non-Patent Citations (2)
| Title |
|---|
| J.C.TAMRAZ: "《脑磁共振局部解剖与功能图谱》", 31 December 2001 * |
| 范文义: "《"3S"理论与技术》", 31 March 2003 * |
Cited By (35)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109859217A (en) * | 2019-02-20 | 2019-06-07 | 厦门美图之家科技有限公司 | The dividing method in pore region and calculating equipment in facial image |
| CN113424223A (en) * | 2019-02-21 | 2021-09-21 | 索尼集团公司 | Multiple neural network based object segmentation in a sequence of color image frames |
| CN113424223B (en) * | 2019-02-21 | 2023-08-08 | 索尼集团公司 | Object segmentation based on multiple neural networks in color image frame sequences |
| CN110033019A (en) * | 2019-03-06 | 2019-07-19 | 腾讯科技（深圳）有限公司 | Method for detecting abnormality, device and the storage medium of human body |
| CN110033019B (en) * | 2019-03-06 | 2021-07-27 | 腾讯科技（深圳）有限公司 | Method and device for detecting abnormality of human body part and storage medium |
| CN110084745A (en) * | 2019-03-12 | 2019-08-02 | 天津大学 | Image super-resolution rebuilding method based on dense convolutional neural networks in parallel |
| CN110060251A (en) * | 2019-04-26 | 2019-07-26 | 福州大学 | A kind of building surface crack detecting method based on U-Net |
| CN110197729A (en) * | 2019-05-20 | 2019-09-03 | 华南理工大学 | Tranquillization state fMRI data classification method and device based on deep learning |
| CN110210544A (en) * | 2019-05-24 | 2019-09-06 | 上海联影智能医疗科技有限公司 | Image classification method, computer equipment and storage medium |
| CN110210544B (en) * | 2019-05-24 | 2021-11-23 | 上海联影智能医疗科技有限公司 | Image classification method, computer device, and storage medium |
| CN110188813A (en) * | 2019-05-24 | 2019-08-30 | 上海联影智能医疗科技有限公司 | Characteristics of image classification method, computer equipment and storage medium |
| CN110322432A (en) * | 2019-05-27 | 2019-10-11 | 上海联影医疗科技有限公司 | Medical image processing method, device, computer equipment and readable storage medium storing program for executing |
| CN110309855A (en) * | 2019-05-30 | 2019-10-08 | 上海联影智能医疗科技有限公司 | Training method, computer equipment and the storage medium of image segmentation |
| CN110310257A (en) * | 2019-05-31 | 2019-10-08 | 上海联影医疗科技有限公司 | Medical image processing method, device, computer equipment and storage medium |
| CN110310257B (en) * | 2019-05-31 | 2022-03-01 | 上海联影智能医疗科技有限公司 | Medical image processing method, apparatus, computer device and storage medium |
| WO2021017372A1 (en) * | 2019-08-01 | 2021-02-04 | 中国科学院深圳先进技术研究院 | Medical image segmentation method and system based on generative adversarial network, and electronic equipment |
| CN110443254A (en) * | 2019-08-02 | 2019-11-12 | 上海联影医疗科技有限公司 | The detection method of metallic region, device, equipment and storage medium in image |
| CN110443254B (en) * | 2019-08-02 | 2023-06-20 | 上海联影医疗科技股份有限公司 | Method, device, equipment and storage medium for detecting metal area in image |
| CN110796636A (en) * | 2019-09-25 | 2020-02-14 | 中国人民解放军战略支援部队信息工程大学 | CT image bone condition detection method and device based on convolutional neural network |
| WO2021057536A1 (en) * | 2019-09-25 | 2021-04-01 | 腾讯科技（深圳）有限公司 | Image processing method and apparatus, computer device and storage medium |
| CN110675385A (en) * | 2019-09-25 | 2020-01-10 | 腾讯科技（深圳）有限公司 | Image processing method and device, computer equipment and storage medium |
| CN110675385B (en) * | 2019-09-25 | 2024-05-14 | 腾讯科技（深圳）有限公司 | Image processing method, device, computer equipment and storage medium |
| CN111178428B (en) * | 2019-12-27 | 2024-03-22 | 上海联影智能医疗科技有限公司 | Cartilage damage classification method, cartilage damage classification device, computer equipment and storage medium |
| CN111178428A (en) * | 2019-12-27 | 2020-05-19 | 上海联影智能医疗科技有限公司 | Cartilage damage classification method and device, computer equipment and storage medium |
| CN111325729A (en) * | 2020-02-19 | 2020-06-23 | 青岛海信医疗设备股份有限公司 | Biological tissue segmentation method based on biomedical images and communication terminal |
| US11013452B1 (en) | 2020-03-03 | 2021-05-25 | Heuron Co., Ltd. | Artificial intelligence (AI) based parkinson's disease diagnosing apparatus and method |
| EP3876194A1 (en) * | 2020-03-03 | 2021-09-08 | Heuron Co., Ltd. | Artificial intelligence (ai) based parkinson's disease diagnosing apparatus and method |
| CN111681205A (en) * | 2020-05-08 | 2020-09-18 | 上海联影智能医疗科技有限公司 | Image analysis method, computer device, and storage medium |
| CN111951270A (en) * | 2020-06-29 | 2020-11-17 | 珠海格力电器股份有限公司 | Single-target object segmentation method and detection device based on deep learning |
| CN112184635A (en) * | 2020-09-10 | 2021-01-05 | 上海商汤智能科技有限公司 | Target detection method, device, storage medium and equipment |
| CN114273240A (en) * | 2020-09-27 | 2022-04-05 | 深圳顺丰泰森控股(集团)有限公司 | Express delivery single piece separation method, device and system and storage medium |
| WO2022121531A1 (en) * | 2020-12-09 | 2022-06-16 | 歌尔股份有限公司 | Product defect detection method and apparatus |
| CN112505598A (en) * | 2020-12-10 | 2021-03-16 | 上海交通大学 | Quantitative magnetic susceptibility imaging reconstruction method and system, storage medium and terminal |
| CN116389647A (en) * | 2023-06-02 | 2023-07-04 | 深圳市尚哲医健科技有限责任公司 | Emergency first-aid integrated platform |
| CN116389647B (en) * | 2023-06-02 | 2023-08-08 | 深圳市尚哲医健科技有限责任公司 | Emergency first-aid integrated platform |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109242849A (en) | Medical image processing method, device, system and storage medium | |
| US10854339B2 (en) | Systems and methods for associating medical images with a patient | |
| US20160073969A1 (en) | Medical Imaging System Providing Disease Prognosis | |
| US20210151187A1 (en) | Data-Driven Estimation of Predictive Digital Twin Models from Medical Data | |
| CN108896943A (en) | A kind of magnetic resonance quantitative imaging method and device | |
| CN105138813B (en) | Patient data is mapped as another system and method from a kind of physiological status | |
| CN110033858A (en) | A kind of tongue picture analysis method and its storage medium | |
| JP7239701B2 (en) | Organ volume measurement method and apparatus using artificial neural network | |
| Jannin et al. | Validation in medical image processing. | |
| WO2021143780A1 (en) | Multi-label learning-based post-physical examination chronic disease prognosis system | |
| CN109192305A (en) | A kind of heart function automatic analysis method based on deep-cycle neural network | |
| Do et al. | Accuracy, uncertainty, and adaptability of automatic myocardial ASL segmentation using deep CNN | |
| CN110473171A (en) | Brain age detection method, computer equipment and storage medium | |
| TW201941220A (en) | Diagnosis support device, program, learned model, and learning device | |
| Wen et al. | On the deep learning-based age prediction of color fundus images and correlation with ophthalmic diseases | |
| CN114494132A (en) | Disease classification system based on deep learning and fiber bundle spatial statistical analysis | |
| CN109480842B (en) | System and apparatus for diagnosing functional dyspepsia | |
| Shetty et al. | Alzheimer’s Diseases Detection by using Convolution Neural Network | |
| Kalapos et al. | Automated T1 and T2 mapping segmentation on cardiovascular magnetic resonance imaging using deep learning | |
| Donepudi et al. | Brain Metastasis Tumor Detection using Image Segmentation and VGG16 Architecture | |
| CN110599560A (en) | Magnetic resonance imaging method, magnetic resonance imaging device, storage medium and computer equipment | |
| Osman et al. | Collaborative Learning for Annotation‐Efficient Volumetric MR Image Segmentation | |
| KR102647251B1 (en) | Method for evaluating low limb alignment and device for evaluating low limb alignment using the same | |
| CN113517057B (en) | Medical image information identification and storage method based on data representation and neural network | |
| CN117438103A (en) | Method and system for predicting duloxetine analgesic efficacy based on MRI (magnetic resonance imaging) image |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190118 |
|
| RJ01 | Rejection of invention patent application after publication |