CN106780482A - A kind of classification method of medical image - Google Patents
A kind of classification method of medical image Download PDFInfo
- Publication number
- CN106780482A CN106780482A CN201710011904.8A CN201710011904A CN106780482A CN 106780482 A CN106780482 A CN 106780482A CN 201710011904 A CN201710011904 A CN 201710011904A CN 106780482 A CN106780482 A CN 106780482A
- Authority
- CN
- China
- Prior art keywords
- image
- layer
- output
- medical image
- input
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 31
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 17
- 238000000605 extraction Methods 0.000 claims abstract description 12
- 238000012549 training Methods 0.000 claims description 30
- 238000010606 normalization Methods 0.000 claims description 15
- 239000011248 coating agent Substances 0.000 claims description 10
- 238000000576 coating method Methods 0.000 claims description 10
- 238000013461 design Methods 0.000 claims description 10
- 239000013598 vector Substances 0.000 claims description 10
- 239000011159 matrix material Substances 0.000 claims description 7
- 230000004913 activation Effects 0.000 claims description 6
- 238000013136 deep learning model Methods 0.000 claims description 5
- 241001269238 Data Species 0.000 claims description 2
- 238000004422 calculation algorithm Methods 0.000 claims description 2
- 238000004364 calculation method Methods 0.000 claims description 2
- 238000010586 diagram Methods 0.000 claims description 2
- 238000005070 sampling Methods 0.000 claims description 2
- 230000007935 neutral effect Effects 0.000 claims 1
- 230000000694 effects Effects 0.000 abstract description 5
- 238000013135 deep learning Methods 0.000 abstract description 2
- 238000013526 transfer learning Methods 0.000 abstract description 2
- 201000010099 disease Diseases 0.000 description 7
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 7
- 238000003745 diagnosis Methods 0.000 description 4
- 230000001575 pathological effect Effects 0.000 description 4
- 238000013145 classification model Methods 0.000 description 2
- AZFKQCNGMSSWDS-UHFFFAOYSA-N MCPA-thioethyl Chemical compound CCSC(=O)COC1=CC=C(Cl)C=C1C AZFKQCNGMSSWDS-UHFFFAOYSA-N 0.000 description 1
- 238000001574 biopsy Methods 0.000 description 1
- 238000004195 computer-aided diagnosis Methods 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 239000006185 dispersion Substances 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 238000010191 image analysis Methods 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 230000003211 malignant effect Effects 0.000 description 1
- 230000007170 pathology Effects 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 238000004611 spectroscopical analysis Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/60—Rotation of whole images or parts thereof
-
- 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
- G06T2207/10024—Color image
-
- 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
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Biology (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Radiology & Medical Imaging (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The present invention proposes a kind of classification method of medical image based on characteristics of image transfer learning and deep learning, good model is showed in image understanding field by what is trained, feature extraction is carried out to medical image, then the characteristic pattern for obtaining is used to train a convolutional neural networks model that can provide last classification, the present invention can make full use of the existing feature that good model extraction is showed in image understanding field, and preferable effect can be obtained on small-scale medical images data sets.
Description
Technical field
Technical field belonging to the present invention is medical image analysis field, specially a kind of classification method of medical image.
Background technology
Medical image plays the role of important in the diagnosis of many medical courses in general, and such as histopathology image is ill for determining
Organize the type and the order of severity of illnesses to be of great significance, many malignant diseases it is final make a definite diagnosis be required for by
Histopathology image and biopsy.But because the identification of medical image needs substantial amounts of professional knowledge and clinical experience, and
Influenceed larger by the subjective factor of interpreting blueprints person, easily caused the unstable of diagnostic result or even reversion, therefore use engineering
The method of habit, from the experience that the medical image diagnosis data learning of history is interpreted blueprints, and is converted into the program of computer, to interpreting blueprints
The stability and correctness of result are all significant.Additionally, the computer-aided diagnosis of medical image helps to mitigate doctor
Workload, enable them to focus more on the diagnosis of special case.
Existing medical image computer automatic identification method is mainly using machine learning model and Statistical learning model
The relation between the feature representation and the disease represented by it of medical image is extracted, for the medical image of certain disease, design
Or specific feature extracting method is selected, and the characteristic vector or matrix of image are obtained, then design or select suitable grader
Or device is returned, classification of diseases is carried out to medical image or the order of severity is given a mark.Document " Ozdemir, E.＆Demir, C.G.A
Hybrid Classification Model for Digital Pathology Using Structural and
Statistical Pattern Recognition.IEEE Trans.Med.Imaging, 2013,32,474-483 " is proposed
Pathological image, is converted into one by one mixed model based on structure and statistical-simulation spectrometry for being used for pathological image classification
Attributed graph and a series of inquiry refer to subgraph, two category features that can be used in classification medical image are then extracted accordingly, finally
Training objective grader.The method follows the conventional thought of feature extraction and classifier training, can only be directed to specific disease
It is trained, and needs larger sample set.
The weak point of existing method is：
(1) needs design to be adapted to the feature extraction mode of particular type medical image.Different features are for different type
The effect of medical image differ widely, but designing suitable feature needs domain expert to take a lot of time, while needing big
The field experience of amount.
(2) is in order to obtain the preferable model of effect, it is necessary to larger historical diagnostic data collection training pattern.Scale
Small diagnostic data set is prevented from using complicated model when Medical Images Classification model is set up, because that can be to cause
Degree fitting, and simple model it is beyond expression of words go out relation between medical image and target disease/pathological characteristics；But the opposing party
Face, for specific medical courses in general, or for the medical image of certain disease for, to obtain sufficiently large training dataset be difficult
, because the acquisition of medically each sample needs practical operation, thus it is very high to obtain the cost of large data sets.
Based on this, the present invention proposes a kind of Medical Images Classification side based on characteristics of image transfer learning and deep learning
Method, good model is showed by what is trained in image understanding field, carries out feature extraction to medical image, then
The characteristic pattern for obtaining is used to train a convolutional neural networks model that can provide last classification, the present invention can be fully sharp
The feature that good model is extracted is showed in image understanding field with existing, can on small-scale medical images data sets
Obtain preferable effect.
The content of the invention
It is contemplated that overcome the shortcomings of existing classification method of medical image, including existing method needs design to be adapted to spy
Determining the feature extraction mode of type medical image, existing method needs fairly large medical images data sets etc..It is of the invention
Feature is included the expansion of the normalized, medical images data sets of original medical image, is led in image understanding using existing
Domain shows good model and carries out preliminary feature extraction, the training of convolutional neural networks and the classification of medical image, each process
Comprising several steps, its feature is described as follows respectively：
(1) normalized of original medical images
The purpose of normalized is the real number for making the numerical value of the matrix of image be converted between 0 to 1, is converted into and training
The weights of process belong to same interval numerical value.Original medical image is the coloured image of RGB channel, if image size is h*w,
Whole image is a matrix of h*w*3, and each passage is a matrix of h*w, and each passage is normalized respectively, is returned
One method changed is first to obtain the pixel maximum and minimum value of each passage in the width image, and p is designated as respectivelymaxAnd pmin,
For each pixel value of the passage, method for normalizing is：Wherein p*It is the pixel value after normalization, p
It is the pixel value before normalization.In addition specify, if pmaxAnd pmin0 is, then the pixel value after normalizing still is 0.
(2) expansion of medical images data sets
The purpose of the expansion of medical images data sets is the scale for increasing data set, and protrusion can determine that final classification is marked
Characteristics of image, reduce model training when overfitting possibility.
Assuming that original image has n.Image level expansion is carried out by following four means：
- upset：To carrying out flip horizontal and flip vertical by the original image after normalization, the image after upset
It is added in data set as new image, then the incrementss of image are 2n in data set；
- rotation：Rotation to carrying out 90,180,270 degree by the original image after normalization, postrotational image
It is added in data set as new image, then the incrementss of image are 3n in data set；
- add error：Random error, error is added to exist each pixel by the image after upset and rotation
It is uniformly distributed between (0,0.005), the image of addition pixel error is added in data set；
- hollow out pixel：Hollowed out to carrying out pixel by the image after upset and rotation, at random the picture of the p% in figure
Plain (p is between 5-15), hollow out operation be selected pixel fill out at random complete (255,255,255) or completely black in vain (0,0,
0), the image for hollowing out pixel is added in data set；By after image extended operation, the scale of image data set is changed into (2n+
3n) * 2+ (2n+3n)=25n.
(3) characteristic pattern of medical images is extracted
The characteristic pattern for carrying out medical image using the existing model good in the performance of image understanding field is extracted.Specifically do
Method is as follows：
- choose one and show good model (having trained) in image understanding field, such as in recent years
ImageNet image recognition contests (ILSVRC, https://image-net.org/challenges/LSVRC/) on it is in the top
Deep learning model, these models have preferably performance in daily image understanding problem, and have comparing bright every year
Aobvious progress；
- each width medical image in data set is input to selected deep learning model, then model can be according to itself
The weights and activation primitive for training, the image to being input into carry out successively secondary feature extraction, and can form one in each layer is
The characteristic pattern of row, and output layer in model has an output；
- to every piece image, randomly select by spy of the dimension between 28*28-128*128 in all layers after model
Figure is levied, is chosen v times altogether, wherein v is odd number, has problems with to need explanation：
The port number of the characteristic pattern that each layer of ■ may be differed, and the layer for example having is 28*28*512, and that have is 32*32*
256, that have is 64*64*128, and this depends on the design of the model for having trained selected；
■ sets a port number threshold value Tmax, choose port number and be more than or equal to TmaxLayer output characteristic figure, strategy such as
Under：If the characteristic pattern port number of a certain layer is less than Tmax, check whether that other layers are identical with the characteristic pattern dimension of this layer,
It is all to be put together with this layer of dimension identical characteristic pattern, if its quantity is still less than Tmax, then the feature of the dimension is abandoned
Figure, otherwise therefrom randomly selects TmaxIndividual characteristic pattern as image characteristic pattern；If the characteristic pattern port number of a certain layer is more than
Tmax, then directly T is therefrom randomly selectedmaxIndividual characteristic pattern, as the characteristic pattern of image；
So far, the every piece image in training set is expressed as v kind feature charts by a model for having trained
Reach.
(4) design and training of convolutional neural networks
This step trains v convolutional neural networks model, and each model is special using one kind that step (3) is extracted
Levy feature representation of the figure as input picture.The structure of convolutional neural networks is as follows in this step：
Convolutional neural networks are made up of input layer, convolutional layer/pond layer, full articulamentum and output layer.
A. input layer
Input layer receives the feature diagram data that is extracted by step (3), the dimension of its input dimension and characteristic pattern and
Port number after sampling is identical；
B. convolutional layer/pond layer
- convolutional layer：First with the convolutional layer that convolution kernel is 5*5, each layer of port number is identical with input layer.One convolutional layer
One nonlinear active coating of heel, the activation primitive of active coating is as follows：
·Relu：Y=max (x, 0), wherein x are the output of last layer, and y is the output of this active coating；
- pond layer：Using the convolution kernel of 5*5, it is 2 to reduce ratio, runs into aliquant situation using the unilateral benefit 0 of level
Method, i.e. [0 10 0] or [0 00 1]；Pond layer is that the convolutional layer of 5*5 and activation unit are applied in combination with convolution kernel,
Total n of such combination one；
C. full articulamentum：1 full articulamentum, the dimension of the full articulamentum are connected in the output of last pond layer combination
Number is 2 times of last pond layer combination output dimension, and an active coating is closelyed follow after full articulamentum；
D. output layer：Output layer is connected entirely with last active coating, the dimension of output and the categorical measure of medical image
It is equal.
(5) training process
The weights of network are adjusted using the error back propagation learning algorithm of standard, adjustment is according to the defeated of network
Enter is carried out with the difference of output.Specifically, for each input network medical image features figure matrix, network it is defeated
It is a vector equal with classification quantity to go out, and the value of each of vector is the real number value between interval [0,1], represents net
Network is judged as input medical image to belong to the probability of the classification representated by this.Calculating the training error per piece image
When, using the output of first discretization again with the error that 0-1 error calculations are total, i.e., for each output probability, if its value is more than
0.5, then it is discrete to turn to 1, otherwise it is discrete turn to 0, wherein 1 represent belong to such, 0 represent be not belonging to such, if the judgement of model
Consistent with real classification, then error is 0, otherwise error is 1, is then carried out according to error anti-to input weights from output end
To adjustment.
The weights of network are initialized using the random number between [0,1], carry out many wheel training, and all training datas are defeated
Enter in network and complete weighed value adjusting for a wheel, untill the output error of network no longer declines.
(6) Medical Images Classifications
A medical image to be sorted is given, first image normalization is carried out by method described in step of the invention (1),
Then the image to be classified after normalization is input in the good model of the image understanding field performance used with training set,
By the method for step (3), v group characteristic patterns are obtained from model, this v groups characteristic pattern is separately input to step (4) according to dimension
In in v convolutional neural networks training, obtain v output vector (each is real-valued between [0,1]), it is laggard
Row discretization, even its value be more than 0.5, then it is discrete to turn to 1, on the contrary it is discrete turn to 0, then each is voted, if
1 number occupies the majority in v output on a certain position, then judge that the medical image belongs to the category, otherwise judge the image not
Belong to the category.
Specific embodiment
The present invention is tested on medical images data sets disclosed in, achieves preferable effect.It is given below
One embodiment, with disclosed medical images data sets histologyDS (https://www.informed.unal.edu.co/ histologyDS) used as test data set, the data set is the pathological image data set of tissue, there is 2828 images, point
It is 4 classifications, image is jpg forms, and color space is RGB, and size is 720*480 pixels.
(1) data normalization
The step of according to the 5th content of the invention (1), carries out the normalization of image, and normalization is every image each passage point
Do not carry out, find out the maximum pixel and minimum pixel of each passage (R or G or B), be normalized by the method for step (1).
(2) data set expands
(2) o'clock according to the content of the invention expands former medical images data sets：
- upset：The horizontal and vertical upset of image is carried out, the incrementss of view data are 2828*2=5656 after upset；
- rotation：Rotation to carrying out 90,180,270 degree by the original image after normalization, postrotational image
It is added in data set as new image, then the incrementss of image are 2828*3=8484 in data set；
- add error：Random error, error is added to exist each pixel by the image after upset and rotation
It is uniformly distributed between (0,0.005), the image of addition pixel error is added in data set；
- hollow out pixel：Hollowed out to carrying out pixel by the image after upset and rotation, at random the picture of the p% in figure
Plain (p is between 5-15), hollow out operation be selected pixel fill out at random complete (255,255,255) or completely black in vain (0,0,
0), the image for hollowing out pixel is added in data set；
By after image extended operation, the scale of image data set is changed into (2*2828+3*2828) * 2+ (2*2828+3*
2828)=25*2828=70700.
(3) feature extraction of medical images
Choose one and show good deep learning model as Feature Selection Model in image understanding field, in this example
Selection ResNet-152-dag (https://www.vlfeat.org/matconvnet/models/imagenet-resnet- 152-dag.mat) used as Feature Selection Model, the model is accepted challenges in the image understanding challenges of ILSVRC 2012 all
Model in be ranked first (https://www.vlfeat.org/matconvnet/pretrained/), medical image
The every piece image concentrated is input to ResNet-152-dag models, and dimension is 28*28's to 28*28-128*128 wherein
V=5 random selection is carried out in characteristic pattern, the port number of maximum is taken during selection for 1024, final checked the 17th, 25,46,59,
99 layers of characteristic pattern, its dimension/port number is respectively 56*56*256,56*56*256,28*28*512,28*28*512,28*
28*512, because this 5 layers port number is all without departing from largest passages number 1024, so directly choosing all features of these layers
Scheme the characteristic pattern as input medical image.
(3) convolutional neural networks design
Using convolutional neural networks deep learning model, v of the 5 convolutional neural networks models of design to input medical image
=5 characteristic patterns are trained, and model is designed by (5) o'clock of the content of the invention, and table 1 is given towards the 1st, 2 groups of features
The network structure of graph expression, table 2 is given towards the 3rd, 4,5 groups of network structures of feature graph expression.
1. towards the 1st, 2 groups of network structures of feature graph expression of table
2. towards the 3rd, 4,5 groups of network structures of feature graph expression of table
(3) network trainings
Data set, is made Matlab by the network structure in realizing Tables 1 and 2 by configuration file in MatConvNet
Data file .mat forms, the training script cnn_train.m for then being provided using MatConvNet is trained.Training is carried out
30 wheels, the learning rate for using is that preceding 10 0.05,11-20 of wheel take turns 0.005,21-30 wheels 0.0005.The loss function of training is used
zero-one loss.Each model is by after 30 wheel training, system can generate 30 .mat files, and each training in rotation is saved respectively
The parameter of model at the end of white silk, these .mat files are each model taken turns and train, and can be used to divide unknown many example samples
Class.
(4) Medical Images Classifications
Using each model the 30th wheel training at the end of network model as disaggregated model, it is to be sorted for one
Medical image, (1) carries out the normalization of image the step of according to the 5th content of the invention；ResNet-152-dag moulds are input into afterwards
Type, extracts the 17th, 25,46,59,99 layers of characteristic pattern, 5 models that this 5 groups of characteristic pattern inputs are trained, and obtains 5 points
This 5 class vectors are carried out discretization by class vector, and even the value of its a certain position is more than 0.5, then discrete to turn to 1, otherwise from
Dispersion is 0, and then each is voted, if 1 number occupies the majority in 5 outputs on a certain position, judges the medical science
Image belongs to the category, otherwise judges that the image is not belonging to the category.
Claims (7)
1. a kind of classification method of medical image, it is characterised in that methods described specifically includes following steps：Original medical image
Normalized, the expansion of medical images data sets, the characteristic pattern extraction of medical image, the design of convolutional neural networks, convolution
The training of neutral net and the classification of medical image.
2. the method for claim 1, it is characterised in that specific the step of the normalized of the original medical image
Including：
It is the coloured image of RGB channel to obtain original medical image, and image size is h*w, and whole image is a square of h*w*3
Battle array, each passage is a matrix of h*w, and each passage is normalized respectively：Each is first obtained in the width image to lead to
The pixel maximum and minimum value in road, are designated as p respectivelymaxAnd pmin, each pixel value for the passage is normalized：Wherein p*It is the pixel value after normalization, p is the pixel value before normalization.
3. method as claimed in claim 2, it is characterised in that specifically wrap the step of the expansion of the medical images data sets
Include：
- upset：To by normalization after original image carry out flip horizontal and flip vertical, using the image after upset as
New image is added in data set, then the incrementss of image are 2n in data set；
- rotation：To carrying out 90,180,270 degree of rotation by the original image after normalization, using postrotational image as
New image is added in data set, then the incrementss of image are 3n in data set；
- add error：To by upset and rotation after image each pixel add random error, error (0,
0.005) it is uniformly distributed between, the image of addition pixel error is added in data set；
- hollow out pixel：At random being hollowed out by the pixel of the image p% after upset and rotation, the finger that hollows out is selected
Pixel it is random fill out complete white or completely black, the image for hollowing out pixel is added in data set；By after image extended operation, scheming
As the scale of data set is changed into 25n, wherein, n is the number of original image.
4. method as claimed in claim 3, it is characterised in that the step of characteristic pattern of the medical image is extracted specifically is wrapped
Include：
- each width medical image in data set is input to selected deep learning model, then model is good according to self training
Weights and activation primitive, to be input into image carry out successively secondary feature extraction, can form a series of spy in each layer
Levy figure, and output layer in model has an output；
- to every piece image, randomly select by feature of the dimension between 28*28-128*128 in all layers after model
Figure, chooses v times altogether, and wherein v is odd number.
5. method as claimed in claim 4, it is characterised in that specifically include the step of the design of the convolutional neural networks：
This step trains v convolutional neural networks model, each model to make using the characteristic pattern that previous step is extracted
Be the feature representation of input picture, and in this step convolutional neural networks by input layer, convolutional layer/pond layer, full articulamentum
Constituted with output layer：
A. input layer
Input layer receives the feature diagram data extracted by previous step, dimension and the sampling of its input dimension and characteristic pattern
Port number afterwards is identical；
B. convolutional layer/pond layer
- convolutional layer：It is the convolutional layer of 5*5 from convolution kernel, each layer of port number is identical with input layer, after a convolutional layer
With a nonlinear active coating, the activation primitive of active coating is as follows：
·Relu：Y=max (x, 0), wherein x are the output of last layer, and y is the output of this active coating；
- pond layer：Using the convolution kernel of 5*5, it is 2 to reduce ratio；Pond layer is the convolutional layer of 5*5 and activation unit with convolution kernel
It is applied in combination, total n of such combination one；
C. full articulamentum：1 full articulamentum is connected in the output of last pond layer combination, the dimension of the full articulamentum is
2 times of last pond layer combination output dimension, closely follow an active coating after full articulamentum；
D. output layer：Output layer is connected entirely with last active coating, and the dimension of output is equal with the categorical measure of medical image.
6. method as claimed in claim 5, it is characterised in that specifically include the step of the training of the convolutional neural networks：
Using standard error back propagation learning algorithm and according to network input with export difference the weights of network are entered
Row adjustment：For the medical image features figure matrix of each input network, the output of network is one equal with classification quantity
Vector, the value of each of vector is the real number value between interval [0,1], and the vector representation network is input medical science figure
The probability of the classification as representated by being judged as belonging to this；When the training error per piece image is calculated, using first discretization
Output is again with the error that 0-1 error calculations are total, i.e., discrete to turn to 1 if its value is more than 0.5 for each output probability,
Otherwise discrete to turn to 0, wherein 1 representative belongs to such, 0 representative is not belonging to such,
If the judgement of model is consistent with real classification, error is 0, otherwise error is 1, is then carried out from output according to error
Input weights are held reversely to adjust；The weights of network are initialized using the random number between [0,1], carry out many trainings in rotation
Practice, all training datas are input in network and complete weighed value adjusting for a wheel, until the output error of network no longer declines
Untill.
7. method as claimed in claim 6, it is characterised in that specifically include the step of the classification of the medical image：
Give a medical image to be sorted, first by the normalized of the original medical image the step of, the medical science
V group characteristic patterns are obtained after the characteristic pattern extraction step of the step of expansion of image data set and the medical image, the v groups
Characteristic pattern is separately input in the v convolutional neural networks for training according to dimension, obtains carrying out after v output vector discrete
Change, even its value be more than 0.5, then it is discrete to turn to 1, on the contrary it is discrete turn to 0, then each is voted, if a certain
1 number occupies the majority in v output on position, then judge that the medical image belongs to the category, otherwise judge that the image is not belonging to this
Classification.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710011904.8A CN106780482A (en) | 2017-01-08 | 2017-01-08 | A kind of classification method of medical image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710011904.8A CN106780482A (en) | 2017-01-08 | 2017-01-08 | A kind of classification method of medical image |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106780482A true CN106780482A (en) | 2017-05-31 |
Family
ID=58951113
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710011904.8A Pending CN106780482A (en) | 2017-01-08 | 2017-01-08 | A kind of classification method of medical image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106780482A (en) |
Cited By (20)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107169527A (en) * | 2017-06-06 | 2017-09-15 | 西北工业大学 | Classification method of medical image based on collaboration deep learning |
| CN107358157A (en) * | 2017-06-07 | 2017-11-17 | 阿里巴巴集团控股有限公司 | A kind of human face in-vivo detection method, device and electronic equipment |
| CN107492099A (en) * | 2017-08-28 | 2017-12-19 | 京东方科技集团股份有限公司 | Medical image analysis method, medical image analysis system and storage medium |
| CN107527069A (en) * | 2017-08-22 | 2017-12-29 | 京东方科技集团股份有限公司 | Image processing method, device, electronic equipment and computer-readable medium |
| CN107563372A (en) * | 2017-07-20 | 2018-01-09 | 济南中维世纪科技有限公司 | A kind of license plate locating method based on deep learning SSD frameworks |
| CN107679525A (en) * | 2017-11-01 | 2018-02-09 | 腾讯科技（深圳）有限公司 | Image classification method, device and computer-readable recording medium |
| CN107767378A (en) * | 2017-11-13 | 2018-03-06 | 浙江中医药大学 | The multi-modal Magnetic Resonance Image Segmentation methods of GBM based on deep neural network |
| CN107862344A (en) * | 2017-12-01 | 2018-03-30 | 中南大学 | A kind of image classification method |
| CN108109152A (en) * | 2018-01-03 | 2018-06-01 | 深圳北航新兴产业技术研究院 | Medical Images Classification and dividing method and device |
| CN108564570A (en) * | 2018-03-29 | 2018-09-21 | 哈尔滨工业大学（威海） | A kind of method and apparatus of intelligentized pathological tissues positioning |
| CN109753978A (en) * | 2017-11-01 | 2019-05-14 | 腾讯科技（深圳）有限公司 | Image classification method, device and computer readable storage medium |
| CN110096810A (en) * | 2019-05-05 | 2019-08-06 | 中南大学 | A kind of industrial process flexible measurement method based on layer-by-layer data extending deep learning |
| CN110163884A (en) * | 2019-05-17 | 2019-08-23 | 温州大学 | A kind of single image dividing method based on full connection deep learning neural network |
| WO2020102914A1 (en) * | 2018-11-24 | 2020-05-28 | Densitas Incorporated | System and method for assessing medical images |
| CN111374672A (en) * | 2018-12-29 | 2020-07-07 | 西安思博探声生物科技有限公司 | Intelligent knee pad and knee joint injury early warning method |
| CN112041893A (en) * | 2018-04-03 | 2020-12-04 | 大卫·拜伦·道格拉斯 | Radiologist-assisted machine learning with interactive, variable volume 3D cursors |
| CN112819076A (en) * | 2021-02-03 | 2021-05-18 | 中南大学 | Deep migration learning-based medical image classification model training method and device |
| CN113486959A (en) * | 2021-07-07 | 2021-10-08 | 漳州卫生职业学院 | Lung CT image classification method based on feature migration |
| CN114502061A (en) * | 2018-12-04 | 2022-05-13 | 巴黎欧莱雅 | Image-based automatic skin diagnosis using deep learning |
| CN117911794A (en) * | 2024-03-15 | 2024-04-19 | 广州中科智巡科技有限公司 | Model obtaining method and device for image classification, electronic equipment and storage medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104700118A (en) * | 2015-03-18 | 2015-06-10 | 中国科学院自动化研究所 | Pulmonary nodule benignity and malignancy predicting method based on convolutional neural networks |
| CN105631480A (en) * | 2015-12-30 | 2016-06-01 | 哈尔滨工业大学 | Hyperspectral data classification method based on multi-layer convolution network and data organization and folding |
| CN105718952A (en) * | 2016-01-22 | 2016-06-29 | 武汉科恩斯医疗科技有限公司 | Method for focus classification of sectional medical images by employing deep learning network |
| CN106096616A (en) * | 2016-06-08 | 2016-11-09 | 四川大学华西医院 | Magnetic resonance image feature extraction and classification method based on deep learning |
| CN106156793A (en) * | 2016-06-27 | 2016-11-23 | 西北工业大学 | Extract in conjunction with further feature and the classification method of medical image of shallow-layer feature extraction |
-
2017
- 2017-01-08 CN CN201710011904.8A patent/CN106780482A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104700118A (en) * | 2015-03-18 | 2015-06-10 | 中国科学院自动化研究所 | Pulmonary nodule benignity and malignancy predicting method based on convolutional neural networks |
| CN105631480A (en) * | 2015-12-30 | 2016-06-01 | 哈尔滨工业大学 | Hyperspectral data classification method based on multi-layer convolution network and data organization and folding |
| CN105718952A (en) * | 2016-01-22 | 2016-06-29 | 武汉科恩斯医疗科技有限公司 | Method for focus classification of sectional medical images by employing deep learning network |
| CN106096616A (en) * | 2016-06-08 | 2016-11-09 | 四川大学华西医院 | Magnetic resonance image feature extraction and classification method based on deep learning |
| CN106156793A (en) * | 2016-06-27 | 2016-11-23 | 西北工业大学 | Extract in conjunction with further feature and the classification method of medical image of shallow-layer feature extraction |
Non-Patent Citations (2)
| Title |
|---|
| 吴正文: "《卷积神经网络在图像分类中的应用研究》", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
| 楚敏南: "《基于卷积神经网络的图像分类技术研究》", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
Cited By (34)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107169527A (en) * | 2017-06-06 | 2017-09-15 | 西北工业大学 | Classification method of medical image based on collaboration deep learning |
| CN107169527B (en) * | 2017-06-06 | 2020-04-03 | 西北工业大学 | Medical image classification method based on collaborative deep learning |
| CN107358157A (en) * | 2017-06-07 | 2017-11-17 | 阿里巴巴集团控股有限公司 | A kind of human face in-vivo detection method, device and electronic equipment |
| US10671870B2 (en) | 2017-06-07 | 2020-06-02 | Alibaba Group Holding Limited | Determining user authenticity with face liveness detection |
| CN107563372A (en) * | 2017-07-20 | 2018-01-09 | 济南中维世纪科技有限公司 | A kind of license plate locating method based on deep learning SSD frameworks |
| CN107563372B (en) * | 2017-07-20 | 2021-01-29 | 济南中维世纪科技有限公司 | License plate positioning method based on deep learning SSD frame |
| US11636664B2 (en) | 2017-08-22 | 2023-04-25 | Boe Technology Group Co., Ltd. | Training method and device of neural network for medical image processing, and medical image processing method and device |
| CN107527069A (en) * | 2017-08-22 | 2017-12-29 | 京东方科技集团股份有限公司 | Image processing method, device, electronic equipment and computer-readable medium |
| CN107492099A (en) * | 2017-08-28 | 2017-12-19 | 京东方科技集团股份有限公司 | Medical image analysis method, medical image analysis system and storage medium |
| US10706333B2 (en) | 2017-08-28 | 2020-07-07 | Boe Technology Group Co., Ltd. | Medical image analysis method, medical image analysis system and storage medium |
| CN107679525A (en) * | 2017-11-01 | 2018-02-09 | 腾讯科技（深圳）有限公司 | Image classification method, device and computer-readable recording medium |
| CN109753978A (en) * | 2017-11-01 | 2019-05-14 | 腾讯科技（深圳）有限公司 | Image classification method, device and computer readable storage medium |
| CN109753978B (en) * | 2017-11-01 | 2023-02-17 | 腾讯科技（深圳）有限公司 | Image classification method, device and computer readable storage medium |
| CN107679525B (en) * | 2017-11-01 | 2022-11-29 | 腾讯科技（深圳）有限公司 | Image classification method and device and computer readable storage medium |
| CN107767378B (en) * | 2017-11-13 | 2020-08-04 | 浙江中医药大学 | GBM multi-mode magnetic resonance image segmentation method based on deep neural network |
| CN107767378A (en) * | 2017-11-13 | 2018-03-06 | 浙江中医药大学 | The multi-modal Magnetic Resonance Image Segmentation methods of GBM based on deep neural network |
| CN107862344B (en) * | 2017-12-01 | 2021-06-11 | 中南大学 | Image classification method |
| CN107862344A (en) * | 2017-12-01 | 2018-03-30 | 中南大学 | A kind of image classification method |
| CN108109152A (en) * | 2018-01-03 | 2018-06-01 | 深圳北航新兴产业技术研究院 | Medical Images Classification and dividing method and device |
| CN108564570A (en) * | 2018-03-29 | 2018-09-21 | 哈尔滨工业大学（威海） | A kind of method and apparatus of intelligentized pathological tissues positioning |
| CN112041893A (en) * | 2018-04-03 | 2020-12-04 | 大卫·拜伦·道格拉斯 | Radiologist-assisted machine learning with interactive, variable volume 3D cursors |
| WO2020102914A1 (en) * | 2018-11-24 | 2020-05-28 | Densitas Incorporated | System and method for assessing medical images |
| CN114502061A (en) * | 2018-12-04 | 2022-05-13 | 巴黎欧莱雅 | Image-based automatic skin diagnosis using deep learning |
| CN114502061B (en) * | 2018-12-04 | 2024-05-28 | 巴黎欧莱雅 | Image-based automatic skin diagnosis using deep learning |
| CN111374672A (en) * | 2018-12-29 | 2020-07-07 | 西安思博探声生物科技有限公司 | Intelligent knee pad and knee joint injury early warning method |
| CN111374672B (en) * | 2018-12-29 | 2024-01-19 | 西安思博探声生物科技有限公司 | Intelligent knee pad and knee joint injury early warning method |
| CN110096810A (en) * | 2019-05-05 | 2019-08-06 | 中南大学 | A kind of industrial process flexible measurement method based on layer-by-layer data extending deep learning |
| CN110163884A (en) * | 2019-05-17 | 2019-08-23 | 温州大学 | A kind of single image dividing method based on full connection deep learning neural network |
| CN112819076A (en) * | 2021-02-03 | 2021-05-18 | 中南大学 | Deep migration learning-based medical image classification model training method and device |
| CN112819076B (en) * | 2021-02-03 | 2022-06-17 | 中南大学 | Deep migration learning-based medical image classification model training method and device |
| CN113486959A (en) * | 2021-07-07 | 2021-10-08 | 漳州卫生职业学院 | Lung CT image classification method based on feature migration |
| CN113486959B (en) * | 2021-07-07 | 2023-06-16 | 漳州卫生职业学院 | Lung CT image classification method based on feature migration |
| CN117911794A (en) * | 2024-03-15 | 2024-04-19 | 广州中科智巡科技有限公司 | Model obtaining method and device for image classification, electronic equipment and storage medium |
| CN117911794B (en) * | 2024-03-15 | 2024-06-25 | 广州中科智巡科技有限公司 | Model obtaining method and device for image classification, electronic equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106780482A (en) | A kind of classification method of medical image | |
| CN107330889B (en) | A kind of Chinese medicine tongue color coating colour automatic analysis method based on convolutional neural networks | |
| CN108806792B (en) | Deep learning face diagnosis system | |
| CN106446942A (en) | Crop disease identification method based on incremental learning | |
| CN104680524B (en) | A kind of leafy vegetable disease screening method | |
| CN107220980A (en) | A kind of MRI image brain tumor automatic division method based on full convolutional network | |
| CN103324944B (en) | A kind of based on SVM with the false fingerprint detection method of rarefaction representation | |
| CN109101994B (en) | Fundus image screening method and device, electronic equipment and storage medium | |
| CN108009594B (en) | A kind of image-recognizing method based on change grouping convolution | |
| CN107977671A (en) | A kind of tongue picture sorting technique based on multitask convolutional neural networks | |
| CN106709511A (en) | Urban rail transit panoramic monitoring video fault detection method based on depth learning | |
| CN106296699A (en) | Cerebral tumor dividing method based on deep neural network and multi-modal MRI image | |
| Mengqiao et al. | The multimodal brain tumor image segmentation based on convolutional neural networks | |
| CN106228185A (en) | A kind of general image classifying and identifying system based on neutral net and method | |
| CN106780466A (en) | A kind of cervical cell image-recognizing method based on convolutional neural networks | |
| CN108959895B (en) | Electroencephalogram EEG (electroencephalogram) identity recognition method based on convolutional neural network | |
| CN108363979A (en) | Neonatal pain expression recognition method based on binary channels Three dimensional convolution neural network | |
| CN107704886A (en) | A kind of medical image hierarchy system and method based on depth convolutional neural networks | |
| CN106600595A (en) | Human body characteristic dimension automatic measuring method based on artificial intelligence algorithm | |
| CN105205504B (en) | A kind of image attention regional quality evaluation index learning method based on data-driven | |
| CN108596264A (en) | A kind of community discovery method based on deep learning | |
| CN109325410B (en) | Electroencephalogram EEG (electroencephalogram) feature extraction method based on convolutional neural network | |
| CN107657633A (en) | A kind of soil improving straw mulching rate measuring method based on BP neural network and sensor data acquisition | |
| CN106980753B (en) | A kind of data-driven machine learning method based on voxel analysis for neurological disease | |
| CN109994202A (en) | A method of the face based on deep learning generates prescriptions of traditional Chinese medicine |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20170531 |
|
| WD01 | Invention patent application deemed withdrawn after publication |