CN109727238A - The recognition methods of x-ray chest radiograph and device - Google Patents
The recognition methods of x-ray chest radiograph and device Download PDFInfo
- Publication number
- CN109727238A CN109727238A CN201811610295.9A CN201811610295A CN109727238A CN 109727238 A CN109727238 A CN 109727238A CN 201811610295 A CN201811610295 A CN 201811610295A CN 109727238 A CN109727238 A CN 109727238A
- Authority
- CN
- China
- Prior art keywords
- ray chest
- chest radiograph
- normotopia
- side position
- convolutional network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 25
- 238000012549 training Methods 0.000 claims abstract description 53
- 241001269238 Data Species 0.000 claims abstract description 11
- 201000010099 disease Diseases 0.000 claims description 16
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 claims description 16
- 238000001514 detection method Methods 0.000 claims description 11
- 230000004913 activation Effects 0.000 claims description 9
- 230000009977 dual effect Effects 0.000 claims description 9
- 230000003044 adaptive effect Effects 0.000 claims description 5
- 238000005096 rolling process Methods 0.000 claims 1
- 238000005516 engineering process Methods 0.000 abstract description 6
- 238000012360 testing method Methods 0.000 description 4
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 238000007689 inspection Methods 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 210000005241 right ventricle Anatomy 0.000 description 2
- 208000021479 Cardiovascular injury Diseases 0.000 description 1
- 206010067286 Left atrial dilatation Diseases 0.000 description 1
- 206010027076 Mediastinal mass Diseases 0.000 description 1
- 230000006978 adaptation Effects 0.000 description 1
- 210000000709 aorta Anatomy 0.000 description 1
- 208000007474 aortic aneurysm Diseases 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 210000004204 blood vessel Anatomy 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 210000002837 heart atrium Anatomy 0.000 description 1
- 210000005240 left ventricle Anatomy 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 210000000056 organ Anatomy 0.000 description 1
- 210000001147 pulmonary artery Anatomy 0.000 description 1
Landscapes
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a kind of recognition methods of x-ray chest radiograph and devices, in the above-mentioned methods, using multiple corresponding image datas of x-ray chest radiograph for having distinguished normotopia and side position as the first input data of depth convolutional network, the training depth convolutional network；Using multiple corresponding image datas of x-ray chest radiograph for having distinguished normotopia and side position as the second input data of the depth convolutional network, the depth convolutional network after training is verified；In the case where verifying accuracy rate higher than first threshold, by the depth convolutional network after distinguishing the x-ray chest radiograph input training of normotopia and side position, recognition result is obtained.Normotopia rabat and lateral chest film can be efficiently and accurately distinguished from a large amount of x-ray chest radiographs.It solves and identifies and distinguishes among positive lateral chest film by human eye in the related technology, the problem of inefficiency.
Description
Technical field
The present invention relates to the communications fields, recognition methods and device in particular to a kind of x-ray chest radiograph.
Background technique
X-ray chest radiograph, clinically referred to as rabat, is clinically widely used, normotopia rabat can show cardiovascular injuries
Size, form, position and profile can be observed and heart and adjoin the relationship of organ and the variation of intrapulmonary blood vessel, can be used for heart and
The measurement of its radial line.The case where overall picture and left and right ventricles of left front oblique bit slice display aorta and atrium dextrum increase.Right anterior oblique position
Piece helps to observe the variation that left atrial enlargement, pulmonary artery section protrusion and right ventricle pars infundibularis increase.Left side bit slice can observe the heart,
Situations such as anteroposterior diameter and chest deformity of chest, the identification and positioning to aortic aneurysm and mediastinal mass are particularly important.
Since most of x-ray chest radiograph is normotopia rabat and lateral chest film each one, depth is trained by the sample obscured
Learning network, it may appear that very big error.And positive lateral chest film is identified and distinguished among by human eye, efficiency is very low.
Therefore, how normotopia rabat and lateral chest film are efficiently and accurately distinguished from a large amount of x-ray chest radiographs, be current
Urgent problem to be solved.
Summary of the invention
It is a primary object of the present invention to disclose a kind of recognition methods of x-ray chest radiograph and device, at least to solve correlation
The corresponding technical solution that normotopia rabat and lateral chest film are efficiently and accurately distinguished from a large amount of x-ray chest radiographs is also lacked in technology
The problem of.
According to an aspect of the invention, there is provided a kind of recognition methods of x-ray chest radiograph.
The recognition methods of x-ray chest radiograph according to the present invention includes: by multiple x-ray chest radiographs for having distinguished normotopia and side position
First input data of the corresponding image data as depth convolutional network, the above-mentioned depth convolutional network of training；By it is multiple
Second input data of the corresponding image data of x-ray chest radiograph of normotopia and side position as above-mentioned depth convolutional network is distinguished, it is right
Above-mentioned depth convolutional network after training is verified；It, will be wait distinguish just in the case where verifying accuracy rate higher than first threshold
Above-mentioned depth convolutional network after the x-ray chest radiograph input training of position and side position, obtains recognition result.
According to another aspect of the present invention, a kind of identification device of x-ray chest radiograph is provided.
The identification device of x-ray chest radiograph according to the present invention includes: training module, for by it is multiple distinguished normotopia and
First input data of the corresponding image data of x-ray chest radiograph of side position as depth convolutional network, the above-mentioned depth convolution net of training
Network；Authentication module, for using multiple corresponding image datas of x-ray chest radiograph for having distinguished normotopia and side position as above-mentioned depth
Second input data of convolutional network verifies the above-mentioned depth convolutional network after training；First obtains module, is used for
Accuracy rate is verified higher than the above-mentioned depth in the case where first threshold, after the x-ray chest radiograph input wait distinguish normotopia and side position is trained
Convolutional network is spent, recognition result is obtained.
Compared with prior art, the embodiment of the present invention has at least the following advantages: the x-ray chest radiograph of positive side position is divided into not
With sample set train depth convolutional network, the above-mentioned depth convolutional network after training verified, in verifying accuracy rate
In the case where higher than first threshold, the x-ray chest radiograph wait distinguish normotopia and side position is inputted into the above-mentioned depth convolutional network after training,
Normotopia rabat and lateral chest film can be efficiently and accurately distinguished from a large amount of x-ray chest radiographs.It solves and passes through in the related technology
The problem of human eye identifies and distinguishes among positive lateral chest film, inefficiency.
Detailed description of the invention
Fig. 1 is the flow chart of the recognition methods of x-ray chest radiograph according to an embodiment of the present invention；
Fig. 2 is the flow chart of the recognition methods of x-ray chest radiograph according to the preferred embodiment of the invention；
Fig. 3 is the structural block diagram of the identification device of x-ray chest radiograph according to an embodiment of the present invention；
Fig. 4 is the structural block diagram of the identification device of x-ray chest radiograph according to the preferred embodiment of the invention.
Specific embodiment
Specific implementation of the invention is made a detailed description with reference to the accompanying drawings of the specification.
Fig. 1 is the flow chart of the recognition methods of x-ray chest radiograph according to an embodiment of the present invention.As shown in Figure 1, the x-ray chest radiograph
Recognition methods include:
Step S101: multiple corresponding image datas of x-ray chest radiograph for having distinguished normotopia and side position are rolled up as depth
First input data of product network, the above-mentioned depth convolutional network of training；
Step S103: using multiple corresponding image datas of x-ray chest radiograph for having distinguished normotopia and side position as above-mentioned depth
The second input data for spending convolutional network, verifies the above-mentioned depth convolutional network after training；
Step S105: in the case where verifying accuracy rate higher than first threshold, by the x-ray chest radiograph of normotopia to be distinguished and side position
Above-mentioned depth convolutional network after input training, obtains recognition result.
The x-ray chest radiograph of positive side position is divided into different sample sets to train depth convolutional network, to above-mentioned after training
Depth convolutional network is verified, in the case where verifying accuracy rate higher than first threshold, by the X-ray of normotopia to be distinguished and side position
Above-mentioned depth convolutional network after rabat input training, can efficiently and accurately distinguish normotopia chest from a large amount of x-ray chest radiographs
Piece and lateral chest film.It solves and identifies and distinguishes among positive lateral chest film by human eye in the related technology, the problem of inefficiency.
In a preferred implementation process, depth convolutional network (Convolutional Neural Networks can be used
Densenet) 121, densenet121 has a 121 layer network structures, and 121 layers are full connection figure layer, i.e., full articulamentum each
Node is all connected with upper one layer of all nodes, for the characteristic synthetic that front is extracted.The spy being connected entirely due to it
Property, the parameter of general full articulamentum is also most.
Preferably, it is rolled up using multiple corresponding image datas of x-ray chest radiograph for having distinguished normotopia and side position as depth
First input data of product network can also include following processing before training above-mentioned depth convolutional network: collect for training
The x-ray chest radiograph sample of above-mentioned depth convolutional network；Above-mentioned x-ray chest radiograph sample is divided into the x-ray chest radiograph sample of normotopia and side position,
And it marks；Size is adjusted to above-mentioned x-ray chest radiograph sample, and increases above-mentioned x-ray chest radiograph in such a way that Random Level is overturn
The quantity of sample.
For example, in order to increase the quantity of x-ray chest radiograph sample, it, can be by image ruler before image is imported neural network
Very little size is adjusted, and such as narrows down to 224 × 224.Further, it is also possible to increase the number of training data by Random Level overturning
Amount.
Preferably, in above-mentioned steps S101, by multiple corresponding picture numbers of x-ray chest radiograph for having distinguished normotopia and side position
According to the first input data as depth convolutional network, the above-mentioned depth convolutional network of training be may further include: by it is multiple
The x-ray chest radiograph for being distinguished normotopia and side position is divided into multiple groups, every time will be described in the corresponding image data input of one group of x-ray chest radiograph
Depth convolutional network is trained；Dual output list is set by the final output layer that the depth convolutional network connects figure layer entirely
Member obtains each X-ray chest in first input data from the dual output unit using nonlinear s igmoid activation primitive
The output result of piece；It is the x-ray chest radiograph or non-x-ray chest radiograph of the x-ray chest radiograph of normotopia, side position according to the output result judgement；
The weight parameter value of the depth convolutional network is adjusted according to the judgement result dynamic.
For example, output be comprising normotopia x-ray chest radiograph probability value X1 and side position x-ray chest radiograph probability value Y1 the two number
According to array, successively the two numbers can be determined, by the general of the probability value X1 of normotopia x-ray chest radiograph and normotopia x-ray chest radiograph
Rate threshold X 2 is compared, if X1 is greater than or equal to X2, is directly determined as the x-ray chest radiograph of normotopia, if X1 is less than X2,
Continue to determine in next step, the probability value Y1 of side position x-ray chest radiograph is compared with the probability threshold value Y2 of side position x-ray chest radiograph, works as Y1
When more than or equal to Y2, then it is determined as the x-ray chest radiograph of side position, when Y1 is less than Y2, is then determined as non-x-ray chest radiograph, for example, it may be possible to
It is head X-ray etc..
Two data (probability value of the probability value X1 of normotopia x-ray chest radiograph and side position x-ray chest radiograph is exported using dual output unit
Y1), it not only can be determined that the x-ray chest radiograph of normotopia or the x-ray chest radiograph of side position, non-x-ray chest radiograph, such as head can also be weeded out
Portion's X-ray etc..Optionally, of course, in above-mentioned steps S101, multiple x-ray chest radiographs for having distinguished normotopia and side position are corresponding
First input data of the image data as depth convolutional network, the above-mentioned depth convolutional network of training can also wrap further
It includes: multiple x-ray chest radiographs for having distinguished normotopia and side position being divided into multiple groups, every time by the corresponding image of one group of x-ray chest radiograph
Data input above-mentioned depth convolutional network and are trained；Above-mentioned depth convolutional network is connected to the final output layer setting of figure layer entirely
It is obtained in above-mentioned first input data using nonlinear s igmoid activation primitive from above-mentioned single output unit for single output unit
The output result of each x-ray chest radiograph；It is the x-ray chest radiograph of normotopia or the x-ray chest radiograph of side position according to above-mentioned output result judgement；Root
The weight parameter value of above-mentioned depth convolutional network is adjusted according to above-mentioned judgement result dynamic.It i.e. can also be using single output unit output
One data can be the probability value of the x-ray chest radiograph of output normotopia, be also possible to the probability value of the x-ray chest radiograph of outlet side position, but
It is only to export a data, non-x-ray chest radiograph can not be weeded out, for example, it may be possible to be head X-ray etc..
In a preferred implementation process, the x-ray chest radiograph sample that multiple (for example, 7000) are used to training and check can be collected
This, wherein 80% training data (i.e. above-mentioned first input data) and 20% verify data (i.e. above-mentioned second input number
According to)；All chest x-ray pieces are marked by the method for eye recognition, are labeled as normotopia rabat (1) and lateral chest film (0),
Using above-mentioned training data as the first input data of depth convolutional network, the above-mentioned depth convolutional network of training；It specifically, can be with
Above-mentioned training data is divided into multiple groups, for example, every group 16 is opened light rabat, one group of data is inputted into depth convolutional network every time,
Depth convolutional network is trained.It sets single for the final output layer that above-mentioned depth convolutional network connects figure layer entirely and exports list
Member obtains each X-ray chest in above-mentioned first input data from above-mentioned single output unit using nonlinear s igmoid activation primitive
The output result of piece.Such as: original the last layer of densenet121 is 1000D fully-connected, softmax, i.e.,
1000 data are exported, are changed after modified are as follows: (output): Dense (None- > 1, Activation (Sigmoid).It is defeated
Result is a probability value out, for example, output is the probability value of normotopia x-ray chest radiograph, is then made a reservation for when this number is greater than or equal to
It can be the x-ray chest radiograph of normotopia according to this output result judgement when threshold value, conversely, when this number is less than predetermined threshold, it can
To be the x-ray chest radiograph of side position according to this output result judgement.
Preferably, it may further include according to the weight parameter value that above-mentioned judgement result adjusts above-mentioned depth convolutional network
It handles below: the weight parameter value of the above-mentioned depth convolutional network of random initializtion；Joined using the standard of adaptive moments estimation algorithm
Number β 1, β 2 and learning rate dynamic adjust the weight parameter value of above-mentioned depth convolutional network.
Wherein, the weight parameter value of depth convolutional network, with standard parameter β 1, the β 2 of adaptive moments estimation Adam algorithm and
Learning rate is relevant, such as: parameter a1=β 1*1* β 2+lr in weight parameter value etc., wherein lr is study
Rate.That is how the weight parameter value of depth convolutional network changes, and is standard parameter β 1, β 2 by adaptation moments estimation Adam algorithm
It is determined with learning rate.During preferred implementation, then first random initializtion network weight parameter value is joined using the standard of Adam
The end-to-end training of progress of the number (β 1=0.9 and β 2=0.999) to network, sets 0.01 for initial learning rate.Every time
One group of data in training data are inputted, constantly dynamic is carried out to the weight parameter value of depth convolutional network and is adjusted, is passed through in this way
Multiple data input and multiple adjustment are crossed, the case where accuracy rate is higher than scheduled threshold value (i.e. above-mentioned first threshold) is being verified
Under, it is determined that the training of depth convolutional network is completed.In practice using above-mentioned parameter by test, by 100 training get off with
Afterwards, verify data accuracy rate can achieve 99.9% or more.
Preferably, it by the above-mentioned depth convolutional network after distinguishing the x-ray chest radiograph input training of normotopia and side position, obtains and knows
It can also include following processing after other result: by the x-ray chest radiograph input normotopia rabat disease inspection of normotopia in above-mentioned recognition result
Survey grid network obtains the corresponding probability of illness of x-ray chest radiograph of normotopia；By the x-ray chest radiograph input side position of side position in above-mentioned recognition result
Rabat disease detection network obtains the corresponding probability of illness of x-ray chest radiograph of side position.
Preferably, the corresponding probability of illness of the x-ray chest radiograph illness corresponding with the x-ray chest radiograph for obtaining side position for obtaining normotopia is general
After rate, it can also include following processing: judge in above-mentioned recognition result, if having the X-ray for the normotopia for belonging to the same patient
The x-ray chest radiograph of rabat and side position；If so, then in conjunction with the corresponding probability of illness of x-ray chest radiograph of the normotopia, the x-ray chest radiograph of side position
The probability of illness of the above-mentioned same patient is calculated using Weighted Average Algorithm for corresponding probability of illness.
In a preferred implementation process, after obtaining the result of normotopia or lateral chest film, according to different as a result, can
The probability of illness of target rabat is obtained to be utilized respectively normotopia rabat disease detection network and lateral chest film disease detection network；Such as
Fruit positive side position x-ray chest radiograph is same people's, then can obtain the probability of illness of above-mentioned positive side position most by Weighted Average Algorithm
Whole probability of illness.Calculation formula is as follows:
Final probability of illness=(x1f1+x2f2+x3f3+x4f4+ ...+xnfn)/n；
Wherein, f1, f2 ... ... fn are weighted value, and x1, x2 ... ... xn are the corresponding trouble of x-ray chest radiograph of normotopia or side position
Sick probability, n are the x-ray chest radiograph number of same people, if having a normotopia x-ray chest radiograph and side position X-ray for same patient
Rabat, then n is 2.
Above-mentioned preferred embodiment is further described below in conjunction with Fig. 2.
Fig. 2 is the flow chart of the recognition methods of x-ray chest radiograph according to the preferred embodiment of the invention.As shown in Fig. 2, the X-ray
The recognition methods of rabat includes:
Step S201: by the above-mentioned depth convolutional network after distinguishing the x-ray chest radiograph input training of normotopia and side position
Densenet121 exports the array of the probability value Y1 of the x-ray chest radiograph for the probability value X1 of the x-ray chest radiograph comprising normotopia and side position.
Step S203: the probability value X1 of the x-ray chest radiograph of output is compared with predetermined threshold X2, when the x-ray chest radiograph
When probability value is greater than or equal to predetermined threshold, S205 is thened follow the steps, otherwise, executes step S207.
Step S205: testing result is the x-ray chest radiograph of normotopia, by the x-ray chest radiograph input normotopia rabat disease inspection of the normotopia
Survey grid network obtains the corresponding probability of illness of x-ray chest radiograph of normotopia.
Step S207: the probability value Y1 of the side position x-ray chest radiograph of output and the probability threshold value Y2 of side position x-ray chest radiograph are compared
Compared with, when Y1 is greater than or equal to Y2, S209 is thened follow the steps, otherwise, execution step S211.
Step S209: testing result is the x-ray chest radiograph of side position, by the x-ray chest radiograph input normotopia rabat disease inspection of side position
Survey grid network obtains the corresponding probability of illness of x-ray chest radiograph of side position.
Step S211: testing result is non-x-ray chest radiograph, for example, it may be possible to be head X-ray etc..
Step S213: judgement identification after x-ray chest radiograph in, if having the normotopia for belonging to the same patient x-ray chest radiograph and
The x-ray chest radiograph of side position；If so, thening follow the steps S215, otherwise, S217 or step S219 is thened follow the steps.
Step S215: general in conjunction with the corresponding probability of illness of x-ray chest radiograph of the normotopia, the corresponding illness of x-ray chest radiograph of side position
The probability of illness of the same patient is calculated using Weighted Average Algorithm for rate.
Final probability of illness=(x1f1+x2f2)/2；
Wherein, f1 is the corresponding weighted value of x-ray chest radiograph of normotopia, and f2 is the corresponding weighted value of x-ray chest radiograph of side position, and x1 is
The corresponding probability of illness of the x-ray chest radiograph of normotopia, x2 are the corresponding probability of illness of x-ray chest radiograph of side position.
Step S217: the corresponding probability of illness of x-ray chest radiograph of normotopia is exported.
Step S219: the corresponding probability of illness of the x-ray chest radiograph of outlet side position.
Fig. 3 is the structural block diagram of the identification device of x-ray chest radiograph according to an embodiment of the present invention.As shown in figure 3, the X-ray chest
The identification device of piece includes: training module 30, for by it is multiple distinguished normotopia and side position the corresponding images of x-ray chest radiograph
First input data of the data as depth convolutional network, the above-mentioned depth convolutional network of training；Authentication module 32, being used for will be multiple
The corresponding image data of x-ray chest radiograph for having distinguished normotopia and side position inputs number as the second of above-mentioned depth convolutional network
According to being verified to the above-mentioned depth convolutional network after training；First obtains module 34, for being higher than first in verifying accuracy rate
In the case where threshold value, by the above-mentioned depth convolutional network after distinguishing the x-ray chest radiograph input training of normotopia and side position, identification is obtained
As a result.
Using above-mentioned apparatus, the x-ray chest radiograph of positive side position is divided into different sample sets to train depth by training module 30
Convolutional network, the above-mentioned depth convolutional network after 32 pairs of training of authentication module are verified, and first obtains module 34 in verifying standard
In the case that true rate is higher than first threshold, by the above-mentioned depth convolution after distinguishing the x-ray chest radiograph input training of normotopia and side position
Network can efficiently and accurately distinguish normotopia rabat and lateral chest film from a large amount of x-ray chest radiographs.It solves in the related technology
Positive lateral chest film is identified and distinguished among by human eye, the problem of inefficiency.
Preferably, as shown in figure 4, above-mentioned training module 30 may further include: grouped element 300, being used for will be multiple
The x-ray chest radiograph for having distinguished normotopia and side position is divided into multiple groups, every time will be in the corresponding image data input of one group of x-ray chest radiograph
Depth convolutional network is stated to be trained；Acquiring unit 302, for above-mentioned depth convolutional network to be connected to the final output of figure layer entirely
Layer is set as dual output unit, using nonlinear s igmoid activation primitive, obtains above-mentioned first input from above-mentioned dual output unit
The output result of each x-ray chest radiograph in data；Judging unit 304, for being the X-ray chest of normotopia according to above-mentioned output result judgement
Piece, side position x-ray chest radiograph or non-x-ray chest radiograph；Adjustment unit 306, for adjusting above-mentioned depth according to above-mentioned judgement result dynamic
Spend the weight parameter value of convolutional network.
Optionally, above-mentioned acquiring unit 302 can be also used for the depth convolutional network connecting the final defeated of figure layer entirely
Layer is set as single output unit out, and using nonlinear s igmoid activation primitive, it is defeated to obtain described first from single output unit
Enter the output result of each x-ray chest radiograph in data；Above-mentioned judging unit 304 can also be just according to the output result judgement
The x-ray chest radiograph of position or the x-ray chest radiograph of side position.
Preferably, above-mentioned adjustment unit 306 is further used for the weight parameter of the above-mentioned depth convolutional network of random initializtion
Value adjusts the power of above-mentioned depth convolutional network using standard parameter β 1, β 2 and the learning rate dynamic of adaptive moments estimation algorithm
Weight parameter value.
Preferably, as shown in figure 4, above-mentioned apparatus can also include: the second acquisition module 36,34 phase of module is obtained with first
Connection obtains the X-ray of normotopia for the x-ray chest radiograph of normotopia in above-mentioned recognition result to be inputted normotopia rabat disease detection network
The corresponding probability of illness of rabat；Third obtains module 38, is connected with the first acquisition module 34, being used for will be in above-mentioned recognition result
The x-ray chest radiograph of side position inputs lateral chest film disease detection network, obtains the corresponding probability of illness of x-ray chest radiograph of side position；Judge mould
Block 40 obtains module 38 with the second acquisition module 36 and third respectively and is connected, for judging in above-mentioned recognition result, if having
Belong to the x-ray chest radiograph of the normotopia of the same patient and the x-ray chest radiograph of side position；Computing module 42 is connected with judgment module 40, uses
When the x-ray chest radiograph of the x-ray chest radiograph and side position that have the normotopia for belonging to the same patient in the output of above-mentioned judgment module, just in conjunction with this
Position the corresponding probability of illness of x-ray chest radiograph, side position the corresponding probability of illness of x-ray chest radiograph, calculated using Weighted Average Algorithm
The probability of illness of the above-mentioned same patient out.
In conclusion the x-ray chest radiograph of positive side position to be divided into different samples by above-described embodiment provided by the invention
Collection verifies the depth convolutional network after training to train depth convolutional network, is higher than in verifying accuracy rate predetermined
It, can be efficient by the depth convolutional network after distinguishing the x-ray chest radiograph input training of normotopia and side position in the case where threshold value
And normotopia rabat and lateral chest film are accurately distinguished from a large amount of x-ray chest radiographs.It solves and is known in the related technology by human eye
Not and positive lateral chest film is distinguished, the problem of inefficiency.In the positive side position of identification x-ray chest radiograph and then in conjunction with positive lateral chest film
Disease detection network obtains positive side position disease and changes disease probability, can for belonging to the normotopia and side position x-ray chest radiograph of a patient
To obtain the higher disease detection conclusion of accuracy rate by Weighted Average Algorithm.
Disclosed above is only several specific embodiments of the invention, and still, the present invention is not limited to this, any ability
What the technical staff in domain can think variation should all fall into protection scope of the present invention.
Claims (10)
1. a kind of recognition methods of x-ray chest radiograph characterized by comprising
Using multiple corresponding image datas of x-ray chest radiograph for having distinguished normotopia and side position as the first defeated of depth convolutional network
Enter data, the training depth convolutional network；
Using multiple corresponding image datas of x-ray chest radiograph for having distinguished normotopia and side position as the of the depth convolutional network
Two input datas verify the depth convolutional network after training；
It, will be after distinguishing the x-ray chest radiograph input training of normotopia and side position in the case where verifying accuracy rate higher than first threshold
The depth convolutional network obtains recognition result.
2. the method according to claim 1, wherein by multiple x-ray chest radiographs for having distinguished normotopia and side position
First input data of the corresponding image data as depth convolutional network, the training depth convolutional network include:
Multiple x-ray chest radiographs for having distinguished normotopia and side position are divided into multiple groups, every time by the corresponding image of one group of x-ray chest radiograph
Data input the depth convolutional network and are trained；
Dual output unit is set by the final output layer that the depth convolutional network connects figure layer entirely, use is non-linear
Sigmoid activation primitive obtains the output result of each x-ray chest radiograph in first input data from the dual output unit；
It is the x-ray chest radiograph or non-x-ray chest radiograph of the x-ray chest radiograph of normotopia, side position according to the output result judgement；
The weight parameter value of the depth convolutional network is adjusted according to the judgement result dynamic.
3. the method according to claim 1, wherein by multiple x-ray chest radiographs for having distinguished normotopia and side position
First input data of the corresponding image data as depth convolutional network, the training depth convolutional network include:
Multiple x-ray chest radiographs for having distinguished normotopia and side position are divided into multiple groups, every time by the corresponding image of one group of x-ray chest radiograph
Data input the depth convolutional network and are trained；
Single output unit is set by the final output layer that the depth convolutional network connects figure layer entirely, use is non-linear
Sigmoid activation primitive obtains the output result of each x-ray chest radiograph in first input data from single output unit；
It is the x-ray chest radiograph of normotopia or the x-ray chest radiograph of side position according to the output result judgement；
The weight parameter value of the depth convolutional network is adjusted according to the judgement result dynamic.
4. according to the method in claim 2 or 3, which is characterized in that adjust the depth convolution according to the judgement result
The weight parameter value of network includes:
The weight parameter value of depth convolutional network described in random initializtion；
The power of the depth convolutional network is adjusted using standard parameter β 1, β 2 and the learning rate dynamic of adaptive moments estimation algorithm
Weight parameter value.
5. the method according to claim 1, wherein the x-ray chest radiograph of normotopia to be distinguished and side position is inputted training
The depth convolutional network afterwards, obtain recognition result after, further includes:
The x-ray chest radiograph of normotopia in the recognition result is inputted into normotopia rabat disease detection network, obtains the x-ray chest radiograph pair of normotopia
The probability of illness answered；
The x-ray chest radiograph of side position in the recognition result is inputted into lateral chest film disease detection network, obtains the x-ray chest radiograph pair of side position
The probability of illness answered.
6. according to the method described in claim 5, it is characterized in that, obtaining the corresponding probability of illness of x-ray chest radiograph of normotopia and obtaining
After the corresponding probability of illness of x-ray chest radiograph for taking side position, further includes:
Judge in the recognition result, if having the x-ray chest radiograph for the normotopia for belonging to the same patient and the x-ray chest radiograph of side position；
If so, then being adopted in conjunction with the corresponding probability of illness of x-ray chest radiograph of the normotopia, the corresponding probability of illness of x-ray chest radiograph of side position
The probability of illness of the same patient is calculated with Weighted Average Algorithm.
7. a kind of identification device of x-ray chest radiograph characterized by comprising
Training module, for rolling up multiple corresponding image datas of x-ray chest radiograph for having distinguished normotopia and side position as depth
First input data of product network, the training depth convolutional network；
Authentication module, for using multiple corresponding image datas of x-ray chest radiograph for having distinguished normotopia and side position as the depth
The second input data for spending convolutional network, verifies the depth convolutional network after training；
First obtains module, is used in the case where verifying accuracy rate higher than first threshold, by the X-ray of normotopia to be distinguished and side position
The depth convolutional network after rabat input training, obtains recognition result.
8. device according to claim 7, which is characterized in that the training module includes:
Grouped element, for multiple x-ray chest radiographs for having distinguished normotopia and side position to be divided into multiple groups, every time by one group of X-ray
The corresponding image data of rabat inputs the depth convolutional network and is trained；
Acquiring unit, the final output layer for the depth convolutional network to be connected figure layer entirely are set as dual output unit, adopt
With nonlinear s igmoid activation primitive, the defeated of each x-ray chest radiograph in first input data is obtained from the dual output unit
Result out；
Judging unit, for being the x-ray chest radiograph of normotopia or the x-ray chest radiograph or non-X of side position according to the output result judgement
Light rabat；
Adjustment unit, for adjusting the weight parameter value of the depth convolutional network according to the judgement result dynamic.
9. device according to claim 8, which is characterized in that the adjustment unit is further used for random initializtion institute
The weight parameter value for stating depth convolutional network, using standard parameter β 1, β 2 and the learning rate dynamic of adaptive moments estimation algorithm
Adjust the weight parameter value of the depth convolutional network.
10. device according to claim 7, which is characterized in that further include:
Second obtains module, for the x-ray chest radiograph of normotopia in the recognition result to be inputted normotopia rabat disease detection network, obtains
Take the corresponding probability of illness of the x-ray chest radiograph of normotopia；
Third obtains module, for the x-ray chest radiograph of side position in the recognition result to be inputted lateral chest film disease detection network, obtains
Take the corresponding probability of illness of x-ray chest radiograph of side position；
Judgment module, for judging in the recognition result, if having x-ray chest radiograph and the side position of the normotopia for belonging to the same patient
X-ray chest radiograph；
Computing module, for there is the X-ray of the x-ray chest radiograph for the normotopia for belonging to the same patient and side position in judgment module output
When rabat, in conjunction with the normotopia the corresponding probability of illness of x-ray chest radiograph, side position the corresponding probability of illness of x-ray chest radiograph, using weighting
The probability of illness of the same patient is calculated in average algorithm.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811610295.9A CN109727238A (en) | 2018-12-27 | 2018-12-27 | The recognition methods of x-ray chest radiograph and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811610295.9A CN109727238A (en) | 2018-12-27 | 2018-12-27 | The recognition methods of x-ray chest radiograph and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109727238A true CN109727238A (en) | 2019-05-07 |
Family
ID=66296590
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811610295.9A Pending CN109727238A (en) | 2018-12-27 | 2018-12-27 | The recognition methods of x-ray chest radiograph and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109727238A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113053520A (en) * | 2021-04-02 | 2021-06-29 | 北京掌引医疗科技有限公司 | Training method and device for tuberculosis detection model and auxiliary diagnosis equipment |
| CN113076993A (en) * | 2021-03-31 | 2021-07-06 | 零氪智慧医疗科技(天津)有限公司 | Information processing method and model training method for chest X-ray film recognition |
| CN113129343A (en) * | 2020-01-14 | 2021-07-16 | 通用电气精准医疗有限责任公司 | Method and system for anatomical structure/view classification in X-ray imaging |
| CN113643261A (en) * | 2021-08-13 | 2021-11-12 | 江南大学 | Lung disease diagnosis method based on frequency attention network |
Citations (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102418518A (en) * | 2011-04-12 | 2012-04-18 | 北京师范大学 | Method for identifying water logging grades of oil reservoir by using neural network analogue cross plot |
| CN103186774A (en) * | 2013-03-21 | 2013-07-03 | 北京工业大学 | Semi-supervised learning-based multi-gesture facial expression recognition method |
| CN104318246A (en) * | 2014-10-20 | 2015-01-28 | 西安电子科技大学 | Depth self-adaption ridgelet network based polarimetric SAR (Synthetic Aperture Radar) image classification |
| CN104598686A (en) * | 2015-01-24 | 2015-05-06 | 安徽大学 | Water pump motor modeling and optimizing method based on electromagnetic calculation and neural network |
| CN105631479A (en) * | 2015-12-30 | 2016-06-01 | 中国科学院自动化研究所 | Imbalance-learning-based depth convolution network image marking method and apparatus |
| US20170109881A1 (en) * | 2015-10-14 | 2017-04-20 | The Regents Of The University Of California | Automated segmentation of organ chambers using deep learning methods from medical imaging |
| CN106845401A (en) * | 2017-01-20 | 2017-06-13 | 中国科学院合肥物质科学研究院 | A kind of insect image-recognizing method based on many spatial convoluted neutral nets |
| US20170169314A1 (en) * | 2015-12-15 | 2017-06-15 | Qualcomm Incorporated | Methods for object localization and image classification |
| CN107292234A (en) * | 2017-05-17 | 2017-10-24 | 南京邮电大学 | It is a kind of that method of estimation is laid out based on information edge and the indoor scene of multi-modal feature |
| CN107730484A (en) * | 2017-07-26 | 2018-02-23 | 江西中科九峰智慧医疗科技有限公司 | A kind of abnormal rabat intelligent identification Method and system based on deep learning |
| CN108230277A (en) * | 2018-02-09 | 2018-06-29 | 中国人民解放军战略支援部队信息工程大学 | A kind of dual intensity CT picture breakdown methods based on convolutional neural networks |
| CN108520206A (en) * | 2018-03-22 | 2018-09-11 | 南京大学 | A kind of fungi microscopic image identification method based on full convolutional neural networks |
| WO2018166114A1 (en) * | 2017-03-13 | 2018-09-20 | 平安科技（深圳）有限公司 | Picture identification method and system, electronic device, and medium |
| CN108898175A (en) * | 2018-06-26 | 2018-11-27 | 北京工业大学 | Area of computer aided model building method based on deep learning gastric cancer pathological section |
-
2018
- 2018-12-27 CN CN201811610295.9A patent/CN109727238A/en active Pending
Patent Citations (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102418518A (en) * | 2011-04-12 | 2012-04-18 | 北京师范大学 | Method for identifying water logging grades of oil reservoir by using neural network analogue cross plot |
| CN103186774A (en) * | 2013-03-21 | 2013-07-03 | 北京工业大学 | Semi-supervised learning-based multi-gesture facial expression recognition method |
| CN104318246A (en) * | 2014-10-20 | 2015-01-28 | 西安电子科技大学 | Depth self-adaption ridgelet network based polarimetric SAR (Synthetic Aperture Radar) image classification |
| CN104598686A (en) * | 2015-01-24 | 2015-05-06 | 安徽大学 | Water pump motor modeling and optimizing method based on electromagnetic calculation and neural network |
| US20170109881A1 (en) * | 2015-10-14 | 2017-04-20 | The Regents Of The University Of California | Automated segmentation of organ chambers using deep learning methods from medical imaging |
| US20170169314A1 (en) * | 2015-12-15 | 2017-06-15 | Qualcomm Incorporated | Methods for object localization and image classification |
| CN105631479A (en) * | 2015-12-30 | 2016-06-01 | 中国科学院自动化研究所 | Imbalance-learning-based depth convolution network image marking method and apparatus |
| CN106845401A (en) * | 2017-01-20 | 2017-06-13 | 中国科学院合肥物质科学研究院 | A kind of insect image-recognizing method based on many spatial convoluted neutral nets |
| WO2018166114A1 (en) * | 2017-03-13 | 2018-09-20 | 平安科技（深圳）有限公司 | Picture identification method and system, electronic device, and medium |
| CN107292234A (en) * | 2017-05-17 | 2017-10-24 | 南京邮电大学 | It is a kind of that method of estimation is laid out based on information edge and the indoor scene of multi-modal feature |
| CN107730484A (en) * | 2017-07-26 | 2018-02-23 | 江西中科九峰智慧医疗科技有限公司 | A kind of abnormal rabat intelligent identification Method and system based on deep learning |
| CN108230277A (en) * | 2018-02-09 | 2018-06-29 | 中国人民解放军战略支援部队信息工程大学 | A kind of dual intensity CT picture breakdown methods based on convolutional neural networks |
| CN108520206A (en) * | 2018-03-22 | 2018-09-11 | 南京大学 | A kind of fungi microscopic image identification method based on full convolutional neural networks |
| CN108898175A (en) * | 2018-06-26 | 2018-11-27 | 北京工业大学 | Area of computer aided model building method based on deep learning gastric cancer pathological section |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113129343A (en) * | 2020-01-14 | 2021-07-16 | 通用电气精准医疗有限责任公司 | Method and system for anatomical structure/view classification in X-ray imaging |
| CN113076993A (en) * | 2021-03-31 | 2021-07-06 | 零氪智慧医疗科技(天津)有限公司 | Information processing method and model training method for chest X-ray film recognition |
| CN113076993B (en) * | 2021-03-31 | 2023-04-18 | 零氪智慧医疗科技(天津)有限公司 | Information processing method and model training method for chest X-ray film recognition |
| CN113053520A (en) * | 2021-04-02 | 2021-06-29 | 北京掌引医疗科技有限公司 | Training method and device for tuberculosis detection model and auxiliary diagnosis equipment |
| CN113643261A (en) * | 2021-08-13 | 2021-11-12 | 江南大学 | Lung disease diagnosis method based on frequency attention network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Ji et al. | Learning calibrated medical image segmentation via multi-rater agreement modeling | |
| CN109727238A (en) | The recognition methods of x-ray chest radiograph and device | |
| Mishra et al. | Diabetic retinopathy detection using deep learning | |
| CN110245721A (en) | Training method, device and the electronic equipment of neural network model | |
| CN109919928A (en) | Detection method, device and the storage medium of medical image | |
| CN107832802A (en) | Quality of human face image evaluation method and device based on face alignment | |
| CN110310723A (en) | Bone image processing method, electronic equipment and storage medium | |
| CN109192305B (en) | Heart function automatic analysis method based on deep circulation neural network | |
| CN110517256A (en) | A kind of early carcinoma assistant diagnosis system based on artificial intelligence | |
| CN109508644A (en) | Facial paralysis grade assessment system based on the analysis of deep video data | |
| CN110111885B (en) | Attribute prediction method, attribute prediction device, computer equipment and computer readable storage medium | |
| CN106846380B (en) | A kind of medical image registration method and equipment | |
| Hsu et al. | Hierarchical Network for Facial Palsy Detection. | |
| CN110731773A (en) | abnormal electrocardiogram screening method based on fusion of global and local depth features of electrocardiogram | |
| CN114359629B (en) | Deep migration learning-based X-chest X-ray pneumonia classification and identification method | |
| JP2019503214A (en) | Fast automatic segmentation of hierarchical images by heuristic graph search | |
| CN111028230A (en) | Fundus image optic disc and macula lutea positioning detection algorithm based on YOLO-V3 | |
| WO2022105735A1 (en) | Coronary artery segmentation method and apparatus, electronic device, and computer-readable storage medium | |
| CN110060315A (en) | A kind of image motion artifact eliminating method and system based on artificial intelligence | |
| CN116258726A (en) | Temporal-mandibular joint MRI image important structure segmentation method based on deep learning | |
| CN110969633A (en) | Automatic optimal phase recognition method for cardiac CT imaging | |
| Han et al. | A convolutional neural network-based anthropomorphic model observer for signal-known-statistically and background-known-statistically detection tasks | |
| CN110288574A (en) | A kind of adjuvant Ultrasonographic Diagnosis hepatoncus system and method | |
| Lu et al. | Dynamic spatio-temporal graph convolutional networks for cardiac motion analysis | |
| CN114092463A (en) | Digital breast tomography focus positioning device |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190507 |