CN108510482A - Cervical carcinoma detection method, device, equipment and medium based on gynecatoptron image - Google Patents
Cervical carcinoma detection method, device, equipment and medium based on gynecatoptron image Download PDFInfo
- Publication number
- CN108510482A CN108510482A CN201810241901.8A CN201810241901A CN108510482A CN 108510482 A CN108510482 A CN 108510482A CN 201810241901 A CN201810241901 A CN 201810241901A CN 108510482 A CN108510482 A CN 108510482A
- Authority
- CN
- China
- Prior art keywords
- image
- cervical carcinoma
- gynecatoptron
- convolutional neural
- neural networks
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/12—Edge-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- 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
- G06T2207/30096—Tumor; Lesion
Abstract
The invention discloses a kind of cervical carcinoma detection method, device, terminal device and computer readable storage medium based on gynecatoptron image, method includes：In the cervical carcinoma detection model based on two-way convolutional neural networks：The positioning and extraction that opening of the cervix position is carried out to the gynecatoptron image of acquisition, to generate the ROI image for including uterine neck carninomatosis hair region；The extraction at edge is split to the ROI image by two-way convolutional neural networks, to generate cutting image；By classifying, convolutional neural networks carry out cancer grade separation to the cutting image, to export the lesion grade of cervical carcinoma, it must fast and accurately obtain cervical carcinoma testing result, help lack experience doctor quickly judge diseased region, find atypia diseased region, lesion degree, materials position are judged, to finding that cervical carcinoma and precancerous lesion play help and facilitation in time.
Description
Technical field
The present invention relates to medical image process field more particularly to a kind of cervical carcinoma detection sides based on gynecatoptron image
Method, device, terminal device and computer readable storage medium.
Background technology
Image segmentation algorithm just has existed in traditional digital image processing field, main threshold application dividing method and
Its various mutation.Amplify out image detection algorithm again on the basis of image segmentation algorithm, but they to the grade of image procossing not
Together, detection is primarily directed to image-region rank, and divides and need to handle each pixel, finer and smoother to image procossing.In depth
Spend learning areas, the partitioning algorithm based on convolutional network is also more and more, using it is most be full convolutional network (FCN).
Screening is according to " three ladders " method, i.e., leading Liquid based cytology test (LCT) and human milk before the cancer of existing cervical carcinoma
Head tumor virus (Human Papillomavirus, HPV) detects, and abnormal person carries out uterine neck, vagina lesion biopsy under gynecatoptron, after
With operative treatment.Cervical biopsy is to diagnose the indispensable step of cervical lesions under gynecatoptron.Vaginoscopy is carried out in doctor
When screening and biopsy are directly carried out by image, received the period of training and horizontal different, the image interpretation energy of doctor by doctor
Power directly affects biopsy results.Meanwhile directly the judgement of cervical carcinoma carried out to digital image piece and segmentation be it is the simplest and
Direct method, but judging result is limited by physician specialty ability, is not popularized.Presently the most effective method
It is the diagnosis in cytology field, and has also obtained computerized algorithm auxiliary, and also applies some conventional segmentation methods,
But it is limited by conventional segmentation methods, the selection result also needs to doctor and rejudges, cumbersome.
During implementing the embodiment of the present invention, inventor has found：Clinically the method for cervical carcinoma screening is mainly
LCT/HPV, gynecatoptron and histopathology, wherein cytology screening have been incorporated with traditional image cutting method, and take
Obtain certain effect, but diagnostic error sometimes；Gynecatoptron is also common inspection method, but its accuracy is by doctors experience level
And training cycle influences, directly affect diagnostic result.
In digital image processing field, image segmentation is the algorithm being commonly used, but dividing method mentioned above exists
Respective drawback, threshold segmentation method is relatively simple on application principle, is easy by image intensity profile itself and noise
The influence of the factors such as interference, the threshold value obtained merely with grey level histogram can not make image segmentation obtain it is satisfied as a result, by
To being affected for image irradiation.Additional light filling is needed when doing vaginoscopy, can seriously affect cutting effect.
FCN cutting algorithms based on convolutional neural networks substantially improve the above situation, can well solve illumination not
Ill effect caused by equilibrium, and cutting effect is also more smooth.But effect is poor on segmentation wisp, with network layer
Secondary intensification causes the missing of wisp information more although the deeper network of level can preferably be fitted sample distribution
Seriously.
Invention content
In view of the above-mentioned problems, the cervical carcinoma detection method that the purpose of the present invention is to provide a kind of based on gynecatoptron image,
Device, terminal device and computer readable storage medium must fast and accurately obtain cervical carcinoma testing result, help to lack experience
Doctor quickly judges diseased region, finds atypia diseased region, lesion degree, materials position is judged, to finding uterine neck in time
Cancer and precancerous lesion play facilitation.
In a first aspect, an embodiment of the present invention provides a kind of cervical carcinoma detection methods based on gynecatoptron image, including with
Lower step：
In the cervical carcinoma detection model based on two-way convolutional neural networks：
The positioning and extraction that opening of the cervix position is carried out to the gynecatoptron image of acquisition include that uterine neck carninomatosis sends out region to generate
ROI image；
The extraction at edge is split to the ROI image by two-way convolutional neural networks, to generate cutting image；
By classifying, convolutional neural networks carry out cancer grade separation to the cutting image, to export the lesion of cervical carcinoma
Grade.
In the first realization method of first aspect, it is described by two-way convolutional neural networks to the ROI image into
The extraction on row segmentation side edge, to generate cutting image, specially：
In two-way convolutional neural networks：
The characteristic value superposition that convolution algorithm and multilayer feature are carried out to the ROI image, to obtain and the ROI image ruler
Very little identical thermal map spectrum；
Progress de-convolution operation is composed to the thermal map and high-level characteristic is superimposed with the characteristic value of low-level feature, to extract side
Along generation cutting image.
According to the first realization method of first aspect, in second of realization method of first aspect, the thermal map spectrum
In each pixel corresponding to numerical value be used to indicate that the pixel in the same position of the gynecatoptron image to belong to disease and sends out pixel
Probability；
It is then described that cancer grade separation is carried out to the cutting image by convolutional neural networks of classifying, to export cervical carcinoma
Lesion grade, specially：
According to the cutting image, disease hair pixel non-to each of the gynecatoptron image carries out processes pixel, with life
At processing image；
The superposition calculation for carrying out the characteristic dimension of multilayer feature to the processing image by classification convolutional neural networks, with
Obtain the lesion grade of the cervical carcinoma corresponding to the gynecatoptron image.
It is described according to institute in the third realization method of first aspect according to second of realization method of first aspect
Cutting image is stated, disease hair pixel non-to each of the gynecatoptron image carries out processes pixel, to generate processing image, specifically
For：
According to the cutting image, the position of the non-disease hair pixel of each of described gynecatoptron image is obtained；
When it is described it is non-disease hair pixel uterine neck carninomatosis send out region cutting edge other than when, by it is described it is non-disease hair pixel picture
Plain value is set to 0.
According to second of realization method of first aspect, in the 4th kind of realization method of first aspect, the cervical carcinoma
Lesion grade include low level lesion, high-level lesion, cervical carcinoma, chronic cervicitis and normal-sub uterine neck.
In the 5th kind of realization method of first aspect, the cervical carcinoma detection model based on two-way convolutional neural networks
Training step include：
The gynecatoptron sample image of acquisition is pre-processed, to generate first sample；
Network is generated according to confrontation, sample expansion is carried out to the first sample, to obtain the second sample；
Determining for opening of the cervix position is carried out to the gynecatoptron image in second sample based on the location algorithm of deep learning
Position and extraction, to generate trained ROI image；
Obtain the ROI- true pictures that user carries out disease hair area marking according to training ROI image；
The two-way convolutional neural networks are trained according to the ROI- true pictures, to generate trained cutting drawing
Picture；Wherein, the convolution kernel of two-way every layer of the convolutional neural networks is 3, and total level is 20 layers, and each layer includes convolutional layer and anti-
Convolutional layer；
According to the trained cutting image, processes pixel is carried out to each non-disease hair pixel in second sample, to obtain
Take processes pixel image；
Feedback label of the user to the processes pixel image is obtained, to generate third sample；
The classification convolutional neural networks are trained according to the third sample, to obtain the disease of the cervical carcinoma of classification
Become grade；Wherein, the convolution kernel of every layer of the convolutional neural networks of classification is 3, and total level is 10 layers.
Further include in the 6th kind of realization method of first aspect according to the 5th of first aspect the kind of realization method：
Costing bio disturbance is carried out according to the ROI- true pictures and the trained cutting image；Wherein, if costing bio disturbance letter
Number is DLOSS, thenX indicates the cutting edge coordinate of user annotation in ROI- true pictures, y
Indicate the edge coordinate of the training cutting image of deconvolution neural network forecast.
Second aspect, an embodiment of the present invention provides a kind of cervical carcinoma detection devices based on gynecatoptron image, including with
Lower step：
In the cervical carcinoma detection model based on two-way convolutional neural networks：
Extraction module is positioned, positioning and extraction for carrying out opening of the cervix position to the gynecatoptron image of acquisition, to generate
The ROI image in region is sent out including uterine neck carninomatosis；
Cutting image acquisition module, for being split edge to the ROI image by two-way convolutional neural networks
Extraction, to generate cutting image；
Level results acquisition module, for carrying out cancer ranking score to the cutting image by convolutional neural networks of classifying
Class, to export the lesion grade of cervical carcinoma.
The third aspect, the cervical carcinoma detection terminal equipment based on gynecatoptron image that an embodiment of the present invention provides a kind of, packet
It includes processor, memory and is stored in the memory and is configured as the computer program executed by the processor,
The processor realized when executing the computer program it is any one of above-mentioned described in the cervical carcinoma based on gynecatoptron image
Detection method.
Fourth aspect, an embodiment of the present invention provides a kind of computer readable storage medium, the computer-readable storage
Medium includes the computer program of storage, wherein controls the computer-readable storage medium when the computer program is run
Equipment where matter execute it is any one of above-mentioned described in the cervical carcinoma detection method based on gynecatoptron image.
The cervical carcinoma detection method that an embodiment of the present invention provides a kind of based on gynecatoptron image, device, terminal device and
Computer readable storage medium has the advantages that：
The gynecatoptron image of acquisition is being input to the cervical carcinoma detection model based on two-way convolutional neural networks into temporary dwelling palace
When neck cancer detects, in the cervical carcinoma detection model based on two-way convolutional neural networks：First to the gynecatoptron image of acquisition into
Then the positioning and extraction at temporary dwelling palace eck position pass through two-way convolution god to generate the ROI image for including uterine neck carninomatosis hair region
The extraction at edge is split to the ROI image through network, to generate cutting image, finally by classification convolutional neural networks
Cancer grade separation is carried out to the cutting image, to export the lesion grade of cervical carcinoma, by two-way convolutional neural networks and
Feature Fusion applies different neural network models in different phase, can be partitioned into small cancerous lesion region, soon
Speed and accurate Ground Split and the lesion grade for sorting out cervical carcinoma carry out can be carried out operating after a small amount of training in doctor,
Can also be substantially reduced by the constraint of physician specialty knowledge, the hospital of not medical practitioner or remote districts can also be accurate
Ground carries out cervical carcinoma screening and diagnosis, and the doctor that can help to lack experience quickly judges diseased region, finds atypia lesion
Position judges lesion degree, materials position, and to finding that cervical carcinoma and precancerous lesion play facilitation in time, the present invention passes through
Artificial intelligence identifies the uterine neck image that Via vagina mirror obtains, and timely and accurately judges various cervical lesions positions, instructs doctor accurate
It really obtains pathological tissues and carries out pathological examination, it might even be possible to substitute traditional cytolgical examination, quickly find cervical carcinoma and cancer
Preceding lesion has prodigious society and medical value.
Description of the drawings
In order to illustrate more clearly of technical scheme of the present invention, attached drawing needed in embodiment will be made below
Simply introduce, it should be apparent that, the accompanying drawings in the following description is only some embodiments of the present invention, general for this field
For logical technical staff, without creative efforts, other drawings may also be obtained based on these drawings.
Fig. 1 is the flow signal for the cervical carcinoma detection method based on gynecatoptron image that first embodiment of the invention provides
Figure.
Fig. 2 is the schematic diagram for the gynecatoptron image that the terminal device that first embodiment of the invention provides obtains.
Fig. 3 is the schematic diagram for the ROI image including uterine neck carninomatosis hair region that first embodiment of the invention provides.
Fig. 4 is the schematic diagram of the thermal map spectrum for the acquisition that first embodiment of the invention provides.
Fig. 5 is the schematic diagram of the cutting image for the acquisition that first embodiment of the invention provides.
Fig. 6 is the schematic diagram for the two-way convolutional neural networks that first embodiment of the invention provides.
Fig. 7 is the schematic diagram for the classification convolutional neural networks that first embodiment of the invention provides.
Fig. 8 is the training for the cervical carcinoma detection model based on two-way convolutional neural networks that second embodiment of the invention provides
The flow diagram of step.
Fig. 9 is the structural representation for the cervical carcinoma detection device based on gynecatoptron image that third embodiment of the invention provides
Figure.
Specific implementation mode
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation describes, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts every other
Embodiment shall fall within the protection scope of the present invention.
Referring to Fig. 1, first embodiment of the invention provides a kind of cervical carcinoma detection method based on gynecatoptron image,
It can be executed, and included the following steps by terminal device：
In the cervical carcinoma detection model based on two-way convolutional neural networks：
S11 carries out the gynecatoptron image of acquisition the positioning and extraction at opening of the cervix position, includes uterine neck carninomatosis hair to generate
The ROI image in region.
In embodiments of the present invention, the terminal device can be flat for desktop PC, notebook, palm PC, intelligence
The computing devices such as plate and cloud server, particularly, the terminal device can be cloud server, as long as the terminal in hospital is clapped
The gynecatoptron picture taken the photograph, Cloud Server can return to segmentation picture and lesion grade in 2s.
In embodiments of the present invention, in the cervical carcinoma detection model based on two-way convolutional neural networks, referring to Fig. 2,
When acquiring gynecatoptron picture, does not ensure that there was only opening of the cervix in picture, other body parts is also had under most of situation,
So in order to eliminate the influence at other positions, needs first to orient the position of opening of the cervix and extract the corresponding ROI image in its position
As the picture of subsequent processing, ROI image, i.e. area-of-interest use the specified targets for wanting to read in of ROI, it is possible to reduce processing
Time increases precision, offers convenience to image procossing；Referring to Fig. 3, the terminal device obtains the gynecatoptron figure of user's transmission
Picture carries out the gynecatoptron image by the location algorithm based on deep learning the positioning and extraction at opening of the cervix position, with life
At the ROI image for sending out region including uterine neck carninomatosis, the location algorithm based on deep learning can robustly navigate to palace very much
Collar area, it should be noted that disease hair region is opening of the cervix position.
S12 is split the ROI image by two-way convolutional neural networks the extraction at edge, to generate cutting drawing
Picture.
In embodiments of the present invention, specifically, in two-way convolutional neural networks：Referring to Fig. 4, the terminal device pair
The ROI image carries out the characteristic value superposition of convolution algorithm and multilayer feature, to obtain heat identical with the ROI image size
Collection of illustrative plates, referring to Fig. 5, then folded to thermal map spectrum progress de-convolution operation and the characteristic value of high-level characteristic and low-level feature
Add, cutting image is generated to extract edge, referring to Fig. 6, C indicates that convolutional layer, DC indicate warp lamination, the terminal in Fig. 6
The characteristic value that equipment carries out the ROI image convolution algorithm and multilayer feature is superimposed, using the mutual syncretizing mechanism of multilayer feature,
The fusion is characteristic value superposition, i.e., the feature of two identical dimensionals carries out digital addition, do not change characteristic spectrum dimension, every layer
The width and height of characteristic spectrum are consistent, and abandon the sample level that can reduce characteristic spectrum size (wide and high), and characteristic value is added
Refer to the identical characteristic image of two dimensions, the numerical value that same position is corresponded in two characteristic images carries out summation operation, keeps
The width and height of image, such as a feature isIt is further characterized asSo it is after feature additionImage
Output after convolution algorithm is the feature of this layer, reduces the width and height of image, institute due to not having when carrying out convolution algorithm
It is identical with the characteristic dimension when feature is superimposed, the convolutional layer of first half has obtained the disease hair region of the ROI image
Thermal map spectrum, the width of thermal map spectrum and high identical as gynecatoptron image input by user, the number corresponding to each pixel that thermal map is composed
The pixel that value represents the gynecatoptron image position belongs to the probability that disease is sent out, and the extraneous edge of thermal map spectrum is that disease hair region is coarse
And fuzzy transitional region, there is no apparent edge, then the terminal device composes the thermal map and carries out de-convolution operation,
And high-level characteristic is superimposed with the characteristic value of low-level feature, with the intensification of convolutional network, the feature of Small object can be more and more unknown
It is aobvious, in some instances it may even be possible to it disappears, so Network Low-layer time feature, same coordinate low-level feature and high-level characteristic weighted superposition are also needed to,
A new characteristic spectrum is formed, with the intensification of feature, low-level feature can slowly disappear, but substantially improve original signal damage
Mistake degree, thermal map spectrum obtain apparent edge image by the deconvolution network of latter half, generate cutting image, cutting image
Different zones different colors is presented.It should be noted that in view of the smaller convolution kernel of convolutional neural networks can be kept
Fine and smooth details in image, so every layer of convolution kernel size is 3, it is 2 that when each convolution algorithm, which adds picture to expand size, and is used
Xavier is uniformly distributed to initialize weight, because neural network can preserve every layer of parameter, as neural network weight, is used for
Image convolution operation, i.e. convolution algorithm are exactly that the weight parameter and image carry out convolution algorithm；Total level of bilateral network is 20
Layer, all layers are convolutional layer and warp lamination, and characteristic spectrum size is consistent.
S13, by classifying, convolutional neural networks carry out cancer grade separation to the cutting image, to export cervical carcinoma
Lesion grade.
In embodiments of the present invention, the terminal device is according to the cutting image, to every in the gynecatoptron image
A non-disease hair pixel carries out processes pixel, to generate processing image, specifically, the terminal device according to the cutting image,
The position of the non-disease hair pixel of each of described gynecatoptron image is obtained, if the non-disease hair pixel sends out region in uterine neck carninomatosis
Within cutting edge, then the pixel value of the non-disease hair pixel is constant；When the non-disease hair pixel sends out region in uterine neck carninomatosis
When other than cutting edge, the pixel value of the non-disease hair pixel is set to 0 by the terminal device, referring to Fig. 7, the then end
End equipment carries out the processing image by classification convolutional neural networks the superposition calculation of the characteristic dimension of multilayer feature, to obtain
Take the lesion grade of the cervical carcinoma corresponding to the gynecatoptron image, the lesion grade of the cervical carcinoma include low level lesion,
High-level lesion, cervical carcinoma, chronic cervicitis and normal-sub uterine neck, the classification convolutional neural networks use multilayer feature and melt
The method of conjunction and smaller convolution kernel, fusion method are the increases of characteristic dimension, that is, splice two features and form dimension more
More features, such as a feature isIt is further characterized asSo after merging featuresDescribed point
Class convolutional neural networks, described every layer of convolution kernel size of classification convolutional neural networks are 3, and are uniformly distributed come just using xavier
Beginningization weight, total level is 10 layers, as shown in fig. 7, C is convolutional layer, P is sample level, and FC is full articulamentum.
In conclusion first embodiment of the invention provides a kind of cervical carcinoma detection method based on gynecatoptron image,
When the gynecatoptron image of acquisition is input to the progress cervical carcinoma detection of the cervical carcinoma detection model based on two-way convolutional neural networks,
In the cervical carcinoma detection model based on two-way convolutional neural networks：Opening of the cervix position is carried out to the gynecatoptron image of acquisition first
Positioning and extraction, with generate include uterine neck carninomatosis hair region ROI image, then by two-way convolutional neural networks to described
ROI image is split the extraction at edge, to generate cutting image, finally by classification convolutional neural networks to the cutting drawing
As carrying out cancer grade separation, to export the lesion grade of cervical carcinoma, by two-way convolutional neural networks and Feature Fusion,
Different neural network models is applied in different phase, in order to adapt to wisp segmentation, in each neural computing process
In used multilayer feature integration technology, while remaining the feature of low layer and top layer, small cancerous lesion area can be partitioned into
Domain fast and accurately Ground Split and sorts out the lesion grade of cervical carcinoma, in doctor can be carried out after a small amount of training
Operation, can also be substantially reduced by the constraint of physician specialty knowledge, also may be used in the hospital of not medical practitioner or remote districts
Accurately to carry out cervical carcinoma screening and diagnosis, the doctor that can help to lack experience quickly judges diseased region, finds atypia
Diseased region judges lesion degree, materials position, to finding that cervical carcinoma and precancerous lesion play facilitation, this hair in time
The bright uterine neck image for identifying that Via vagina mirror is obtained by artificial intelligence, timely and accurately judges various cervical lesions positions, instructs
Doctor accurately obtains pathological tissues and carries out pathological examination, it might even be possible to substitute traditional cytolgical examination, quickly find uterine neck
Cancer and precancerous lesion have prodigious society and medical value.
In order to facilitate the understanding of the present invention, some currently preferred embodiments of the present invention will be done and will further be retouched below
It states.
Second embodiment of the invention：
Referring to Fig. 8, on the basis of first embodiment of the invention, the cervical carcinoma based on two-way convolutional neural networks
The training step of detection model includes：
S21 pre-processes the gynecatoptron sample image of acquisition, to generate first sample.
In embodiments of the present invention, specifically, in order to ensure the reliability of gynecatoptron sample image quality, ensureing not change
Data enhancing is done to former gynecatoptron sample image under the premise of becoming gynecatoptron sample image contrast, Enhancement Method is mainly：
Pass through translation, overturning, plus noise etc..Overturning is to carry out 3 directions to picture to rotate and the rotation of artwork mirror image picture；Addition
Noise be common Gaussian noise, to form pretreated first sample.
S22 generates network according to confrontation and carries out sample expansion to the first sample, to obtain the second sample.
In embodiments of the present invention, since the patient for doing vaginoscopy is not very much, useful sample is just less, is
Prevent the sample very few and caused by network over-fitting, need to do sample expansion before training.The terminal device according to
Confrontation generates network (GAN) and carries out sample expansion to the first sample, can generate some samples very true to nature, described right
Antibiosis is a kind of deep learning model at network (GAN), is the method for unsupervised learning most foreground in complex distributions in recent years
One of, model passes through (at least) two modules in frame：Generate model (Generative Model) and discrimination model
The mutual Game Learning of (Discriminative Model) generates fairly good output.
S23 carries out opening of the cervix position based on the location algorithm of deep learning to the gynecatoptron image in second sample
Positioning and extraction, to generate trained ROI image.
In embodiments of the present invention, when acquiring gynecatoptron picture, do not ensure that there was only opening of the cervix in picture, it is most of
Other body parts are also had under situation, so in order to eliminate the influence at other positions, are first oriented the position of opening of the cervix and are carried
Take ROI pictures as the samples pictures of subsequent processing, the terminal device is with the location algorithm based on deep learning to described
Gynecatoptron image in second sample carries out the positioning and extraction at opening of the cervix position, to generate trained ROI image.
S24 obtains the ROI- true pictures that user carries out disease hair area marking according to training ROI image；
In embodiments of the present invention, after acquisition includes the training ROI image in disease hair region at opening of the cervix position, by institute
It states trained ROI image and feeds back to user (i.e. doctor), with the help of the doctor of professional experiences, the edge line mark in disease hair region
It outpours and, the region surrounded by edge line is that true disease sends out region, and then the terminal device obtains user according to training ROI
Image carry out disease hair area marking ROI- true pictures, the coordinate of these real estates of mark is saved, using as
The training sample of the two-way convolutional neural networks.
S25 is trained the two-way convolutional neural networks according to the ROI- true pictures, to generate training cutting
Image；Wherein, the convolution kernel of two-way every layer of the convolutional neural networks be 3, total level be 20 layers, each layer include convolutional layer and
Warp lamination.
In embodiments of the present invention, the terminal device according to the ROI- true pictures to the two-way convolutional Neural net
Network is trained, and to generate trained cutting image, using the mutual syncretizing mechanism of multilayer feature, which is characteristic value superposition,
The feature of i.e. two identical dimensionals carries out digital addition, does not change characteristic spectrum dimension, and the width and height of every layer of characteristic spectrum are protected
The sample level for unanimously abandoning and reducing characteristic spectrum size (wide and high) is held, meanwhile, with the intensification of convolutional network, small mesh
Target feature can increasingly unobvious, in some instances it may even be possible to disappear, so also needing to Network Low-layer time feature, same coordinate low-level feature
With high-level characteristic weighted superposition, a new characteristic spectrum is formed, with the intensification of feature, low-level feature can slowly disappear, but
Substantially improve original loss of signal degree, it is contemplated that the smaller convolution kernel of convolutional neural networks can keep fine and smooth in image
Details, so every layer of convolution kernel size is 3, it is 2 that when each convolution algorithm, which adds picture to expand size, and is uniformly divided using xavier
Cloth initializes weight, and total level of two-way convolutional neural networks is 20 layers, all layers are convolutional layer and warp lamination, feature
Collection of illustrative plates size is consistent.
In embodiments of the present invention, referring to Fig. 6, in the training process, region is sent out in order to distinguish normal region and disease
Boundary uses the fine or not degree that two loss functions carry out scoring model, and total LOSS values are the sum of two loss functions, and LOSS is got over
Mini Mod is trained better.As shown in fig. 6, (two-way convolutional neural networks joining place) uses among two-way convolutional neural networks
The convolutional layer of SoftMaxLoss loss functions, first half has obtained the thermal map spectrum in sick hair region, the width of thermal map spectrum and high and institute
The numerical value for stating each pixel that gynecatoptron sample image is identical, and thermal map is composed represents corresponding described gynecatoptron sample image position
Pixel belong to the probability of disease hair, the extraneous edge of thermal map spectrum is the coarse and fuzzy transitional region in disease hair region, bright
Aobvious edge.Thermal map spectrum obtains apparent edge image by the deconvolution network of latter half, as trains cutting image, instruction
Different colors is presented in the different zones for practicing cutting image, and second loss function is connected behind deconvolution, anti-for judging
Similarity between training cutting image and the ROI- true pictures that convolutional network generates, according to the ROI- true pictures
And the trained cutting image carries out costing bio disturbance, the training cutting image and the ROI- that deconvolution neural network generates are true
Similarity Appraisal process is as follows between image：
1) cutting edge for the training cutting image and the ROI- true pictures that extraction deconvolution network generates, forms side
The coordinate of each pixel on edge is indicated with different vectors respectively.
2) the LOSS values between two vectors are calculated according to the following formula, are indicated with DLOSS.
Wherein x indicates the cutting edge coordinate of user annotation in ROI- true pictures, y
Indicate the edge coordinate of the training cutting image of deconvolution neural network forecast.
S26 carries out processes pixel according to the trained cutting image to each non-disease hair pixel in second sample,
To obtain processes pixel image.
In embodiments of the present invention, according to the trained cutting image, the non-disease of each of the gynecatoptron image is obtained
Send out the position of pixel；If the non-disease hair pixel is within the cutting edge that uterine neck carninomatosis sends out region, the non-disease sends out pixel
Pixel value is constant；If the non-disease hair pixel is other than the cutting edge that uterine neck carninomatosis sends out region, the picture of the non-disease hair pixel
Plain value is set to 0, and processes pixel image is formed after the completion of processing.
S27 obtains feedback label of the user to the processes pixel image, to generate third sample.
In embodiments of the present invention, described after the non-disease hair area pixel processing is completed to form processes pixel image
Terminal device gives the processes pixel image feedback to user (i.e. doctor), by medical practitioner to the processes pixel image mark
Label, are divided into according to cancer grade：Low level lesion, high-level lesion, cervical carcinoma, chronic cervicitis and normal, the terminal is set
It is standby to obtain feedback label of the user to the processes pixel image, to generate third sample, as the classification convolutional Neural net
The training sample of network.
S28 is trained the classification convolutional neural networks according to the third sample, to obtain the cervical carcinoma of classification
Lesion grade；Wherein, the convolution kernel of every layer of the convolutional neural networks of classification is 3, and total level is 10 layers.
In embodiments of the present invention, the terminal device according to the third sample to the classification convolutional neural networks into
Row training, to obtain the lesion grade of the cervical carcinoma of classification, the classification convolutional neural networks use multilayer feature fusion
Method and smaller convolution kernel, fusion method are the increases of characteristic dimension, that is, splice two features and to form dimension more
Feature, the classification convolutional neural networks, described every layer of convolution kernel size of classification convolutional neural networks is 3, and uses xavier
It is uniformly distributed to initialize weight, total level is 10 layers, as shown in fig. 7, C is convolutional layer, P is sample level, and FC is full connection
Layer carries out costing bio disturbance finally using SoftMaxLoss loss functions.
In embodiments of the present invention, using 1000 cervical carcinoma gynecatoptron pictures as training sample original image, each
Cancer grade is respectively 200, and for 200 crane pictures as test original image, each cancer grade is respectively 40,
200 test samples of middle test are not in 1000 training samples.After sample preprocessing and sample expand, training sample
10000 are extended to from 1000, test sample is without expanding, and in test sample, segmentation is assessed using three indexs
Accuracy rate, respectively segmentation precision, over-segmentation rate and less divided rate have reached 95% or more wherein calculating and obtaining segmentation precision,
Over-segmentation rate is less than 3%, and less divided rate is less than 3%, and grade separation accuracy rate is more than 90%.Also subsequently through increase training sample
It can continue to improve segmentation precision and classification accuracy, while reduce over-segmentation rate and less divided rate.
Referring to Fig. 9, third embodiment of the invention provides a kind of cervical carcinoma detection device based on gynecatoptron image, packet
Include following steps：
In the cervical carcinoma detection model based on two-way convolutional neural networks：
Extraction module 11 is positioned, positioning and extraction for carrying out opening of the cervix position to the gynecatoptron image of acquisition, with life
At the ROI image for sending out region including uterine neck carninomatosis.
Cutting image acquisition module 12 is split edge for passing through two-way convolutional neural networks to the ROI image
Extraction, to generate cutting image.
Level results acquisition module 13, for carrying out cancer grade to the cutting image by convolutional neural networks of classifying
Classification, to export the lesion grade of cervical carcinoma.
In the first realization method of 3rd embodiment, the cutting image acquisition module 12 specifically includes：
In two-way convolutional neural networks：
Thermal map composes acquiring unit, and the characteristic value for carrying out convolution algorithm and multilayer feature to the ROI image is superimposed, with
Obtain thermal map spectrum identical with the ROI image size.
Cutting image generation unit carries out de-convolution operation and high-level characteristic and low-level feature for being composed to the thermal map
Characteristic value superposition, with extract edge generate cutting image.
The first realization method according to third embodiment, in second of realization method of 3rd embodiment, the heat
The numerical value corresponding to each pixel in collection of illustrative plates is used to indicate that the pixel in the same position of the gynecatoptron image to belong to disease hair
The probability of pixel.
Then the level results acquisition module 13 specifically includes：
Image generation unit is handled, for according to the cutting image, disease hair non-to each of the gynecatoptron image
Pixel carries out processes pixel, to generate processing image.
Lesion grade acquiring unit, for carrying out multilayer feature to the processing image by convolutional neural networks of classifying
The superposition calculation of characteristic dimension, to obtain the lesion grade of the cervical carcinoma corresponding to the gynecatoptron image.
Second of realization method according to third embodiment, in the third realization method of 3rd embodiment, the place
Reason image generation unit specifically includes：
Position acquisition subelement, for according to the cutting image, obtaining the non-disease hair of each of described gynecatoptron image
The position of pixel.
First pixel processing unit, when it is described it is non-disease hair pixel uterine neck carninomatosis send out region cutting edge other than when, will
The pixel value of the non-disease hair pixel is set to 0.
Second of realization method according to third embodiment, in the 4th kind of realization method of 3rd embodiment, the palace
The lesion grade of neck cancer includes low level lesion, high-level lesion, cervical carcinoma, chronic cervicitis and normal-sub uterine neck.
In the 5th kind of realization method of 3rd embodiment, the cervical carcinoma based on two-way convolutional neural networks detects mould
The training step of type includes：
First sample generation module is pre-processed for the gynecatoptron sample image to acquisition, to generate first sample.
Second sample generation module carries out sample expansion, to obtain for generating network according to confrontation to the first sample
Take the second sample.
Training ROI image generation module, for the location algorithm based on deep learning to the vagina in second sample
Mirror image carries out the positioning and extraction at opening of the cervix position, to generate trained ROI image.
ROI- true picture acquisition modules carry out disease hair area marking for obtaining user according to training ROI image
ROI- true pictures.
Two-way convolutional neural networks training module is used for according to the ROI- true pictures to the two-way convolutional Neural net
Network is trained, to generate trained cutting image；Wherein, the convolution kernel of two-way every layer of the convolutional neural networks is 3, total level
It it is 20 layers, each layer includes convolutional layer and warp lamination.
Processes pixel image collection module is used for according to the trained cutting image, to each non-in second sample
Disease hair pixel carries out processes pixel, to obtain processes pixel image.
Third sample generation module, the feedback label for obtaining user to the processes pixel image, to generate third
Sample.
Classify convolutional neural networks training module, for according to the third sample to the classification convolutional neural networks into
Row training, to obtain the lesion grade of the cervical carcinoma of classification；Wherein, the convolution kernel of every layer of the convolutional neural networks of classification is 3,
Total level is 10 layers.
5th kind of realization method according to third embodiment further include in the 6th kind of realization method of 3rd embodiment：
Costing bio disturbance module, for carrying out costing bio disturbance according to the ROI- true pictures and the trained cutting image；
Wherein, if costing bio disturbance function is DLOSS, thenX indicates user annotation in ROI- true pictures
Cutting edge coordinate, y indicate deconvolution neural network forecast training cutting image edge coordinate.
Fourth embodiment of the invention provides a kind of cervical carcinoma detection terminal equipment based on gynecatoptron image.The embodiment
The cervical carcinoma detection terminal equipment based on gynecatoptron image include：It processor, memory and is stored in the memory
And the computer program that can be run on the processor, such as the cervical carcinoma based on gynecatoptron image detect program.The place
Reason device is realized when executing the computer program in above-mentioned each cervical carcinoma detection method embodiment based on gynecatoptron image
Step, such as step S11 shown in FIG. 1.Alternatively, the processor realizes that above-mentioned each device is real when executing the computer program
Apply the function of each module/unit in example, such as cutting image acquisition module.
Illustratively, the computer program can be divided into one or more module/units, one or more
A module/unit is stored in the memory, and is executed by the processor, to complete the present invention.It is one or more
A module/unit can be the series of computation machine program instruction section that can complete specific function, and the instruction segment is for describing institute
State implementation procedure of the computer program in the cervical carcinoma detection terminal equipment based on gynecatoptron image.
The cervical carcinoma detection terminal equipment based on gynecatoptron image can be desktop PC, notebook, palm
The computing devices such as computer and cloud server.The cervical carcinoma detection terminal equipment based on gynecatoptron image may include, but not
It is only limitted to, processor, memory.It will be understood by those skilled in the art that above-mentioned component is only based on the palace of gynecatoptron image
The example of neck cancer detection terminal equipment does not constitute the restriction to the cervical carcinoma detection terminal equipment based on gynecatoptron image, can
To include component more more or fewer than above-mentioned component, certain components or different components are either combined, such as described is based on
The cervical carcinoma detection terminal equipment of gynecatoptron image can also include input-output equipment, network access equipment, bus etc..
Alleged processor can be central processing unit (Central Processing Unit, CPU), can also be it
His general processor, digital signal processor (Digital Signal Processor, DSP), application-specific integrated circuit
(Application Specific Integrated Circuit, ASIC), ready-made programmable gate array (Field-
Programmable Gate Array, FPGA) either other programmable logic device, discrete gate or transistor logic,
Discrete hardware components etc..General processor can be microprocessor or the processor can also be any conventional processor
Deng the processor is the control centre of the cervical carcinoma detection terminal equipment based on gynecatoptron image, utilizes various interfaces
The entire various pieces of the cervical carcinoma detection terminal equipment based on gynecatoptron image with connection.
The memory can be used for storing the computer program and/or module, and the processor is by running or executing
Computer program in the memory and/or module are stored, and calls the data being stored in memory, described in realization
The various functions of cervical carcinoma detection terminal equipment based on gynecatoptron image.The memory can mainly include storing program area and
Storage data field, wherein storing program area can storage program area, application program (such as the sound needed at least one function
Playing function, image player function etc.) etc..In addition, memory may include high-speed random access memory, can also include non-
Volatile memory, such as hard disk, memory, plug-in type hard disk, intelligent memory card (Smart Media Card, SMC), safe number
Word (Secure Digital, SD) block, flash card (Flash Card), at least one disk memory, flush memory device or its
His volatile solid-state part.
Wherein, if module/unit of the cervical carcinoma detection terminal integration of equipments based on gynecatoptron image is with software
The form of functional unit is realized and when sold or used as an independent product, can be stored in a computer-readable storage
In medium.Based on this understanding, the present invention realizes all or part of flow in above-described embodiment method, can also pass through meter
Calculation machine program is completed to instruct relevant hardware, and the computer program can be stored in a computer readable storage medium
In, the computer program is when being executed by processor, it can be achieved that the step of above-mentioned each embodiment of the method.Wherein, the calculating
Machine program includes computer program code, and the computer program code can be source code form, object identification code form, can hold
Style of writing part or certain intermediate forms etc..The computer-readable medium may include：The computer program code can be carried
Any entity or device, recording medium, USB flash disk, mobile hard disk, magnetic disc, CD, computer storage, read-only memory (ROM,
Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunications letter
Number and software distribution medium etc..It should be noted that the content that the computer-readable medium includes can be managed according to the administration of justice
Local legislation and the requirement of patent practice carry out increase and decrease appropriate, such as in certain jurisdictions, according to legislation and patent
Practice, computer-readable medium do not include electric carrier signal and telecommunication signal.
It should be noted that the apparatus embodiments described above are merely exemplary, wherein described be used as separating component
The unit of explanation may or may not be physically separated, and the component shown as unit can be or can also
It is not physical unit, you can be located at a place, or may be distributed over multiple network units.It can be according to actual
It needs that some or all of module therein is selected to achieve the purpose of the solution of this embodiment.In addition, device provided by the invention
In embodiment attached drawing, the connection relation between module indicates there is communication connection between them, specifically can be implemented as one or
A plurality of communication bus or signal wire.Those of ordinary skill in the art are without creative efforts, you can to understand
And implement.
The above is the preferred embodiment of the present invention, it is noted that for those skilled in the art
For, various improvements and modifications may be made without departing from the principle of the present invention, these improvements and modifications are also considered as
Protection scope of the present invention.
Claims (10)
1. a kind of cervical carcinoma detection method based on gynecatoptron image, which is characterized in that include the following steps：
In the cervical carcinoma detection model based on two-way convolutional neural networks：
The positioning and extraction that opening of the cervix position is carried out to the gynecatoptron image of acquisition include that uterine neck carninomatosis sends out region to generate
ROI image；
The extraction at edge is split to the ROI image by two-way convolutional neural networks, to generate cutting image；
By classifying, convolutional neural networks carry out cancer grade separation to the cutting image, to export the lesion etc. of cervical carcinoma
Grade.
2. the cervical carcinoma detection method according to claim 1 based on gynecatoptron image, which is characterized in that described by double
The extraction at edge is split to the ROI image to convolutional neural networks, to generate cutting image, specially：
In two-way convolutional neural networks：
The characteristic value superposition that convolution algorithm and multilayer feature are carried out to the ROI image, to obtain and the ROI image size phase
Same thermal map spectrum；
Progress de-convolution operation is composed to the thermal map and high-level characteristic is superimposed with the characteristic value of low-level feature, to extract edge life
At cutting image.
3. the cervical carcinoma detection method according to claim 2 based on gynecatoptron image, which is characterized in that the thermal map spectrum
In each pixel corresponding to numerical value be used to indicate that the pixel in the same position of the gynecatoptron image to belong to disease and sends out pixel
Probability；
It is then described that cancer grade separation is carried out to the cutting image by convolutional neural networks of classifying, to export the disease of cervical carcinoma
Become grade, specially：
According to the cutting image, disease hair pixel non-to each of the gynecatoptron image carries out processes pixel, to generate place
Manage image；
By convolutional neural networks of classifying to the superposition calculation for handling image and carrying out the characteristic dimension of multilayer feature, to obtain
The lesion grade of cervical carcinoma corresponding to the gynecatoptron image.
4. according to claim 3 according to the cutting image, which is characterized in that it is described according to the cutting image, it is right
The non-disease hair pixel of each of the gynecatoptron image carries out processes pixel, to generate processing image, specially：
According to the cutting image, the position of the non-disease hair pixel of each of described gynecatoptron image is obtained；
When it is described it is non-disease hair pixel uterine neck carninomatosis send out region cutting edge other than when, by it is described it is non-disease hair pixel pixel value
It is set to 0.
5. the cervical carcinoma detection method according to claim 3 based on gynecatoptron image, which is characterized in that the cervical carcinoma
Lesion grade include low level lesion, high-level lesion, cervical carcinoma, chronic cervicitis and normal-sub uterine neck.
6. the cervical carcinoma detection method according to claim 1 based on gynecatoptron image, which is characterized in that described based on double
Training step to the cervical carcinoma detection model of convolutional neural networks includes：
The gynecatoptron sample image of acquisition is pre-processed, to generate first sample；
Network is generated according to confrontation, sample expansion is carried out to the first sample, to obtain the second sample；
Based on the location algorithm of deep learning to the gynecatoptron image in second sample carry out opening of the cervix position positioning and
Extraction, to generate trained ROI image；
Obtain the ROI- true pictures that user carries out disease hair area marking according to training ROI image；
The two-way convolutional neural networks are trained according to the ROI- true pictures, to generate trained cutting image；Its
In, the convolution kernel of two-way every layer of the convolutional neural networks is 3, and total level is 20 layers, and each layer includes convolutional layer and deconvolution
Layer；
According to the trained cutting image, processes pixel is carried out to each non-disease hair pixel in second sample, to obtain picture
Element processing image；
Feedback label of the user to the processes pixel image is obtained, to generate third sample；
The classification convolutional neural networks are trained according to the third sample, to obtain the lesion etc. of the cervical carcinoma of classification
Grade；Wherein, the convolution kernel of every layer of the convolutional neural networks of classification is 3, and total level is 10 layers.
7. the cervical carcinoma detection method based on gynecatoptron image according to the claim 6, which is characterized in that also wrap
It includes：
Costing bio disturbance is carried out according to the ROI- true pictures and the trained cutting image；Wherein, if costing bio disturbance function is
DLOSS, thenX indicates that the cutting edge coordinate of user annotation in ROI- true pictures, y indicate
The edge coordinate of the training cutting image of deconvolution neural network forecast.
8. a kind of cervical carcinoma detection device based on gynecatoptron image, which is characterized in that include the following steps：
In the cervical carcinoma detection model based on two-way convolutional neural networks：
Extraction module is positioned, positioning and extraction for carrying out opening of the cervix position to the gynecatoptron image of acquisition include to generate
Uterine neck carninomatosis sends out the ROI image in region；
Cutting image acquisition module, the extraction for being split edge to the ROI image by two-way convolutional neural networks,
To generate cutting image；
Level results acquisition module, for carrying out cancer grade separation to the cutting image by convolutional neural networks of classifying,
To export the lesion grade of cervical carcinoma.
9. a kind of cervical carcinoma detection terminal equipment based on gynecatoptron image, including processor, memory and it is stored in described
In memory and it is configured as the computer program executed by the processor, when the processor executes the computer program
Realize the cervical carcinoma detection method based on gynecatoptron image as claimed in any of claims 1 to 7 in one of claims.
10. a kind of computer readable storage medium, which is characterized in that the computer readable storage medium includes the calculating of storage
Machine program, wherein equipment where controlling the computer readable storage medium when the computer program is run is executed as weighed
Profit requires the cervical carcinoma detection method based on gynecatoptron image described in any one of 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810241901.8A CN108510482B (en) | 2018-03-22 | 2018-03-22 | Cervical cancer detection device based on colposcope images |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810241901.8A CN108510482B (en) | 2018-03-22 | 2018-03-22 | Cervical cancer detection device based on colposcope images |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108510482A true CN108510482A (en) | 2018-09-07 |
| CN108510482B CN108510482B (en) | 2020-12-04 |
Family
ID=63378218
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810241901.8A Expired - Fee Related CN108510482B (en) | 2018-03-22 | 2018-03-22 | Cervical cancer detection device based on colposcope images |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108510482B (en) |
Cited By (20)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109427060A (en) * | 2018-10-30 | 2019-03-05 | 腾讯科技（深圳）有限公司 | A kind of method, apparatus, terminal device and the medical system of image identification |
| CN109493340A (en) * | 2018-11-28 | 2019-03-19 | 武汉大学人民医院（湖北省人民医院） | Esophagus fundus ventricularis varication assistant diagnosis system and method under a kind of gastroscope |
| CN109584229A (en) * | 2018-11-28 | 2019-04-05 | 武汉大学人民医院（湖北省人民医院） | A kind of real-time assistant diagnosis system of Endoscopic retrograde cholangio-pancreatiography art and method |
| CN109949271A (en) * | 2019-02-14 | 2019-06-28 | 腾讯科技（深圳）有限公司 | A kind of detection method based on medical image, the method and device of model training |
| CN110033456A (en) * | 2019-03-07 | 2019-07-19 | 腾讯科技（深圳）有限公司 | A kind of processing method of medical imaging, device, equipment and system |
| CN110033445A (en) * | 2019-04-10 | 2019-07-19 | 司法鉴定科学研究院 | Medicolegal examination automatic identification system and recognition methods based on deep learning |
| CN110136113A (en) * | 2019-05-14 | 2019-08-16 | 湖南大学 | A kind of vagina pathology image classification method based on convolutional neural networks |
| CN110334565A (en) * | 2019-03-21 | 2019-10-15 | 江苏迪赛特医疗科技有限公司 | A kind of uterine neck neoplastic lesions categorizing system of microscope pathological photograph |
| CN110490850A (en) * | 2019-02-14 | 2019-11-22 | 腾讯科技（深圳）有限公司 | A kind of lump method for detecting area, device and Medical Image Processing equipment |
| CN110516665A (en) * | 2019-08-23 | 2019-11-29 | 上海眼控科技股份有限公司 | Identify the neural network model construction method and system of image superposition character area |
| CN110706794A (en) * | 2019-09-26 | 2020-01-17 | 中国科学院深圳先进技术研究院 | Medical image processing system and medical image processing method |
| CN111436972A (en) * | 2020-04-13 | 2020-07-24 | 王时灿 | Three-dimensional ultrasonic gynecological disease diagnosis device |
| CN111476794A (en) * | 2019-01-24 | 2020-07-31 | 武汉兰丁医学高科技有限公司 | UNET-based cervical pathological tissue segmentation method |
| CN111914841A (en) * | 2020-08-07 | 2020-11-10 | 温州医科大学 | CT image processing method and device |
| CN112435242A (en) * | 2020-11-25 | 2021-03-02 | 江西中科九峰智慧医疗科技有限公司 | Lung image processing method and device, electronic equipment and storage medium |
| CN112884707A (en) * | 2021-01-15 | 2021-06-01 | 复旦大学附属妇产科医院 | Cervical precancerous lesion detection system, equipment and medium based on colposcope |
| WO2021114832A1 (en) * | 2020-05-28 | 2021-06-17 | 平安科技（深圳）有限公司 | Sample image data enhancement method, apparatus, electronic device, and storage medium |
| WO2021139447A1 (en) * | 2020-09-30 | 2021-07-15 | 平安科技（深圳）有限公司 | Abnormal cervical cell detection apparatus and method |
| CN113710166A (en) * | 2020-03-19 | 2021-11-26 | 艾多特公司 | Carotid artery ultrasonic diagnosis system |
| TWI767506B (en) * | 2020-02-26 | 2022-06-11 | 大陸商上海商湯智能科技有限公司 | Image recognition method, training method and equipment of recognition model |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8488863B2 (en) * | 2008-11-06 | 2013-07-16 | Los Alamos National Security, Llc | Combinational pixel-by-pixel and object-level classifying, segmenting, and agglomerating in performing quantitative image analysis that distinguishes between healthy non-cancerous and cancerous cell nuclei and delineates nuclear, cytoplasm, and stromal material objects from stained biological tissue materials |
| CN106339591A (en) * | 2016-08-25 | 2017-01-18 | 汤平 | Breast cancer prevention self-service health cloud service system based on deep convolutional neural network |
| CN106780466A (en) * | 2016-12-21 | 2017-05-31 | 广西师范大学 | A kind of cervical cell image-recognizing method based on convolutional neural networks |
| CN107609503A (en) * | 2017-09-05 | 2018-01-19 | 刘宇红 | Intelligent cancerous tumor cell identifying system and method, cloud platform, server, computer |
| US20180061046A1 (en) * | 2016-08-31 | 2018-03-01 | International Business Machines Corporation | Skin lesion segmentation using deep convolution networks guided by local unsupervised learning |
-
2018
- 2018-03-22 CN CN201810241901.8A patent/CN108510482B/en not_active Expired - Fee Related
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8488863B2 (en) * | 2008-11-06 | 2013-07-16 | Los Alamos National Security, Llc | Combinational pixel-by-pixel and object-level classifying, segmenting, and agglomerating in performing quantitative image analysis that distinguishes between healthy non-cancerous and cancerous cell nuclei and delineates nuclear, cytoplasm, and stromal material objects from stained biological tissue materials |
| CN106339591A (en) * | 2016-08-25 | 2017-01-18 | 汤平 | Breast cancer prevention self-service health cloud service system based on deep convolutional neural network |
| US20180061046A1 (en) * | 2016-08-31 | 2018-03-01 | International Business Machines Corporation | Skin lesion segmentation using deep convolution networks guided by local unsupervised learning |
| CN106780466A (en) * | 2016-12-21 | 2017-05-31 | 广西师范大学 | A kind of cervical cell image-recognizing method based on convolutional neural networks |
| CN107609503A (en) * | 2017-09-05 | 2018-01-19 | 刘宇红 | Intelligent cancerous tumor cell identifying system and method, cloud platform, server, computer |
Non-Patent Citations (2)
| Title |
|---|
| TAO XU 等: "Multimodal Deep Learning", 《MICCAI 2016》 * |
| 谢珍珠 等: "边缘增强深层网络的图像超分辨率重建", 《中国图象图形学报》 * |
Cited By (28)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11610310B2 (en) | 2018-10-30 | 2023-03-21 | Tencent Technology (Shenzhen) Company Limited | Method, apparatus, system, and storage medium for recognizing medical image |
| CN109427060A (en) * | 2018-10-30 | 2019-03-05 | 腾讯科技（深圳）有限公司 | A kind of method, apparatus, terminal device and the medical system of image identification |
| US11410306B2 (en) | 2018-10-30 | 2022-08-09 | Tencent Technology (Shenzhen) Company Limited | Method, apparatus, system, and storage medium for recognizing medical image |
| CN109493340A (en) * | 2018-11-28 | 2019-03-19 | 武汉大学人民医院（湖北省人民医院） | Esophagus fundus ventricularis varication assistant diagnosis system and method under a kind of gastroscope |
| CN109584229A (en) * | 2018-11-28 | 2019-04-05 | 武汉大学人民医院（湖北省人民医院） | A kind of real-time assistant diagnosis system of Endoscopic retrograde cholangio-pancreatiography art and method |
| CN111476794A (en) * | 2019-01-24 | 2020-07-31 | 武汉兰丁医学高科技有限公司 | UNET-based cervical pathological tissue segmentation method |
| CN111476794B (en) * | 2019-01-24 | 2023-10-20 | 武汉兰丁智能医学股份有限公司 | Cervical pathological tissue segmentation method based on UNET |
| CN110490850A (en) * | 2019-02-14 | 2019-11-22 | 腾讯科技（深圳）有限公司 | A kind of lump method for detecting area, device and Medical Image Processing equipment |
| CN109949271B (en) * | 2019-02-14 | 2021-03-16 | 腾讯科技（深圳）有限公司 | Detection method based on medical image, model training method and device |
| CN109949271A (en) * | 2019-02-14 | 2019-06-28 | 腾讯科技（深圳）有限公司 | A kind of detection method based on medical image, the method and device of model training |
| CN110458883A (en) * | 2019-03-07 | 2019-11-15 | 腾讯科技（深圳）有限公司 | A kind of processing system of medical imaging, method, apparatus and equipment |
| CN110033456B (en) * | 2019-03-07 | 2021-07-09 | 腾讯科技（深圳）有限公司 | Medical image processing method, device, equipment and system |
| CN110033456A (en) * | 2019-03-07 | 2019-07-19 | 腾讯科技（深圳）有限公司 | A kind of processing method of medical imaging, device, equipment and system |
| CN110334565A (en) * | 2019-03-21 | 2019-10-15 | 江苏迪赛特医疗科技有限公司 | A kind of uterine neck neoplastic lesions categorizing system of microscope pathological photograph |
| CN110033445A (en) * | 2019-04-10 | 2019-07-19 | 司法鉴定科学研究院 | Medicolegal examination automatic identification system and recognition methods based on deep learning |
| CN110136113B (en) * | 2019-05-14 | 2022-06-07 | 湖南大学 | Vagina pathology image classification method based on convolutional neural network |
| CN110136113A (en) * | 2019-05-14 | 2019-08-16 | 湖南大学 | A kind of vagina pathology image classification method based on convolutional neural networks |
| CN110516665A (en) * | 2019-08-23 | 2019-11-29 | 上海眼控科技股份有限公司 | Identify the neural network model construction method and system of image superposition character area |
| CN110706794A (en) * | 2019-09-26 | 2020-01-17 | 中国科学院深圳先进技术研究院 | Medical image processing system and medical image processing method |
| TWI767506B (en) * | 2020-02-26 | 2022-06-11 | 大陸商上海商湯智能科技有限公司 | Image recognition method, training method and equipment of recognition model |
| CN113710166A (en) * | 2020-03-19 | 2021-11-26 | 艾多特公司 | Carotid artery ultrasonic diagnosis system |
| CN111436972A (en) * | 2020-04-13 | 2020-07-24 | 王时灿 | Three-dimensional ultrasonic gynecological disease diagnosis device |
| WO2021114832A1 (en) * | 2020-05-28 | 2021-06-17 | 平安科技（深圳）有限公司 | Sample image data enhancement method, apparatus, electronic device, and storage medium |
| CN111914841A (en) * | 2020-08-07 | 2020-11-10 | 温州医科大学 | CT image processing method and device |
| CN111914841B (en) * | 2020-08-07 | 2023-10-13 | 温州医科大学 | CT image processing method and device |
| WO2021139447A1 (en) * | 2020-09-30 | 2021-07-15 | 平安科技（深圳）有限公司 | Abnormal cervical cell detection apparatus and method |
| CN112435242A (en) * | 2020-11-25 | 2021-03-02 | 江西中科九峰智慧医疗科技有限公司 | Lung image processing method and device, electronic equipment and storage medium |
| CN112884707A (en) * | 2021-01-15 | 2021-06-01 | 复旦大学附属妇产科医院 | Cervical precancerous lesion detection system, equipment and medium based on colposcope |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108510482B (en) | 2020-12-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108510482A (en) | Cervical carcinoma detection method, device, equipment and medium based on gynecatoptron image | |
| CN105894517B (en) | CT image liver segmentation method and system based on feature learning | |
| CN106056595B (en) | Based on the pernicious assistant diagnosis system of depth convolutional neural networks automatic identification Benign Thyroid Nodules | |
| CN106780448B (en) | A kind of pernicious categorizing system of ultrasonic Benign Thyroid Nodules based on transfer learning and Fusion Features | |
| Cao et al. | Fracture detection in x-ray images through stacked random forests feature fusion | |
| Cha et al. | Urinary bladder segmentation in CT urography using deep‐learning convolutional neural network and level sets | |
| Joshi et al. | Classification of brain cancer using artificial neural network | |
| CN108257135A (en) | The assistant diagnosis system of medical image features is understood based on deep learning method | |
| WO2021082691A1 (en) | Segmentation method and apparatus for lesion area of eye oct image, and terminal device | |
| Xu et al. | DeepLN: a framework for automatic lung nodule detection using multi-resolution CT screening images | |
| CN109919928A (en) | Detection method, device and the storage medium of medical image | |
| CN107506770A (en) | Diabetic retinopathy eye-ground photography standard picture generation method | |
| KR102058348B1 (en) | Apparatus and method for classification of angiomyolipoma wihtout visible fat and clear cell renal cell carcinoma in ct images using deep learning and sahpe features | |
| CN109389129A (en) | A kind of image processing method, electronic equipment and storage medium | |
| CN107229952A (en) | The recognition methods of image and device | |
| CN107045721A (en) | One kind extracts pulmonary vascular method and device from chest CT image | |
| Jony et al. | Detection of lung cancer from CT scan images using GLCM and SVM | |
| CN111754453A (en) | Pulmonary tuberculosis detection method and system based on chest radiography image and storage medium | |
| CN112263217B (en) | Improved convolutional neural network-based non-melanoma skin cancer pathological image lesion area detection method | |
| CN106530298A (en) | Three-way-decision-based liver tumor CT image classification method | |
| CN109902682A (en) | A kind of mammary gland x line image detection method based on residual error convolutional neural networks | |
| CN109492547A (en) | A kind of tubercle recognition methods, device and storage medium | |
| Ma et al. | Automated pectoral muscle identification on MLO‐view mammograms: Comparison of deep neural network to conventional computer vision | |
| CN108596174A (en) | A kind of lesion localization method of skin disease image | |
| CN110188767A (en) | Keratonosus image sequence feature extraction and classifying method and device based on deep neural network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20201204 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |