CN109859159A - A kind of cervical lesions region segmentation method and device based on multi-modal segmentation network - Google Patents
A kind of cervical lesions region segmentation method and device based on multi-modal segmentation network Download PDFInfo
- Publication number
- CN109859159A CN109859159A CN201811469200.6A CN201811469200A CN109859159A CN 109859159 A CN109859159 A CN 109859159A CN 201811469200 A CN201811469200 A CN 201811469200A CN 109859159 A CN109859159 A CN 109859159A
- Authority
- CN
- China
- Prior art keywords
- image
- iodine
- acetic acid
- network
- modal
- 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
Landscapes
- Image Analysis (AREA)
- Image Processing (AREA)
- Apparatus For Radiation Diagnosis (AREA)
Abstract
The invention discloses a kind of cervical lesions region segmentation methods and device based on multi-modal segmentation network, belong to medical image processing technology field, the mode of interconnection is taken to merge the feature of two kinds of images in the characteristic extraction procedure of acetic acid image and iodine image.In order to merge the feature of two kinds of images, the iodine image of the acetic acid characteristics of image of previous convolution block and next convolution block is carried out to the concatenation of channel level, then iodine image branch carries out subsequent feature learning again；Similarly, the acetic acid characteristics of image of the iodine characteristics of image of previous convolution block and next convolution block is subjected to concatenation, then acetic acid image branch is carrying out subsequent feature learning.Such cross-connection system is continued until the 5th convolution block, and the acetic acid image and iodine characteristics of image that export from the 5th convolution block maintain the feature of two kinds of images substantially.Then, the partitioning portion that the feature that acetic acid image branch and iodine image branch learn is respectively enterd to FCN model, is split prediction.
Description
Technical field
The present invention relates to medical image processing technology field, specifically, being related to a kind of based on multi-modal segmentation network
Cervical lesions region segmentation method and device.
Background technique
One of the main reason for cervical carcinoma is threat whole world women life, and the unique cause of disease of the mankind is explicitly pernicious at present
Tumour.Currently, some screening technologies by detection squamous intraepithelial lesion (squamous intraepithelial lesion,
SIL) prevent cervical carcinoma.And squamous intraepithelial lesion is divided into two classes, high-level squamous intraepithelial lesion (High-grade
Squamous intraepithelial lesion, HSIL) and low level squamous intraepithelial lesion (Low-grade
Squamous intraepithelial lesion, HSIL).The further deterioration of HSIL will form cervical carcinoma.Therefore, exist
HSIL needs to take further treatment and prevention measure in clinical application；LSIL only needs constantly to investigate, and takes some slight
Remedy measures.
The screening means of cervical lesions mainly have HPV to check, PAP is checked, digital cervicography (digital
) and vaginoscopy (colposcopy) cervicography.The multi-modal gynecatoptron image that the present invention uses then from
Vaginoscopy.Comprising the concrete steps that after direct exposure uterine neck for vaginoscopy, successively uses 0.9% physiological saline, 3%-
5% acetum, Dobell's solution are applied to cervical surface, examine cervix for examiner by the uterine neck image of shooting
Whether there is the presence of lesion lesion in squama column boundary and columnar epithelium region, and then assess the property and type of lesion, determines disease
The range of change finally instructs the precise location of selection biopsy according to these information, avoids blindness biopsy, improves positive rates of biopsy
With the accuracy rate of diagnosis.
However, the subjective experience of doctor is largely dependent upon for the diagnostic result of vaginoscopy, judgement
The positive rate and accuracy rate of diagnosis of direct relation biopsy whether accurate.The hair analyzed with artificial intelligence technology and medical imaging
The method of exhibition, many machine learning and deep learning is all applied in the auxiliary diagnosis of medical imaging, helps doctor to make more smart
Really diagnose.
Image, semantic segmentation (Semantic Segmentation) is that one of artificial intelligence computer visual field is important
Research direction, task are to complete the segmentation of an object or a specific region or scene.In medical image, semantic segmentation
It is typically used to lesion, organ or cell in segmented image, tissue etc., in order to the subsequent lesion classification of doctor or diagnosis.
CVPR meeting paper Fully Convolutional Networks for Semantic of the Jonathan Long in 2015
It is proposed in Segmentation, using not making in full convolutional neural networks (Fully Convolutional Networks, FCN)
With full connection but convolution and deconvolution (Deconvolution) is used to carry out semantic segmentation task, achieve it is breakthrough at
Fruit becomes one of the main method of semantic segmentation model from this FCN.
In the checking process of gynecatoptron, doctor can compare repeatedly smear acetic acid after and smear iodine solution after epithelium of cervix uteri
Gynecatoptron image after smearing acetic acid (is called acetic acid image, the gynecatoptron image after smearing iodine solution becomes iodine figure by variation
Picture), their common region is found, lesion is identified, to improve the accuracy rate of biopsy.Therefore, the present invention is based on FCN method and palaces
The acetic acid image and iodine image of neck gynecatoptron propose a kind of cervical lesions cutting techniques based on multi-modal segmentation network, auxiliary
Doctor is helped to realize the cervical lesions identification under gynecatoptron.
Summary of the invention
It is an object of the present invention to provide it is a kind of based on it is multi-modal segmentation network cervical lesions region segmentation method and device,
Improve the diagnosis efficiency and accuracy rate of cervical lesions.
To achieve the goals above, the cervical lesions region segmentation method provided by the invention based on multi-modal segmentation network
The following steps are included:
1) the acetic acid image and iodine image for obtaining multiple cervical samples, lesion region to all images and are marked,
Obtain cervical samples training set；
2) the acetic acid image of same group of image in training set is inputted into the first FCN network, iodine image inputs the 2nd FCN net
The structure of network, the first FCN network and the 2nd FCN network is identical, collectively forms multi-modal cervical lesions region segmentation network；
3) the characteristic extraction part of the first FCN network and the 2nd FCN network carry out interconnection, fusion acetic acid image and
The feature of iodine image obtains the characteristic pattern of acetic acid image and the characteristic pattern of iodine image；
4) it is split in the characteristic pattern of the partitioning portion Dichlorodiphenyl Acetate image of the first FCN network, obtains acetic acid image segmentation
As a result；The characteristic pattern of iodine image is split in the partitioning portion of the 2nd FCN network, obtains iodine image segmentation result；
5) multi-modal cervical lesions region segmentation net is calculated according to the segmentation result and its label of acetic acid image and iodine image
The loss of network, and update according to the loss parameter of multi-modal cervical lesions region segmentation network；
6) the acetic acid image of next group of image and iodine image are separately input in the first FCN network and the 2nd FCN network,
Repeat step 4)~6) multi-modal cervical lesions region segmentation network is trained, until convergence, obtains multi-modal cervix disease
Become region segmentation model；
7) the acetic acid image of cervical samples to be detected and iodine image are inputted into multi-modal cervical lesions region segmentation model
In obtain segmentation prognostic chart.
In above-mentioned technical proposal, on the basis of existing technology, pass through the intersection of acetic acid image and the characteristic pattern of iodine image
The feature of acetic acid image and iodine image is merged in connection, makes full use of and learns the potential correlativity of the two, promote lesion characteristics
Study, improve the accuracy rate of segmentation, pass through last segmentation figure and assist diagnosis, improve diagnosis efficiency.
In order to improve the learning rate of model, so that the Generalization Capability of model is more preferable, preferably, further comprising the steps of:
Cervical samples verifying collection is obtained by the method for step 1)；
After multi-modal cervical lesions region segmentation model training is complete, using verifying concentrate same group of image acetic acid image and
Iodine image tests multi-modal cervical lesions region segmentation model, adjusts model learning rate.
Preferably, further comprising the steps of:
Cervical samples test set is obtained by the method for step 1)；
Using the acetic acid image and iodine image of same group of image in test set to trained multi-modal cervical lesions region
Parted pattern is tested, and segmentation accuracy rate, recall rate and mIOU value are calculated.
Preferably, training set, verifying integrates and the ratio data of test set is 7:2:1.
Often check the iodine on the vinegar white region and iodine image of acetic acid image simultaneously in clinical diagnosis due to doctor
Not pigmented section, and acetic acid image and iodine image are compared repeatedly, their common lesions and lesion region are found, to improve biopsy
Accuracy rate.Therefore, there is certain correlativities to a certain extent for acetic acid characteristics of image and iodine characteristics of image, for as far as possible
The practical diagnosis situation for simulating doctor, preferably, step 3) includes:
The n-th convolution block of acetic acid characteristics of image and the 2nd FCN network that the n-th convolution block of first FCN network extracts
The iodine characteristics of image of extraction merges before the N+1 convolution block of the 2nd FCN network carries out convolution operation；
Meanwhile the 2nd FCN network n-th convolution block extract acetic acid characteristics of image and the first FCN network n-th volume
The iodine characteristics of image that block is extracted merges before the N+1 convolution block of the first FCN network carries out convolution operation；
Until completing all convolution operations obtains the characteristic pattern of acetic acid image and the characteristic pattern of iodine image.
Preferably, acetic acid characteristics of image is merged with iodine characteristics of image by the splicing in feature channel.
Preferably, the characteristic extraction part of the first FCN network and the 2nd FCN network is based on VGG-16, including five volumes
Block, each convolution block are equipped with a pond layer later.
Preferably, in step 4), the step of characteristic pattern of Dichlorodiphenyl Acetate image or iodine image is split, includes:
Result after 5th convolution block pond is subjected to twice of up-sampling, later with after Volume Four block pond by
Element result is added；
Result obtained in the previous step is carried out twice of up-sampling and with the result after third convolution block pond by element phase
Add, then carry out twice of up-sampling, obtains segmentation result to the end.
Cervical lesions region segmentation device provided by the invention based on multi-modal segmentation network, comprising: memory, storage
Computer executable instructions and the data for using or producing when executing the computer executable instructions；Processor, with institute
Memory communication connection is stated, and is configured to execute the computer executable instructions of memory storage, computer executable instructions exist
It is performed, realizes the above-mentioned cervical lesions region segmentation method based on multi-modal segmentation network.
It should be noted that " cervical lesions region segmentation method or device " is directed to corresponding in image in the present invention
The segmentation in region.
Compared with prior art, the invention has the benefit that
The present invention proposes that a kind of mode of interconnection is realized in the characteristic extraction part of acetic acid image and iodine image
The fusion of feature, so that network can learn also learn the feature to iodine image to the feature of acetic acid image, the two mutually promotees
Into mutual fusion, the accuracy of the segmentation result of acetic acid image and iodine image is improved.
Detailed description of the invention
Fig. 1 is the multi-modal cervical lesions region segmentation model training flow chart of the embodiment of the present invention；
Fig. 2 is the structural schematic diagram of the FCN network of the embodiment of the present invention；
Fig. 3 is the structural schematic diagram of the multi-modal cervical lesions region segmentation model of the embodiment of the present invention；
Fig. 4 is the flow chart that in the embodiment of the present invention cervical samples to be detected are split with prediction.
Specific embodiment
To make the object, technical solutions and advantages of the present invention clearer, with reference to embodiments and its attached drawing is to this hair
It is bright to be described further.
Embodiment
The cervical lesions region segmentation device based on multi-modal segmentation network of the present embodiment includes: memory, storage meter
Calculation machine executable instruction and the data for using or producing when executing the computer executable instructions；Processor, and it is described
Memory communication connection, and be configured to execute the computer executable instructions of memory storage, computer executable instructions are in quilt
When execution, the step of realization below based on the multi-modal cervical lesions region segmentation method for dividing network:
Step 1: the screening and pretreatment of gynecatoptron image
Vaginoscopy can successively smear physiological saline, 3%~5% acetic acid and Dobell's solution, mistake on uterine neck
Cheng doctor can shoot multiple physiological saline images, acetic acid image and iodine image.It is only simple clear due to smearing physiological saline
Clean effect removes other interfering substances, so that entire uterine neck is completely exposed, epithelium of cervix uteri will not occur specifically to react, therefore,
The present embodiment only with each patient acetic acid image and iodine image.
It is likely present some medical instruments, text, large-area hemorrhage and reflective image in acetic acid and iodine image, is
Retain the preferable image of quality and preferably learn characteristics of image, filters out an acetic acid image and an iodine for each patient
Image；Give the image filtered out to doctor expert's mark, the region of mark has lesion region (HSIL and LSIL), by all samples
Originally it is divided into training set, verifying collection and test set, ratio data 7:2:1.
Step 2: multi-modal cervical lesions region segmentation model training
The input of the multi-modal cervical lesions region segmentation model of the present embodiment be vaginoscopy under acetic acid image and
Iodine image, using the acetic acid image and iodine image in training set when training multi-modal cervical lesions region segmentation model.Whole instruction
Practice process as shown in Figure 1, the acetic acid image of same patient's multiple groups (a generally batch=8) and iodine image are input to
In multi-modal cervical lesions parted pattern, constantly training pattern, so that result phase of the segmentation result as far as possible with doctor's mark
Closely, the prediction result of model and the loss of legitimate reading, gradient anti-pass are generated by loss function, and update model parameter；So
Next group of acetic acid image and iodine image are input in multi-modal cervical lesions region segmentation model afterwards, continue the instruction of above-mentioned steps
Practice；It is finally restrained until model and terminates training.
When training, model can also be tested on verifying collection, i.e., be input to the data that verifying is concentrated trained
In the model crossed, by hyper parameters such as the indexs regularized learning algorithm rates such as penalty values, accuracy rate, recall rate, so that the generalization of model
It can be more preferable；The segmentation result that model finally exports as acetic acid image and iodine image.
The specific structure is shown in FIG. 3 for multi-modal cervical lesions region segmentation model, is based primarily upon the segmentation of the FCN of Fig. 2
The characteristic extraction part distribution of network, FCN network is based on VGG-16, which has 5 convolution blocks, and each convolution block heel one
Pond layer；The rear portion of FCN network is then up-sampling part, and the result after the 5th convolution block pond adopt on twice
Sample is added with the result after Volume Four block pond by element later, then obtained result is carried out twice and is up-sampled, and with
Result after third convolution block pond is added by element, then carries out twice of up-sampling, obtains segmentation prediction result to the end.
The multi-modal cervical lesions region segmentation model that the present embodiment proposes carries out extension extension on FCN network, mainly
The mode of interconnection is introduced in the characteristic extraction part of FCN network to merge the feature of acetic acid and iodine image.The model has
Acetic acid image and iodine image Liang Ge branch, each branch are a FCN networks, intermediate to be melted by interconnection progress feature
It closes, rest part is consistent with single FCN network.
Often check the iodine on the vinegar white region and iodine image of acetic acid image simultaneously in clinical diagnosis due to doctor
Not pigmented section, and acetic acid image and iodine image are compared repeatedly, their common lesions and lesion region are found, to improve biopsy
Accuracy rate.Therefore, there is certain correlativities to a certain extent for acetic acid characteristics of image and iodine characteristics of image, for as far as possible
The practical diagnosis situation of doctor is simulated, the present embodiment takes interconnection in the characteristic extraction procedure of acetic acid image and iodine image
Mode merge the features of two kinds of images.
As shown in figure 3, the first to five convolution block indicates the characteristic extraction part of network, each convolution block indicates pond
Result later；In order to merge the feature of two kinds of images, by the acetic acid characteristics of image of previous convolution block and next convolution block
Iodine image carry out channel level concatenation, then iodine image branch carries out subsequent feature learning again；It similarly, will be previous
The acetic acid characteristics of image of the iodine characteristics of image of a convolution block and next convolution block carries out concatenation, then acetic acid image branch
Carrying out subsequent feature learning.Such cross-connection system is continued until the 5th convolution block, exports from the 5th convolution block
Acetic acid image and iodine characteristics of image maintain the features of two kinds of images substantially.Then, by acetic acid image branch and iodine image point
The feature that branch learns respectively enters the partitioning portion (part is consistent with the up-sampling part in FCN) of FCN model, is divided
Cut prediction.
Step 3: cervical lesions region segmentation prediction
When there is new patient's colposcopy altimetric image (acetic acid image and iodine image in test set), as long as 3%- will be passed through
Image is input in the cervical lesions region segmentation model of step 2 training after 5% acetum, Dobell's solution processing
It is partitioned into the cervical lesions region on acetic acid image and on iodine image, detailed process is shown in Fig. 4.
It should be noted that " cervical lesions region segmentation method or device " is directed to phase in image in the present embodiment
Answer the segmentation in region.
Claims (8)
1. a kind of cervical lesions region segmentation method based on multi-modal segmentation network, which comprises the following steps:
1) the acetic acid image and iodine image for obtaining multiple cervical samples, are marked the lesion region of all images, obtain palace
Neck sample training collection；
2) the acetic acid image of same group of image in training set being inputted into the first FCN network, iodine image inputs the 2nd FCN network, the
The structure of one FCN network and the 2nd FCN network is identical, collectively forms multi-modal cervical lesions region segmentation network；
3) interconnection is carried out in the characteristic extraction part of the first FCN network and the 2nd FCN network, merges acetic acid image and iodine figure
The feature of picture obtains the characteristic pattern of acetic acid image and the characteristic pattern of iodine image；
4) it is split in the characteristic pattern of the partitioning portion Dichlorodiphenyl Acetate image of the first FCN network, obtains acetic acid image segmentation result；
The characteristic pattern of iodine image is split in the partitioning portion of the 2nd FCN network, obtains iodine image segmentation result；
5) multi-modal cervical lesions region segmentation network is calculated according to the segmentation result and its label of acetic acid image and iodine image
It loses, and updates the parameter of multi-modal cervical lesions region segmentation network according to the loss；
6) the acetic acid image of next group of image and iodine image are separately input in the first FCN network and the 2nd FCN network, are repeated
Step 4)~6) multi-modal cervical lesions region segmentation network is trained, until convergence, obtains multi-modal cervical lesions area
Regional partition model；
7) the acetic acid image of cervical samples to be detected and iodine image are inputted in multi-modal cervical lesions region segmentation model and is obtained
To segmentation prognostic chart.
2. cervical lesions region segmentation method according to claim 1, which is characterized in that further comprising the steps of:
Cervical samples verifying collection is obtained by the method for step 1)；
The primary training of the multi-modal every completion of cervical lesions region segmentation network, the acetic acid image of same group of image is concentrated using verifying
Multi-modal cervical lesions region segmentation network is verified with iodine image, adjusts e-learning rate.
3. cervical lesions region segmentation method according to claim 2, which is characterized in that further comprising the steps of:
Cervical samples test set is obtained by the method for step 1)；
Using the acetic acid image and iodine image of same group of image in test set to trained multi-modal cervical lesions region segmentation
Model is tested, and segmentation accuracy rate, recall rate and mIOU value are calculated.
4. cervical lesions region segmentation method according to claim 1, which is characterized in that step 3) includes:
The n-th convolution block of acetic acid characteristics of image and the 2nd FCN network that the n-th convolution block of first FCN network extracts extracts
Iodine characteristics of image the 2nd FCN network the N+1 convolution block carry out convolution operation before merge；
Meanwhile the 2nd FCN network n-th convolution block extract acetic acid characteristics of image and the first FCN network n-th convolution block
The iodine characteristics of image of extraction merges before the N+1 convolution block of the first FCN network carries out convolution operation；
Until completing all convolution operations obtains the characteristic pattern of acetic acid image and the characteristic pattern of iodine image.
5. cervical lesions region segmentation method according to claim 4, which is characterized in that by acetic acid characteristics of image and iodine figure
As feature is merged by the splicing in feature channel.
6. cervical lesions region segmentation method according to claim 1, which is characterized in that the first FCN network and the 2nd FCN
The characteristic extraction part of network is based on VGG-16, including five convolution blocks, and a pond layer is equipped with after each convolution block.
7. cervical lesions region segmentation method according to claim 6, which is characterized in that in step 4), Dichlorodiphenyl Acetate image
Or the characteristic pattern of iodine image the step of being split, includes:
By after the 5th convolution block pond result carry out twice up-sampling, later with the result after Volume Four block pond by
Element is added；
Result obtained in the previous step is carried out twice to up-sample and be added with the result after third convolution block pond by element, then
Twice of up-sampling is carried out, segmentation result to the end is obtained.
8. a kind of cervical lesions region segmentation device based on multi-modal segmentation network, comprising: memory, storage computer can be held
The data for row instruction and using or produce when executing the computer executable instructions；Processor is logical with the memory
Letter connection, and be configured to execute the computer executable instructions of memory storage, which is characterized in that the computer is executable to be referred to
It enables when executed, realizes the cervical lesions region segmentation side based on multi-modal segmentation network as described in claim 1~7
Method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811469200.6A CN109859159B (en) | 2018-11-28 | 2018-11-28 | Cervical lesion region segmentation method and device based on multi-mode segmentation network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811469200.6A CN109859159B (en) | 2018-11-28 | 2018-11-28 | Cervical lesion region segmentation method and device based on multi-mode segmentation network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109859159A true CN109859159A (en) | 2019-06-07 |
| CN109859159B CN109859159B (en) | 2020-10-13 |
Family
ID=66890554
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811469200.6A Active CN109859159B (en) | 2018-11-28 | 2018-11-28 | Cervical lesion region segmentation method and device based on multi-mode segmentation network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109859159B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110491479A (en) * | 2019-07-16 | 2019-11-22 | 北京邮电大学 | A kind of construction method of sclerotin status assessment model neural network based |
| CN110826576A (en) * | 2019-10-10 | 2020-02-21 | 浙江大学 | Cervical lesion prediction system based on multi-mode feature level fusion |
| CN112750115A (en) * | 2021-01-15 | 2021-05-04 | 杭州电子科技大学 | Multi-modal cervical carcinoma pre-lesion image recognition method based on graph neural network |
| CN112750115B (en) * | 2021-01-15 | 2024-06-04 | 浙江大学医学院附属邵逸夫医院 | Multi-mode cervical cancer pre-lesion image recognition method based on graph neural network |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103325128A (en) * | 2013-05-16 | 2013-09-25 | 深圳市理邦精密仪器股份有限公司 | Method and device intelligently identifying characteristics of images collected by colposcope |
| CN106097347A (en) * | 2016-06-14 | 2016-11-09 | 福州大学 | A kind of multimodal medical image registration and method for visualizing |
| WO2018076023A1 (en) * | 2016-10-21 | 2018-04-26 | Nantomics, Llc | Digital histopathology and microdissection |
| CN108257129A (en) * | 2018-01-30 | 2018-07-06 | 浙江大学 | The recognition methods of cervical biopsy region aids and device based on multi-modal detection network |
| CN108319977A (en) * | 2018-01-30 | 2018-07-24 | 浙江大学 | Cervical biopsy area recognizing method based on the multi-modal network of channel information and device |
| CN108388841A (en) * | 2018-01-30 | 2018-08-10 | 浙江大学 | Cervical biopsy area recognizing method and device based on multiple features deep neural network |
| US20180247195A1 (en) * | 2017-02-28 | 2018-08-30 | Anixa Diagnostics Corporation | Methods for using artificial neural network analysis on flow cytometry data for cancer diagnosis |
| US20180260951A1 (en) * | 2017-03-08 | 2018-09-13 | Siemens Healthcare Gmbh | Deep Image-to-Image Recurrent Network with Shape Basis for Automatic Vertebra Labeling in Large-Scale 3D CT Volumes |
-
2018
- 2018-11-28 CN CN201811469200.6A patent/CN109859159B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103325128A (en) * | 2013-05-16 | 2013-09-25 | 深圳市理邦精密仪器股份有限公司 | Method and device intelligently identifying characteristics of images collected by colposcope |
| CN106097347A (en) * | 2016-06-14 | 2016-11-09 | 福州大学 | A kind of multimodal medical image registration and method for visualizing |
| WO2018076023A1 (en) * | 2016-10-21 | 2018-04-26 | Nantomics, Llc | Digital histopathology and microdissection |
| US20180247195A1 (en) * | 2017-02-28 | 2018-08-30 | Anixa Diagnostics Corporation | Methods for using artificial neural network analysis on flow cytometry data for cancer diagnosis |
| US20180260951A1 (en) * | 2017-03-08 | 2018-09-13 | Siemens Healthcare Gmbh | Deep Image-to-Image Recurrent Network with Shape Basis for Automatic Vertebra Labeling in Large-Scale 3D CT Volumes |
| CN108257129A (en) * | 2018-01-30 | 2018-07-06 | 浙江大学 | The recognition methods of cervical biopsy region aids and device based on multi-modal detection network |
| CN108319977A (en) * | 2018-01-30 | 2018-07-24 | 浙江大学 | Cervical biopsy area recognizing method based on the multi-modal network of channel information and device |
| CN108388841A (en) * | 2018-01-30 | 2018-08-10 | 浙江大学 | Cervical biopsy area recognizing method and device based on multiple features deep neural network |
Non-Patent Citations (1)
| Title |
|---|
| CANER HAZIRBAS 等: "FuseNet: Incorporating Depth into Semantic Segmentation via Fusion-Based CNN Architecture", 《COMPUTER VISION–ACCV 2016》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110491479A (en) * | 2019-07-16 | 2019-11-22 | 北京邮电大学 | A kind of construction method of sclerotin status assessment model neural network based |
| CN110826576A (en) * | 2019-10-10 | 2020-02-21 | 浙江大学 | Cervical lesion prediction system based on multi-mode feature level fusion |
| CN112750115A (en) * | 2021-01-15 | 2021-05-04 | 杭州电子科技大学 | Multi-modal cervical carcinoma pre-lesion image recognition method based on graph neural network |
| CN112750115B (en) * | 2021-01-15 | 2024-06-04 | 浙江大学医学院附属邵逸夫医院 | Multi-mode cervical cancer pre-lesion image recognition method based on graph neural network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109859159B (en) | 2020-10-13 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109146884A (en) | Endoscopy monitoring method and device | |
| CN109961443A (en) | Liver neoplasm dividing method and device based on the guidance of more phase CT images | |
| CN109886986A (en) | A kind of skin lens image dividing method based on multiple-limb convolutional neural networks | |
| CN108446730A (en) | A kind of CT pulmonary nodule detection methods based on deep learning | |
| CN107564580A (en) | Gastroscope visual aids processing system and method based on integrated study | |
| CN108388841B (en) | Cervical biopsy region identification method and device based on multi-feature deep neural network | |
| CN109903284A (en) | A kind of other method and system of HER2 immunohistochemistry image automatic judging | |
| CN110310289A (en) | Lung tissue's image partition method based on deep learning | |
| CN108319977A (en) | Cervical biopsy area recognizing method based on the multi-modal network of channel information and device | |
| CN108257129A (en) | The recognition methods of cervical biopsy region aids and device based on multi-modal detection network | |
| CN104545792B (en) | The arteriovenous retinal vessel optic disc localization method of eye fundus image | |
| CN109636808A (en) | A kind of lobe of the lung dividing method based on full convolutional neural networks | |
| CN109543719A (en) | Uterine neck atypia lesion diagnostic model and device based on multi-modal attention model | |
| CN108564026A (en) | Network establishing method and system for Thyroid Neoplasms smear image classification | |
| CN110097545A (en) | Eye fundus image generation method based on deep learning | |
| CN104573712A (en) | Arteriovenous retinal blood vessel classification method based on eye fundus image | |
| CN109636805A (en) | A kind of uterine neck image lesion region segmenting device and method based on classification priori | |
| CN108564123A (en) | A kind of Thyroid Neoplasms smear image classification method and its device | |
| CN109859159A (en) | A kind of cervical lesions region segmentation method and device based on multi-modal segmentation network | |
| CN110648344A (en) | Diabetic retinopathy classification device based on local focus characteristics | |
| CN104318565B (en) | Interactive method for retinal vessel segmentation based on bidirectional region growing of constant-gradient distance | |
| CN115546605A (en) | Training method and device based on image labeling and segmentation model | |
| CN113160119A (en) | Diabetic retinopathy image classification method based on deep learning | |
| CN109241963A (en) | Blutpunkte intelligent identification Method in capsule gastroscope image based on Adaboost machine learning | |
| CN115206495A (en) | Renal cancer pathological image analysis method and system based on CoAtNet deep learning and intelligent microscopic 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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |