CN110163102A - A kind of cervical cell image classification recognition methods based on convolutional neural networks - Google Patents
A kind of cervical cell image classification recognition methods based on convolutional neural networks Download PDFInfo
- Publication number
- CN110163102A CN110163102A CN201910310937.1A CN201910310937A CN110163102A CN 110163102 A CN110163102 A CN 110163102A CN 201910310937 A CN201910310937 A CN 201910310937A CN 110163102 A CN110163102 A CN 110163102A
- Authority
- CN
- China
- Prior art keywords
- dense
- cervical cell
- neural networks
- convolutional neural
- network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/60—Type of objects
- G06V20/69—Microscopic objects, e.g. biological cells or cellular parts
Abstract
The cervical cell image classification recognition methods based on convolutional neural networks that the invention discloses a kind of.Prepare cervical cell image first as training sample, construct convolutional neural networks later, then training sample is inputted in the convolutional neural networks model and is trained, deconditioning saves network weight parameter after iterating to certain number.Target image is partitioned into the weight parameter and network structure obtained with nucleolate region to be predicted, later load training, region to be predicted input, which is wherein calculated, can be obtained classification results when use.It improves the accuracy and efficiency of cervical cell diagnosis, and many and diverse, time-consuming diagnostic process is optimized, the work load of doctor is greatly reduced.
Description
Technical field
A kind of cervical cell image classification recognition methods based on convolutional neural networks belongs to cell image processing technique neck
Domain, in particular to image classification and convolutional neural networks technology.
Background technique
Cervical exfoliated cell checks the primary dcreening operation means as gynemetrics's cervical lesions, has been widely used in clinical work
In work.In practical diagnosis and treatment process, pathologist needs to be visually inspected ten hundreds of cells under the microscope.And every disease
Reason doctor needs to handle daily the sample of a large amount of sufferers, can usually generate diagosis fatigue, and mistaken diagnosis phenomenon happens occasionally.Therefore, it needs
A kind of efficient and quantitative cervical cell diagnostic method is wanted, the diagosis burden of pathologist is mitigated, while improving cervical cell knowledge
Other accuracy rate.
Convolutional neural networks (Convolutional Neural Networks, CNN) are a kind of comprising convolution operation,
It is specifically used to handle the feed-forward type deep neural network with the data of similar network.Convolutional layer included in network and
Pond layer is the nucleus module for realizing convolutional neural networks feature extraction functions.If the input of convolution is the two dimension comprising dry passage
It is carried out convolution operation with several convolution kernels and is exported after being biased by characteristic pattern.Pass through existing cervical cell figure
As data, optimize the objective function of given neural network, the parameter in learning network, so that between network output and true tag
Error it is minimum, to obtain to input the convolutional neural networks of cervical cell image precise classification.
Summary of the invention
(1) technical problems to be solved
The purpose of the present invention is in view of the deficiencies of the prior art, propose a kind of cervical cell figure based on convolutional neural networks
As classifying identification method.The present invention strengthens different convolution using dense convolutional neural networks, dense convolutional neural networks
Connection between layer, improves feature reuse, gradient disappearance and overfitting problem has been effectively relieved, while substantially reducing nerve
The parameter of network improves network performance and result accuracy.We change for this problem of cervical cell image classification
Into improving cervical cell rate of correct diagnosis and efficiency, improve the diagnostic process of pathologist.
(2) technical solution
Cervical cell image classification recognition methods based on convolutional neural networks has following committed step:
Step 1: the cervical cell image for preparing largely to have marked is as training sample.All images are divided into normal table
Confluent monolayer cells, normal indsole confluent monolayer cells, granulocyte, gland cell, atypical squamous cell, foraging by burrowing, high karyoplasmic ratio cell, lymph
Cell, packed cell, monocyte and a kind of sample of rubbish ten.
Step 2: building dense convolutional neural networks.Dense convolutional neural networks are mainly by two important features --- and it is thick
Close link block (Dense Block) and transition zone (Transition Layer).Dense link block is to realize that network is dense
The structure of connection；Transition zone is made of the operation of convolution sum pondization, is the module realized dimensionality reduction and reduce redundancy.With markup information
Triple channel cervical cell image be input to after network by a convolution operation, several dense link blocks and several transition zones
Afterwards, classification finally is exported by Softmax classifier.
Step 3: building cervical cell Image Classifier.By ten a kind of cervical cell image inputs with markup information
The dense convolutional neural networks that step 2 is put up are trained.Objective function is continued to optimize by back-propagation algorithm, is adjusted
The parameter of network obtains the classifier that can identify ten a kind of cervical cell images.
Step 4: cervical cell type in prediction target image.Target image is partitioned into nucleolate to be predicted
Region, the weight parameter and network structure that load training obtains, region to be predicted input, which is wherein calculated, can be obtained classification knot
Fruit.
(3) beneficial effect
The present invention uses dense convolutional neural networks structure, in conjunction with a large amount of mark cervical cell image datas, constructs one
A classifier that can classify to ten a kind of cervical cell images.Cervical cell image is divided using the method for the present invention
Class improves the efficiency of pathological diagnosis, reduces misdiagnosis rate, and the method for the present invention classification speed is fast, is easy in engineering practice,
With wide application value and market prospects.
Detailed description of the invention
Fig. 1 is work flow diagram of the invention.
Fig. 2 is cervical cell image training sample of the invention.
Specific embodiment
Technical solution for a better understanding of the present invention, with reference to the accompanying drawing and this hair is discussed in detail in specific embodiment
It is bright.
The present invention is a kind of cervical cell image classification recognition methods based on convolutional neural networks, and this method mainly includes
Following steps:
Step 1: the cervical cell image for preparing largely to have marked is as training sample.All images are divided into normal surface layer
Cell, normal indsole confluent monolayer cells, granulocyte, gland cell, atypical squamous cell, foraging by burrowing, high karyoplasmic ratio cell, lymph are thin
Born of the same parents, packed cell, monocyte and a kind of sample of rubbish ten.
Step 2: building dense convolutional neural networks.Dense convolutional neural networks are mainly by two important features --- and it is dense
Link block (Dense Block) and transition zone (Transition Layer).Dense link block is to realize the dense company of network
The structure connect；Transition zone is made of the operation of convolution sum pondization, is the module realized dimensionality reduction and reduce redundancy.With markup information
Triple channel cervical cell image is input to after network by a convolution operation, several dense link blocks and several transition zones
Afterwards, classification is exported by Softmax classifier.
Step 3: building cervical cell Image Classifier.By ten a kind of cervical cell image input steps with markup information
The rapid two dense convolutional neural networks put up are trained.Objective function is continued to optimize by back-propagation algorithm, adjusts net
The parameter of network obtains the classifier that can identify ten a kind of cervical cell images.
Step 4: cervical cell type in prediction target image.Target image is partitioned into nucleolate area to be predicted
Domain, the weight parameter and network structure that load training obtains, region to be predicted input, which is wherein calculated, can be obtained classification results.
Specific implementation flow of the invention is as shown in Figure 1, each section specific implementation details are as follows:
1. preparing sample
Dye simultaneously sealing film-making to cervical cell using Papanicolaou's vaginal smear technique, scans pathological section later and imported into meter
In calculation machine, all cervical cell images are divided into normal cells of superficial layer, normal indsole confluent monolayer cells, granulocyte, gland cell, non-
Atypical squamous cells, foraging by burrowing, high karyoplasmic ratio cell, lymphocyte, packed cell, monocyte and a kind of sample of rubbish ten
This.
2. building dense convolutional neural networks
Dense convolutional neural networks are mainly by two important features --- dense link block (Dense Block) and transition
Layer (Transition Layer).Dense link block is to realize the structure of the dense connection of network；In each Dense Block
By the way of dense connection, i.e., each layer of input feature vector is from the output of all layers of front in portion.Every layer has one to answer
Closing operation Hl, it is the combination of three operations: normalization -> ReLU function activation -> convolution algorithm, l layers of output is xl
xl=Hl([x0,x1,...,xl-1])
Transition zone is made of convolution sum pond layer, is the module realized dimensionality reduction and reduce redundancy.Convolutional layer is by several volumes
Product unit composition, the parameter of each convolution unit is optimized by back-propagation algorithm, the mesh of convolution algorithm
Be extract input picture different characteristic.And the purpose of pond layer is to reduce the feature vector of output while improving result.Net
The last of network includes one Softmax layers, and Softmax function can be by a K dimensional vector z containing any real number " compressed " to another
In a K dimension reality vector σ (z) so that the range of each element is between (0,1), and all elements and be 1.
The form of the function is usually provided by following formula:
Each value of this vector indicates the probability for this sample being belonged to each class.
Triple channel cervical cell image with markup information is input to after network by a convolution operation, several dense
After link block and several transition zones, classification is exported by classifier.
3. constructing cervical cell Image Classifier
The dense convolutional neural networks that ten a kind of cervical cell image input steps 2 with markup information are put up into
Row training.Objective function is continued to optimize by back-propagation algorithm, adjusts the parameter of network, back-propagation algorithm is that BP is calculated
Method is a kind of learning algorithm for being suitable for multilayer neural networks, which can find out the parameter of the modification of each layer of weight,
Until error reaches desired value, the classifier that can identify eight class cervical cell images is finally obtained.
4. predicting cervical cell type in target image
Target image is partitioned into the weight parameter obtained with nucleolate region to be predicted, load training and network knot
Structure, region to be predicted input, which is wherein calculated, can be obtained classification results.
Claims (3)
1. a kind of cervical cell image classification recognition methods based on convolutional neural networks has following committed step:
(1) the cervical cell image for preparing largely to have marked is as training sample.All images are divided into normal cells of superficial layer, just
It is normal indsole confluent monolayer cells, granulocyte, gland cell, atypical squamous cell, foraging by burrowing, high karyoplasmic ratio cell, lymphocyte, agglomerating
Cell, monocyte and a kind of sample of rubbish ten；
(2) dense convolutional neural networks are built.Dense convolutional neural networks are mainly by two important features --- dense connection mould
Block (Dense Block) and transition zone (Transition Layer).Dense link block is to realize the knot of the dense connection of network
Structure；Transition zone is made of the operation of convolution sum pondization, is the module realized dimensionality reduction and reduce redundancy.Triple channel with markup information
Cervical cell image is input to after network after a convolution operation, several dense link blocks and several transition zones, is passed through
Softmax classifier exports classification；
(3) cervical cell Image Classifier is constructed.Ten a kind of cervical cell image input steps two with markup information are taken
The dense convolutional neural networks built up are trained.Objective function is continued to optimize by back-propagation algorithm, adjusts the ginseng of network
Number obtains the classifier that can identify ten a kind of cervical cell images；
(4) cervical cell type in target image is predicted.Target image is partitioned into nucleolate region to be predicted, load
The weight parameter and network structure that training obtains, region to be predicted input, which is wherein calculated, can be obtained classification results.
2. a kind of cervical cell image classification recognition methods based on convolutional neural networks according to claim 1, special
Sign is: the dense convolutional neural networks of the step (2) are mainly by two important features --- dense link block (Dense
) and transition zone (Transition Layer) Block.Dense link block is to realize the structure of the dense connection of network；Each
Inside Dense Block by the way of dense connection, i.e., each layer of input feature vector is from the output of all layers of front.
Every layer has a composition operation Hl, it is the combination of three operations: normalization -> ReLU function activation -> convolution algorithm, l layers
Output is
xl=Hl([x0,x1,...,xl-1]) (1)
Transition zone is made of convolution sum pond layer, is the module realized dimensionality reduction and reduce redundancy.Convolutional layer is by several convolution lists
Member composition, the parameter of each convolution unit is optimized by back-propagation algorithm, and the purpose of convolution algorithm is
Extract the different characteristic of input picture.And the purpose of pond layer is to reduce the feature vector of output while improving result.Network
It finally include one Softmax layers, Softmax function can be by a K dimensional vector z containing any real number " compressed " to another K
Tie up in reality vector σ (z) so that the range of each element is between (0,1), and all elements and be 1；
The form of the function is usually provided by following formula:
Each value of this vector indicates the probability for this sample being belonged to each class.
Triple channel cervical cell image with markup information is input to after network by a convolution operation, several dense connections
After module and several transition zones, classification is exported by Softmax classifier.
3. a kind of cervical cell image classification recognition methods based on convolutional neural networks according to claim 1, special
Sign is: the dense convolution mind that the step (3) puts up ten a kind of cervical cell image input steps 2 with markup information
It is trained through network.Objective function is continued to optimize by back-propagation algorithm, adjusts the parameter of network, back-propagation algorithm is
It is a kind of learning algorithm for being suitable for multilayer neural networks for BP algorithm, which can find out the modification of each layer of weight
Parameter finally obtain the classifier that can identify ten a kind of cervical cell images until error reaches desired value.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910310937.1A CN110163102A (en) | 2019-04-18 | 2019-04-18 | A kind of cervical cell image classification recognition methods based on convolutional neural networks |
| CN201910812991.6A CN110363188A (en) | 2019-04-18 | 2019-08-30 | Cervical cell image classification method based on convolutional neural networks |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910310937.1A CN110163102A (en) | 2019-04-18 | 2019-04-18 | A kind of cervical cell image classification recognition methods based on convolutional neural networks |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110163102A true CN110163102A (en) | 2019-08-23 |
Family
ID=67639512
Family Applications (2)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910310937.1A Pending CN110163102A (en) | 2019-04-18 | 2019-04-18 | A kind of cervical cell image classification recognition methods based on convolutional neural networks |
| CN201910812991.6A Pending CN110363188A (en) | 2019-04-18 | 2019-08-30 | Cervical cell image classification method based on convolutional neural networks |
Family Applications After (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910812991.6A Pending CN110363188A (en) | 2019-04-18 | 2019-08-30 | Cervical cell image classification method based on convolutional neural networks |
Country Status (1)
| Country | Link |
|---|---|
| CN (2) | CN110163102A (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110796636A (en) * | 2019-09-25 | 2020-02-14 | 中国人民解放军战略支援部队信息工程大学 | CT image bone condition detection method and device based on convolutional neural network |
| CN111462122A (en) * | 2020-03-26 | 2020-07-28 | 中国科学技术大学 | Automatic cervical cell nucleus segmentation method and system |
| CN111611926A (en) * | 2020-05-21 | 2020-09-01 | 重庆现代建筑产业发展研究院 | Method and device for detecting and identifying landmark building |
| CN112950585A (en) * | 2021-03-01 | 2021-06-11 | 中国人民解放军陆军军医大学 | Cervical cancer cell intelligent detection method based on liquid-based thin-layer cell detection technology TCT |
| CN113222044A (en) * | 2021-05-25 | 2021-08-06 | 合肥工业大学 | Cervical fluid-based cell classification method based on ternary attention and scale correlation fusion |
| CN114708589A (en) * | 2022-04-02 | 2022-07-05 | 大连海事大学 | Cervical cell classification method based on deep learning |
| CN116386857B (en) * | 2023-06-07 | 2023-11-10 | 深圳市森盈智能科技有限公司 | Pathological analysis system and method |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111047577B (en) * | 2019-12-12 | 2021-02-26 | 太原理工大学 | Abnormal urine red blood cell classification statistical method and system |
| CN111209972A (en) * | 2020-01-09 | 2020-05-29 | 中国科学院计算技术研究所 | Image classification method and system based on hybrid connectivity deep convolution neural network |
| CN111274903B (en) * | 2020-01-15 | 2022-12-06 | 合肥工业大学 | Cervical cell image classification method based on graph convolution neural network |
| CN111444829B (en) * | 2020-03-25 | 2023-08-18 | 南京航空航天大学 | Fusion reasoning and learning decision classification method for liquid-based cytology examination |
| CN113469119B (en) * | 2021-07-20 | 2022-10-04 | 合肥工业大学 | Cervical cell image classification method based on visual converter and image convolution network |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9984201B2 (en) * | 2015-01-18 | 2018-05-29 | Youhealth Biotech, Limited | Method and system for determining cancer status |
| CN108875624B (en) * | 2018-06-13 | 2022-03-25 | 华南理工大学 | Face detection method based on multi-scale cascade dense connection neural network |
| CN109190441B (en) * | 2018-06-21 | 2022-11-08 | 丁彦青 | Intelligent classification method, diagnostic instrument and storage medium for female genital tract cell pathology |
-
2019
- 2019-04-18 CN CN201910310937.1A patent/CN110163102A/en active Pending
- 2019-08-30 CN CN201910812991.6A patent/CN110363188A/en active Pending
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110796636A (en) * | 2019-09-25 | 2020-02-14 | 中国人民解放军战略支援部队信息工程大学 | CT image bone condition detection method and device based on convolutional neural network |
| CN111462122A (en) * | 2020-03-26 | 2020-07-28 | 中国科学技术大学 | Automatic cervical cell nucleus segmentation method and system |
| CN111611926A (en) * | 2020-05-21 | 2020-09-01 | 重庆现代建筑产业发展研究院 | Method and device for detecting and identifying landmark building |
| CN112950585A (en) * | 2021-03-01 | 2021-06-11 | 中国人民解放军陆军军医大学 | Cervical cancer cell intelligent detection method based on liquid-based thin-layer cell detection technology TCT |
| CN113222044A (en) * | 2021-05-25 | 2021-08-06 | 合肥工业大学 | Cervical fluid-based cell classification method based on ternary attention and scale correlation fusion |
| CN114708589A (en) * | 2022-04-02 | 2022-07-05 | 大连海事大学 | Cervical cell classification method based on deep learning |
| CN116386857B (en) * | 2023-06-07 | 2023-11-10 | 深圳市森盈智能科技有限公司 | Pathological analysis system and method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110363188A (en) | 2019-10-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110163102A (en) | A kind of cervical cell image classification recognition methods based on convolutional neural networks | |
| Zhou et al. | Cgc-net: Cell graph convolutional network for grading of colorectal cancer histology images | |
| Patil et al. | White blood cells image classification using deep learning with canonical correlation analysis | |
| CN110020682B (en) | Attention mechanism relation comparison network model method based on small sample learning | |
| CN107180248A (en) | Strengthen the hyperspectral image classification method of network based on associated losses | |
| CN108304826A (en) | Facial expression recognizing method based on convolutional neural networks | |
| CN108288035A (en) | The human motion recognition method of multichannel image Fusion Features based on deep learning | |
| CN110675411B (en) | Cervical squamous intraepithelial lesion recognition algorithm based on deep learning | |
| CN107742117A (en) | A kind of facial expression recognizing method based on end to end model | |
| CN110197205A (en) | A kind of image-recognizing method of multiple features source residual error network | |
| CN104217214A (en) | Configurable convolutional neural network based red green blue-distance (RGB-D) figure behavior identification method | |
| CN109410194A (en) | A kind of cancer of the esophagus pathology image processing method based on deep learning | |
| CN107330480A (en) | Hand-written character Computer Identification | |
| CN110363253A (en) | A kind of Surfaces of Hot Rolled Strip defect classification method based on convolutional neural networks | |
| CN107392224A (en) | A kind of crop disease recognizer based on triple channel convolutional neural networks | |
| CN111127423B (en) | Rice pest and disease identification method based on CNN-BP neural network algorithm | |
| CN110533024A (en) | Biquadratic pond fine granularity image classification method based on multiple dimensioned ROI feature | |
| CN107944410A (en) | A kind of cross-cutting facial characteristics analytic method based on convolutional neural networks | |
| CN107451565A (en) | A kind of semi-supervised small sample deep learning image model classifying identification method | |
| CN109214298A (en) | A kind of Asia women face value Rating Model method based on depth convolutional network | |
| CN110084136A (en) | Context based on super-pixel CRF model optimizes indoor scene semanteme marking method | |
| CN112766283A (en) | Two-phase flow pattern identification method based on multi-scale convolution network | |
| Yu et al. | A recognition method of soybean leaf diseases based on an improved deep learning model | |
| CN114266717A (en) | Parallel capsule network cervical cancer cell detection method based on Inception module | |
| Menaka et al. | Chromenet: A CNN architecture with comparison of optimizers for classification of human chromosome images |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20190823 |
|
| WD01 | Invention patent application deemed withdrawn after publication |