CN108898175A - Area of computer aided model building method based on deep learning gastric cancer pathological section - Google Patents
Area of computer aided model building method based on deep learning gastric cancer pathological section Download PDFInfo
- Publication number
- CN108898175A CN108898175A CN201810665639.XA CN201810665639A CN108898175A CN 108898175 A CN108898175 A CN 108898175A CN 201810665639 A CN201810665639 A CN 201810665639A CN 108898175 A CN108898175 A CN 108898175A
- Authority
- CN
- China
- Prior art keywords
- model
- gastric cancer
- training
- data
- pathological section
- 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
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/03—Recognition of patterns in medical or anatomical images
Abstract
The invention discloses the area of computer aided model building methods based on deep learning gastric cancer pathological section, belong to field of artificial intelligence.This method has used the identification of one 121 layers dense connection convolutional neural networks progress image.Dense block structure in DenseNet gets shallow-layer feature by the high-level partial of network, mitigates over-fitting well.The model number of plies is more simultaneously, can fit increasingly complex, more smooth decision function.Although there are many number of plies, the number of parameters of the model is simultaneously few, has saved resource occupation well.To further avoid over-fitting, using the training mechanism of transfer learning.Model can first carry out pre-training on ImageNet data set, allow model to obtain very strong image characteristics extraction ability, the main optimization of model can preferably concentrate on how extracting the feature of affected areas in formal training, greatly improve the utilization efficiency of data.
Description
Technical field
The invention belongs to field of artificial intelligence, mainly a kind of to be based on deep learning algorithm, and have used attention
The gastric cancer pathological section Computer assisted identification model building method of mechanism reinforcing effect.
Background technique
Gastric cancer is that the fourth-largest common cancer, the death rate are even more to be in second in all cancers in the world.Therefore gastric cancer
It has been increasingly becoming a public health problem of whole people's common concern.And if gastric cancer tumor generate early stage, patient's energy
It is diagnosed, then significant effect will be obtained for the treatment of patients with gastric cancer, greatly reduces the death rate of stomach cancer.Gastric cancer
Reason sectioning image is exactly after obtaining stomach tissue by aspiration biopsy, to histotomy Hematoxylin & Eosin
(Hematoxylin&Eosin, H&E) dyeing, then the image obtained with digital camera by microscope photographing.Gastric cancer pathological section
Important directive function can be provided for the treatment and diagnosis of clinician.However in the analysis of Pathology Doctors ' and diagnosis process
In, doctor needs to judge in conjunction with the clinical diagnosis experience of oneself long-term accumulation whether have canceration in gastric cancer pathological section, this
Artificial diagnostic method, professional knowledge and working experience to doctor have high requirement, and vulnerable to diagosis person's subjectivity
The influence of the factors such as mood and tired diagosis, so the diagnosis process that this height relies on human factor has subjective differences,
The dot fault of doctor may all bring serious consequence to patient.
Computer-aided diagnosis (Computer-Aided Diagnosis, CAD) is current manual's intelligence in medical domain
Important application has become the big research hotspot of current one.In recent years, with the continuous development of big data technology, Ren Menke
Easily to obtain effective medical information, and nowadays computing capability is substantially improved, and computer has pathological image non-
The processing and analysis ability of Chang Qiang great.It is very high that virologist has realized that CAD has the evaluation analysis of pathological tissue image
Robustness and high efficiency, these results can support judgement of the virologist to disease well, and mention for further treatment
More accurate reference is supplied.Therefore, by the powerful calculating ability of modern computer, the gastric cancer pathology an of high robust is constructed
It is feasible for being sliced CAD model, and is of great significance.
Currently, research of the CAD on the field is also very rare.Support vector machines is mainly used in Existing methods
(SupportVector Machine, SVM) classifies.Cosatto et al. is based on area-of-interest (Regions-of-
Interest, ROI) different zones of each gastric cancer pathological section are split, are marked, according to the color manual extraction of slice
Feature is constructed training set, then carries out semi-supervised training using how exemplary mode, classified using SVM to sample.These
Conventional method still possesses certain defect.Firstly, may require that research people when carrying out pathological image segmentation and feature extraction
Member possesses the professional knowledge of related fields, is otherwise difficult to morphological feature, textural characteristics and the correlation in sample slice
Structure is described.Meanwhile pathological image is often sufficiently complex, conventional method is likely difficult to extract the high quality of distinction
Feature causes accuracy of identification more general.
Summary of the invention
The present invention uses artificial intelligence technology, realizes the identification of the gastric cancer pathological section based on deep learning, assists doctor
Clinical diagnosis.Gastric cancer pathological section real data set is collected first, is carried out data mark via specialist, is analyzed data set matter
Amount, and carry out data enhancing and data prediction.Then the convolution that high efficiency extraction goes out gastric cancer pathology sectioning image feature is constructed
Neural network (Convolutional NeuralNetwork, CNN), is finally instructed using established data the set pair analysis model
Practice and test, completes the identification to gastric cancer pathological section.
The problem of present invention mainly faces following points：
1, the gastric cancer pathological image of high quality is very precious, and picture material is extremely complex, and data are few, difficulty is big, in training
It is easy to appear the situations of over-fitting for model in the process；
2, picture size is larger, all in million grades of pixels, it is difficult to calculate, a large amount of meter can be occupied by directly carrying out analysis identification
Calculate resource, it is contemplated that actual use, model must have higher recognition efficiency and lower resource occupation；
3, may have very large stretch of region in a width gastric cancer cases sectioning image is all normal region, this may interfere with mould
Extraction of the type to illness feature；
4, during actually diagnosis, doctor and patient can prefer to have finer detection to illness sample when diagnosis,
To reduce malpractice, it is patient that the loss that patient's mistaken diagnosis is normal person, which is much larger than normal person's mistaken diagnosis, therefore model is right
The gastric cancer pathological section of illness, i.e. positive sample have higher susceptibility.
In view of the above-mentioned problems, present invention uses one 121 layers dense connection convolutional neural networks (Densely
Connected Convolutional Network, DenseNet) carry out image identification.Dense piece in DenseNet
(Dense Block) structure can allow the high-level partial of network to get shallow-layer feature, mitigate over-fitting well.Simultaneously
The model number of plies is more, can fit increasingly complex, more smooth decision function.Although there are many number of plies, the ginseng of the model
Number quantity is simultaneously few, has saved resource occupation well.To further avoid over-fitting, the present invention additionally uses transfer learning
Training mechanism.Model can first carry out pre-training on ImageNet data set, and model is allowed to obtain very strong image characteristics extraction energy
Power, in formal training, the main optimization of model can preferably concentrate on how extracting the feature of affected areas, greatly mention
The utilization efficiency of high data.
The model formidably ability in feature extraction is benefited from, input picture has been compressed to 224 × 224 size, substantially
Model parameter quantity is reduced, accelerates to calculate.Simultaneously because the pre-training on ImageNet data set, stomach cancer slices data set
Data distribution is closer to ImageNet data set, there will be better training result, therefore all data before entering model all
The average value of ImageNet data set has been used to be standardized with variance.Also data have been carried out with random water when training
Flat, vertical overturning and 15 ° of the row data that are rotated into enhance.
In order to further emphasize that information included in affected areas, this method introduce attention mechanism in image
Among DenseNet.Doctor also excessively pays close attention to normal segments when diagnosing to pathological section, but can be by more energy collection
In in the region for having illness suspicion.Therefore the transition zone between first of model, second, dense piece of third
It joined after (Transition Layer) and pay attention to power module (Attention Module).Into the characteristic pattern of the module
(Feature Map) is by the way that continuously up-sampling and down-sampling, that is, context when passing through trained spontaneously allow model twice
The probability that every bit in characteristic pattern represents illness feature is practised, reinforces the illness correlation in characteristic pattern further according to probability size later and believes
Breath.Up-sampling first passes through convolution operation twice with down-sampling twice and pondization operates and increases receptive field, compressive features figure, then
Pass through deconvolution operation twice and upper storage reservoirization operation reduction input size.After first time up-samples, output can pass through
The results added of one great-jump-forward connection (Skip Connection) and first time down-sampling, this can promote network from multiple senses
Information is extracted by open country.Since characteristic pattern can be spliced after the convolutional coding structure by DenseNet with original input, number
It will be expanded once according to every by a convolution, be unfavorable for very much the compression of characteristic pattern.Therefore the present invention changes in paying attention to power module
Convolution is carried out with the residual error structure of ResNet, continues to retain the feature that extracts in floor portions with this.
The present invention has used intersection entropy function in the training process, in order to enhance the susceptibility that model is sliced illness, this
Invention part of positive sample in intersecting entropy function added a weight greater than 1.
In conclusion the area of computer aided model building method based on deep learning gastric cancer pathological section, this method include
Following steps：
Step 1 constructs one 121 layers of DenseNet model, as shown in figure 8, the trunk portion of the model be by 4 by
The dense structure gradually deepened is formed with 4 transition zone alternative splicings.The base of the dense structure of transition layer structure and composition therein
This convolutional coding structure distribution such as Fig. 2 and Fig. 1.It, all can be by each secondary volume before before each convolution operation starts in each dense structure
Long-pending result is spliced in channel direction, realizes the feature transmitting of great-jump-forward.The last layer of model is that a Sigmoid is mono- defeated
Full articulamentum out, the result of output model classification.In addition to the last layer, all layers of parameter is initialized as the model structure
The good parameter of pre-training on ImageNet data set；
Step 2 carries out related pretreatment to gastric cancer pathological section data collection；Specifically include following steps：
The image of every gastric cancer pathological section is compressed to 224 × 224 using the mode of bilinear interpolation by step 2.1
Size；
Step 2.2 carries out standard to gastric cancer pathological section data collection using the average value and variance of ImageNet data set
Change processing；
Gastric cancer pathological section data concentration gastric cancer pathology sectioning image is randomly divided into three groups by step 2.3：Training set,
Verifying collection and test set, preferably, the data in training set account for the 80% of all data, verifying collection and the number in test set
According to accounting for the 80% of all data；
Step 3 is trained using the data the set pair analysis model pre-processed.
Preferably, step 3 specifically includes following steps：
Step 3.1, model training algorithm be adjusted to model be trained using the Adam optimization algorithm of standard, lose letter
Number is set as being added to the intersection entropy function of 1.5 weight of value in positive sample part, and batch size is 10 when training；
When step 3.2, training, before gastric cancer pathology sectioning image enters model, there is 20% probability to will do it following operation
In one of which to carry out data enhancing：Randomly carry out horizontal, vertical overturning and 15 ° of positive hour hands or rotation counterclockwise
Turn, wherein the point beyond boundary can be replaced by white (RGB (255,255,255)), the i.e. background color of gastric cancer pathological section；
Step 3.3, model 2 epoch of training on gastric cancer pathological section data collection make it obtain extraction gastric cancer pathology and cut
The basic capacity of piece feature；
Step 3.4, between first of model, second, dense piece of third, addition pays attention to power module.Continue to instruct
Practice 60 epoch, model is allowed spontaneously will more to focus on diseased part in next training process.In every instruction
After practicing an epoch, model is allowed to predict verifying collection, while the accuracy of record cast prediction.Final choice verifying
Collect the smallest model of penalty values as final result.
Step 3.5 saves optimal models, and uses the accuracy rate of the test set DATA REASONING category of model.
Preferably, the whole training platform of model is based on cloud, Keras frame is built on linux system, after
End uses TensorFlow.Trained GPU is GTX1080, the operation driving for using CUDA to calculate as video card, and is used
CuDNN accelerates deep learning.
Compared with prior art, the present invention has following clear superiority：
1, with the help of deep learning algorithm, this CAD model automatically extracts the feature in gastric cancer pathological section, complete
The constraint of relevant medical professional knowledge is got rid of entirely；
2, the constructed model accuracy of identification out of the present invention is higher, hence it is evident that is higher than existing conventional method；
3, the dependence of data volume is greatly decreased, due to the advantage and transfer learning of DenseNet model structure itself
Using the model can also train high-precision model in the case where data volume is less to avoid over-fitting；
4, model convergence rate is very fast, and resource occupation when training is lower.
5, model parameter is less, and when carrying out specimen discerning, operation is high-efficient, and resource occupation is low, can put into reality well
Among border is practical；
6, model has higher susceptibility to illness slice, reduces a possibility that malpractice occurs.
Detailed description of the invention：
Fig. 1 is the dense convolutional coding structure (Dense Conv Block) in identification model；
Fig. 2 is the transition zone (Transition Layer) in identification model between dense convolutional coding structure；
Fig. 3 is in identification model for substantially compressing the input processing structure (Input Block) of input picture；
Fig. 4 is the residual error convolutional coding structure (Residual Conv Block) in identification model；
Fig. 5 is the down-sampling structure (Down Sample Block) paid attention in power module；
Fig. 6 is the up-sampling structure (Up Sample Block) paid attention in power module；
Fig. 7 is the basic structure for paying attention to power module (Attention Module)；
Fig. 8 adds the structure before attention model for this method identification model；
Fig. 9 is the structure after this method identification model addition attention model.
Specific embodiment
Yi Xiajiehejutishishili,Bing Canzhaofutu,Dui Benfamingjinyibuxiangxishuoming.
Hardware used in the present invention is the work station that can carry out deep learning.Used auxiliary tool is deep
Spend learning training frame Keras.
The computer-aided diagnosis model building method of gastric cancer pathological section provided by the present invention based on deep learning
It mainly includes the following steps that：
Step 1, construct one 121 layers of DenseNet model, as shown in figure 8, the trunk portion of the model be by 4 by
The dense structure gradually deepened is formed with 4 transition zone alternative splicings.The base of the dense structure of transition layer structure and composition therein
This convolutional coding structure distribution such as Fig. 2 and Fig. 1.It, all can be by each secondary volume before before each convolution operation starts in each dense structure
Long-pending result is spliced in channel direction, realizes the feature transmitting of great-jump-forward.The last layer of model is that a Sigmoid is mono- defeated
Full articulamentum out, the result of output model classification.In addition to the last layer, all layers of parameter is initialized as the model structure
The good parameter of pre-training on ImageNet data set
Step 2, related pretreatment is carried out to gastric cancer pathological section data collection.
Step 2.1, the image of every gastric cancer pathological section is compressed to 224 × 224 using the mode of bilinear interpolation
Size.
Step 2.2, standard is carried out to gastric cancer pathological section data collection using the average value of ImageNet data set and variance
Change processing, i.e. the original pixel value x for i-th point in imageiHave：
Wherein μ and σ2Respectively represent the mean value and variance of ImageNet data set.
Step 2.3, gastric cancer pathological section data concentration gastric cancer pathology sectioning image is randomly divided into three groups：Training set,
Verifying collection and test set.Data in training set account for the 80% of all data, and verifying collection accounts for all with the data in test set
The 80% of data.
Step 2.4, before training set data enters model, randomly progress horizontal, vertical overturning and 15 ° are being had just
Hour hands or counterclockwise rotation, wherein the point beyond boundary can be replaced by white (RGB (255,255,255)), i.e. gastric cancer pathology
The background color of slice.
Step 3, it is trained using the data the set pair analysis model handled well.
Step 3.1, model training algorithm be adjusted to model be trained using the Adam optimization algorithm of standard,
Batchsize is set as 10, and loss function is set as being added to the intersection entropy function of 1.5 weight of value, final friendship in positive sample part
Pitching entropy function is：
Loss (y, y ')=wy-log (y ')+(1-y)-log (1-y '),
Wherein y is desired output, and y ' is reality output, and w is weight and value is 1.5.
When step 3.2, training, before gastric cancer pathology sectioning image enters model, there is 20% probability to will do it following operation
In one of which to carry out data enhancing：Randomly carry out horizontal, vertical overturning and 15 ° of positive hour hands or rotation counterclockwise
Turn, wherein the point beyond boundary can be replaced by white (RGB (255,255,255)), the i.e. background color of gastric cancer pathological section；
Step 3.3, model 2 epoch of training on gastric cancer pathological section data collection make it obtain extraction gastric cancer pathology and cut
The basic capacity of piece feature.
Step 3.4, as shown in figure 8, adding attention mould between first of model, second, dense piece of third
Block, shown in part as dashed lines.Notice that the structure of power module is as shown in Figure 7.Continue 60 epoch of training, model is allowed to connect
It spontaneously will more focus on diseased part in the training process got off.After one epoch of every training, model is allowed
Verifying collection is predicted, while the accuracy of record cast prediction.The final choice verifying collection the smallest model conduct of penalty values
Final result.
Step 3.5 saves optimal models, and uses the accuracy rate of the test set DATA REASONING category of model.
Above embodiments are only exemplary embodiment of the present invention, are not used in the limitation present invention, protection scope of the present invention
It is defined by the claims.Those skilled in the art can within the spirit and scope of the present invention make respectively the present invention
Kind modification or equivalent replacement, this modification or equivalent replacement also should be regarded as being within the scope of the present invention.
Claims (3)
1. the area of computer aided model building method based on deep learning gastric cancer pathological section, it is characterised in that：This method includes
Following steps：
Step 1, the DenseNet model for constructing one 121 layers, the trunk portion of the model is the dense knot gradually deepened by 4
Structure is formed with 4 transition zone alternative splicings；It, all can be by before each time before each convolution operation starts in each dense structure
The result of convolution is spliced in channel direction, realizes the feature transmitting of great-jump-forward；The last layer of model is that a Sigmoid is mono-
The full articulamentum of output, the result of output model classification；In addition to the last layer, all layers of parameter is initialized as the model knot
Structure good parameter of pre-training on ImageNet data set；
Step 2 carries out related pretreatment to gastric cancer pathological section data collection；Specifically include following steps：
Step 2.1, the size that the image of every gastric cancer pathological section is compressed to using the mode of bilinear interpolation to 224 × 224；
Step 2.2 is standardized place to gastric cancer pathological section data collection using the average value and variance of ImageNet data set
Reason；
Gastric cancer pathological section data concentration gastric cancer pathology sectioning image is randomly divided into three groups by step 2.3：Training set, verifying
Collection and test set, the data in training set account for the 80% of all data, and verifying collection accounts for all data with the data in test set
80%；
Step 3 is trained using the data the set pair analysis model pre-processed；
Step 3 specifically includes following steps：
Step 3.1, model training algorithm be adjusted to model be trained using the Adam optimization algorithm of standard, loss function is set
For the intersection entropy function for being added to 1.5 weight of value in positive sample part, batch size is 10 when training；
When step 3.2, training, before gastric cancer pathology sectioning image enters model, there is 20% probability to will do it in following operation
One of which is to carry out data enhancing：Horizontal, vertical overturning and 15 ° of positive hour hands or rotation counterclockwise are randomly carried out,
In beyond the point on boundary can be replaced by white, the i.e. background color of gastric cancer pathological section；
Step 3.3, model 2 epoch of training on gastric cancer pathological section data collection obtain it and extract gastric cancer pathological section spy
The basic capacity of sign；
Step 3.4, between first of model, second, dense piece of third, addition pays attention to power module；Continue training 60
A epoch allows model spontaneously will more to focus on diseased part in next training process；In every training one
After a epoch, model is allowed to predict verifying collection, while the accuracy of record cast prediction；Final choice verifying collection damage
Mistake is worth the smallest model as final result；
Step 3.5 saves optimal models, and uses the accuracy rate of the test set DATA REASONING category of model.
2. the area of computer aided model building method according to claim 1 based on deep learning gastric cancer pathological section,
It is characterized in that：The whole training platform of model is based on cloud, and Keras frame is built on linux system, and rear end uses
TensorFlow；Trained GPU is GTX1080, the operation driving for using CUDA to calculate as video card, and using cuDNN to depth
Degree study is accelerated.
3. the area of computer aided model building method according to claim 1 based on deep learning gastric cancer pathological section,
It is characterized in that：Step 1, one 121 layers of DenseNet model is constructed, the trunk portion of the model is gradually deepened by 4
Dense structure is formed with 4 transition zone alternative splicings；In transition layer structure and the basic convolutional coding structure of the dense structure of composition, often
It, all can be real by the result of convolution is spliced in channel direction each time before before each convolution operation starts in a dense structure
The feature transmitting of existing great-jump-forward；The last layer of model is the mono- full articulamentum exported of a Sigmoid, output model classification
As a result；In addition to the last layer, it is good that all layers of parameter is initialized as model structure pre-training on ImageNet data set
Parameter
Step 2, related pretreatment is carried out to gastric cancer pathological section data collection；
Step 2.1, the image of every gastric cancer pathological section is compressed to 224 × 224 size using the mode of bilinear interpolation；
Step 2.2, place is standardized to gastric cancer pathological section data collection using the average value of ImageNet data set and variance
Reason, i.e. the original pixel value x for i-th point in imageiHave：
Wherein μ and σ2Respectively represent the mean value and variance of ImageNet data set；
Step 2.3, gastric cancer pathological section data concentration gastric cancer pathology sectioning image is randomly divided into three groups：Training set, verifying
Collection and test set；Data in training set account for the 80% of all data, and verifying collection accounts for all data with the data in test set
80%；
Step 2.4, before training set data enters model, horizontal randomly progress, vertical overturning and 15 ° of positive hour hands are had
Or rotation counterclockwise, wherein the point beyond boundary can be replaced by white, the i.e. background color of gastric cancer pathological section；
Step 3, it is trained using the data the set pair analysis model handled well；
Step 3.1, model training algorithm be adjusted to model be trained using the Adam optimization algorithm of standard, batch size
10 are set as, loss function is set as being added to the intersection entropy function of 1.5 weight of value, final intersection entropy function in positive sample part
For：
Loss (y, y ')=wy-log (y ')+(1-y)-log (1-y '),
Wherein y is desired output, and y ' is reality output, and w is weight and value is 1.5；
When step 3.2, training, before gastric cancer pathology sectioning image enters model, there is 20% probability to will do it in following operation
One of which is to carry out data enhancing：Horizontal, vertical overturning and 15 ° of positive hour hands or rotation counterclockwise are randomly carried out,
In beyond the point on boundary can be replaced by white, the i.e. background color of gastric cancer pathological section；
Step 3.3, model 2 epoch of training on gastric cancer pathological section data collection obtain it and extract gastric cancer pathological section spy
The basic capacity of sign；
Step 3.4, between first of model, second, dense piece of third, addition pays attention to power module, as dashed lines
Shown in part；Continue 60 epoch of training, model is allowed spontaneously will more to focus in next training process
Diseased part；After one epoch of every training, model is allowed to predict verifying collection, while record cast prediction is correct
Rate；The final choice verifying collection the smallest model of penalty values is as final result；
Step 3.5 saves optimal models, and uses the accuracy rate of the test set DATA REASONING category of model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810665639.XA CN108898175B (en) | 2018-06-26 | 2018-06-26 | Computer-aided model construction method based on deep learning gastric cancer pathological section |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810665639.XA CN108898175B (en) | 2018-06-26 | 2018-06-26 | Computer-aided model construction method based on deep learning gastric cancer pathological section |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108898175A true CN108898175A (en) | 2018-11-27 |
| CN108898175B CN108898175B (en) | 2020-11-20 |
Family
ID=64346080
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810665639.XA Active CN108898175B (en) | 2018-06-26 | 2018-06-26 | Computer-aided model construction method based on deep learning gastric cancer pathological section |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108898175B (en) |
Cited By (20)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109685141A (en) * | 2018-12-25 | 2019-04-26 | 哈工大机器人（合肥）国际创新研究院 | A kind of robotic article sorting visible detection method based on deep neural network |
| CN109727238A (en) * | 2018-12-27 | 2019-05-07 | 贵阳朗玛信息技术股份有限公司 | The recognition methods of x-ray chest radiograph and device |
| CN109784347A (en) * | 2018-12-17 | 2019-05-21 | 西北工业大学 | Image classification method based on multiple dimensioned dense convolutional neural networks and spectrum attention mechanism |
| CN110378463A (en) * | 2019-07-15 | 2019-10-25 | 北京智能工场科技有限公司 | A kind of artificial intelligence model standardized training platform and automated system |
| CN110472676A (en) * | 2019-08-05 | 2019-11-19 | 首都医科大学附属北京朝阳医院 | Stomach morning cancerous tissue image classification system based on deep neural network |
| CN110633385A (en) * | 2019-09-24 | 2019-12-31 | 天津医科大学 | Retrieval and compression method of medical image |
| CN110706793A (en) * | 2019-09-25 | 2020-01-17 | 天津大学 | Attention mechanism-based thyroid nodule semi-supervised segmentation method |
| CN110738231A (en) * | 2019-07-25 | 2020-01-31 | 太原理工大学 | Method for classifying mammary gland X-ray images by improving S-DNet neural network model |
| CN110766670A (en) * | 2019-10-18 | 2020-02-07 | 厦门粉红思黛医学科技有限公司 | Mammary gland molybdenum target image tumor localization algorithm based on deep convolutional neural network |
| CN111126175A (en) * | 2019-12-05 | 2020-05-08 | 厦门大象东方科技有限公司 | Facial image recognition algorithm based on deep convolutional neural network |
| CN111680553A (en) * | 2020-04-29 | 2020-09-18 | 北京联合大学 | Pathological image identification method and system based on depth separable convolution |
| CN112488234A (en) * | 2020-12-10 | 2021-03-12 | 武汉大学 | End-to-end histopathology image classification method based on attention pooling |
| CN112767329A (en) * | 2021-01-08 | 2021-05-07 | 北京安德医智科技有限公司 | Image processing method and device and electronic equipment |
| CN113011306A (en) * | 2021-03-15 | 2021-06-22 | 中南大学 | Method, system and medium for automatic identification of bone marrow cell images in continuous maturation stage |
| CN113109327A (en) * | 2021-03-09 | 2021-07-13 | 杭州市林业科学研究院 | Prediction method of dry rot of hickory nut |
| CN113112459A (en) * | 2021-03-26 | 2021-07-13 | 四川大学 | Oral squamous cell carcinoma differentiation degree prediction system |
| CN113192633A (en) * | 2021-05-24 | 2021-07-30 | 山西大学 | Stomach cancer fine-grained classification method based on attention mechanism |
| CN113628754A (en) * | 2021-08-12 | 2021-11-09 | 武剑 | Cerebrovascular disease dynamic prediction model construction method and system based on artificial intelligence |
| CN113947607A (en) * | 2021-09-29 | 2022-01-18 | 电子科技大学 | Cancer pathology image survival prognosis model construction method based on deep learning |
| CN114862770A (en) * | 2022-04-18 | 2022-08-05 | 华南理工大学 | Gastric cancer pathological section image segmentation prediction method based on SEnet |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20060285743A1 (en) * | 2005-06-20 | 2006-12-21 | Shih-Jong J. Lee | Object based boundary refinement method |
| CN104834943A (en) * | 2015-05-25 | 2015-08-12 | 电子科技大学 | Brain tumor classification method based on deep learning |
| CN108021916A (en) * | 2017-12-31 | 2018-05-11 | 南京航空航天大学 | Deep learning diabetic retinopathy sorting technique based on notice mechanism |
| CN108038519A (en) * | 2018-01-30 | 2018-05-15 | 浙江大学 | A kind of uterine neck image processing method and device based on dense feature pyramid network |
| CN108108757A (en) * | 2017-12-18 | 2018-06-01 | 深圳市唯特视科技有限公司 | A kind of diabetic foot ulcers sorting technique based on convolutional neural networks |
-
2018
- 2018-06-26 CN CN201810665639.XA patent/CN108898175B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20060285743A1 (en) * | 2005-06-20 | 2006-12-21 | Shih-Jong J. Lee | Object based boundary refinement method |
| CN104834943A (en) * | 2015-05-25 | 2015-08-12 | 电子科技大学 | Brain tumor classification method based on deep learning |
| CN108108757A (en) * | 2017-12-18 | 2018-06-01 | 深圳市唯特视科技有限公司 | A kind of diabetic foot ulcers sorting technique based on convolutional neural networks |
| CN108021916A (en) * | 2017-12-31 | 2018-05-11 | 南京航空航天大学 | Deep learning diabetic retinopathy sorting technique based on notice mechanism |
| CN108038519A (en) * | 2018-01-30 | 2018-05-15 | 浙江大学 | A kind of uterine neck image processing method and device based on dense feature pyramid network |
Non-Patent Citations (3)
| Title |
|---|
| YUEXIANG LI: "Deep Learning Based Gastric Cancer Indentification", 《2018 IEEE 15TH INTERNATIONAL SYMPOSIUM ON BIOMEDICAL IMAGING》 * |
| 于观贞: "人工智能在肿瘤病理诊断和评估中的应用与思考", 《第二军医大学学报》 * |
| 黄奕晖,冯前进: "基于三维全卷积DenseNet的脑胶质瘤MRI分割", 《南方医科大学学报》 * |
Cited By (31)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109784347A (en) * | 2018-12-17 | 2019-05-21 | 西北工业大学 | Image classification method based on multiple dimensioned dense convolutional neural networks and spectrum attention mechanism |
| CN109784347B (en) * | 2018-12-17 | 2022-04-26 | 西北工业大学 | Image classification method based on multi-scale dense convolution neural network and spectral attention mechanism |
| CN109685141A (en) * | 2018-12-25 | 2019-04-26 | 哈工大机器人（合肥）国际创新研究院 | A kind of robotic article sorting visible detection method based on deep neural network |
| CN109685141B (en) * | 2018-12-25 | 2022-10-04 | 合肥哈工慧拣智能科技有限公司 | Robot article sorting visual detection method based on deep neural network |
| CN109727238A (en) * | 2018-12-27 | 2019-05-07 | 贵阳朗玛信息技术股份有限公司 | The recognition methods of x-ray chest radiograph and device |
| CN110378463A (en) * | 2019-07-15 | 2019-10-25 | 北京智能工场科技有限公司 | A kind of artificial intelligence model standardized training platform and automated system |
| CN110378463B (en) * | 2019-07-15 | 2021-05-14 | 北京智能工场科技有限公司 | Artificial intelligence model standardization training platform and automatic system |
| CN110738231B (en) * | 2019-07-25 | 2022-12-27 | 太原理工大学 | Method for classifying mammary gland X-ray images by improving S-DNet neural network model |
| CN110738231A (en) * | 2019-07-25 | 2020-01-31 | 太原理工大学 | Method for classifying mammary gland X-ray images by improving S-DNet neural network model |
| CN110472676A (en) * | 2019-08-05 | 2019-11-19 | 首都医科大学附属北京朝阳医院 | Stomach morning cancerous tissue image classification system based on deep neural network |
| CN110633385A (en) * | 2019-09-24 | 2019-12-31 | 天津医科大学 | Retrieval and compression method of medical image |
| CN110633385B (en) * | 2019-09-24 | 2023-05-12 | 天津医科大学 | Medical image retrieval and compression method |
| CN110706793A (en) * | 2019-09-25 | 2020-01-17 | 天津大学 | Attention mechanism-based thyroid nodule semi-supervised segmentation method |
| CN110766670A (en) * | 2019-10-18 | 2020-02-07 | 厦门粉红思黛医学科技有限公司 | Mammary gland molybdenum target image tumor localization algorithm based on deep convolutional neural network |
| CN111126175A (en) * | 2019-12-05 | 2020-05-08 | 厦门大象东方科技有限公司 | Facial image recognition algorithm based on deep convolutional neural network |
| CN111680553A (en) * | 2020-04-29 | 2020-09-18 | 北京联合大学 | Pathological image identification method and system based on depth separable convolution |
| CN112488234A (en) * | 2020-12-10 | 2021-03-12 | 武汉大学 | End-to-end histopathology image classification method based on attention pooling |
| CN112488234B (en) * | 2020-12-10 | 2022-04-29 | 武汉大学 | End-to-end histopathology image classification method based on attention pooling |
| CN112767329B (en) * | 2021-01-08 | 2021-09-10 | 北京安德医智科技有限公司 | Image processing method and device and electronic equipment |
| CN112767329A (en) * | 2021-01-08 | 2021-05-07 | 北京安德医智科技有限公司 | Image processing method and device and electronic equipment |
| CN113109327B (en) * | 2021-03-09 | 2023-11-17 | 杭州市林业科学研究院 | Method for predicting dry rot of hickory |
| CN113109327A (en) * | 2021-03-09 | 2021-07-13 | 杭州市林业科学研究院 | Prediction method of dry rot of hickory nut |
| CN113011306A (en) * | 2021-03-15 | 2021-06-22 | 中南大学 | Method, system and medium for automatic identification of bone marrow cell images in continuous maturation stage |
| CN113112459A (en) * | 2021-03-26 | 2021-07-13 | 四川大学 | Oral squamous cell carcinoma differentiation degree prediction system |
| CN113192633B (en) * | 2021-05-24 | 2022-05-31 | 山西大学 | Stomach cancer fine-grained classification method based on attention mechanism |
| CN113192633A (en) * | 2021-05-24 | 2021-07-30 | 山西大学 | Stomach cancer fine-grained classification method based on attention mechanism |
| CN113628754B (en) * | 2021-08-12 | 2022-04-08 | 武剑 | Cerebrovascular disease dynamic prediction model construction method and system based on artificial intelligence |
| CN113628754A (en) * | 2021-08-12 | 2021-11-09 | 武剑 | Cerebrovascular disease dynamic prediction model construction method and system based on artificial intelligence |
| CN113947607A (en) * | 2021-09-29 | 2022-01-18 | 电子科技大学 | Cancer pathology image survival prognosis model construction method based on deep learning |
| CN113947607B (en) * | 2021-09-29 | 2023-04-28 | 电子科技大学 | Cancer pathological image survival prognosis model construction method based on deep learning |
| CN114862770A (en) * | 2022-04-18 | 2022-08-05 | 华南理工大学 | Gastric cancer pathological section image segmentation prediction method based on SEnet |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108898175B (en) | 2020-11-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108898175A (en) | Area of computer aided model building method based on deep learning gastric cancer pathological section | |
| Liu et al. | A cascaded deep convolutional neural network for joint segmentation and genotype prediction of brainstem gliomas | |
| Gecer et al. | Detection and classification of cancer in whole slide breast histopathology images using deep convolutional networks | |
| WO2020151536A1 (en) | Brain image segmentation method, apparatus, network device and storage medium | |
| CN108364006B (en) | Medical image classification device based on multi-mode deep learning and construction method thereof | |
| Li et al. | Colorectal polyp segmentation using a fully convolutional neural network | |
| Ding et al. | Multi-scale fully convolutional network for gland segmentation using three-class classification | |
| CN109272048B (en) | Pattern recognition method based on deep convolutional neural network | |
| CN104751178B (en) | Lung neoplasm detection means and method based on shape template matching combining classification device | |
| Bai et al. | Liver tumor segmentation based on multi-scale candidate generation and fractal residual network | |
| CN111798425B (en) | Intelligent detection method for mitotic image in gastrointestinal stromal tumor based on deep learning | |
| CN110070540A (en) | Image generating method, device, computer equipment and storage medium | |
| Nofallah et al. | Machine learning techniques for mitoses classification | |
| CN108010013A (en) | A kind of lung CT image pulmonary nodule detection methods | |
| Zhao et al. | A local and global feature disentangled network: toward classification of benign-malignant thyroid nodules from ultrasound image | |
| CN114565761A (en) | Deep learning-based method for segmenting tumor region of renal clear cell carcinoma pathological image | |
| CN114398979A (en) | Ultrasonic image thyroid nodule classification method based on feature decoupling | |
| Liang et al. | Reducing false-positives in lung nodules detection using balanced datasets | |
| CN115546605A (en) | Training method and device based on image labeling and segmentation model | |
| Huang et al. | FRE-Net: Full-region enhanced network for nuclei segmentation in histopathology images | |
| Mehedi et al. | Kidney tumor segmentation and classification using deep neural network on ct images | |
| Banerjee et al. | A CADe system for gliomas in brain MRI using convolutional neural networks | |
| CN112381818B (en) | Medical image identification enhancement method for subclass diseases | |
| Ahmed et al. | COMA-Net: Towards generalized medical image segmentation using complementary attention guided bipolar refinement modules | |
| Han et al. | A deep learning quantification algorithm for HER2 scoring of gastric cancer |
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 |