CN109118485A - Digestive endoscope image classification based on multitask neural network cancer detection system early - Google Patents
Digestive endoscope image classification based on multitask neural network cancer detection system early Download PDFInfo
- Publication number
- CN109118485A CN109118485A CN201810928147.5A CN201810928147A CN109118485A CN 109118485 A CN109118485 A CN 109118485A CN 201810928147 A CN201810928147 A CN 201810928147A CN 109118485 A CN109118485 A CN 109118485A
- Authority
- CN
- China
- Prior art keywords
- image
- detection
- pixel
- branch
- classification
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30028—Colon; Small intestine
Abstract
The invention belongs to medical image Intelligent treatment technical field, specially a kind of digestive endoscope image classification based on multitask neural network cancer detection system early.Present system includes: (1) feature extraction core network;(2) digestive endoscope image classification branch;(3) alimentary canal morning carninomatosis becomes region detection branch;The present invention uses the deep neural network structure of multitask, two multiple convolutional layers of task sharing of classification and detection.Endoscopic image is input in neural network model, by a propagated forward, can obtain detection and classification results simultaneously, and calculation amount is effectively reduced, and promotes classification and detection accuracy.The experimental results showed that endoscopic image accurately can be divided into normal and early two class of cancer and detect irregular lesion region in early cancer image by the present invention, reducing human factor influences, and improves the efficiency of clinical diagnosis.
Description
Technical field
The invention belongs to medical image Intelligent treatment technical fields, and in particular to a kind of digestive endoscope image classification is early
Cancer detection system, more specifically to a kind of digestive endoscope image classification based on multitask neural network, cancer is examined early
Examining system.
Background technique
The cancer of the esophagus is China and the common upper gastrointestinal cancer of developing country.It is total that esophageal cancer in China new cases account for the world
40% or more of case load, morbidity and mortality are apparently higher than world average level.Esophageal precancerous lesion, such as intraepithelial neoplasia (cin)
It is the Etiological of the cancer of the esophagus with Barrett oesophagus.Since superior gastrointestinal endoscope screening is in terms of reducing morbidity and mortality
Significant effect has been proposed as the Main Diagnosis method of screening precancerous lesion.Main oesophagus endoscopic diagnosis tool has white light
Observation, electronics Magnification chromoendoscopy, endoscopic ultrasonography, iodine staining etc..These methods are each advantageous, but some human factors, than
Such as doctors experience is insufficient, fatigue, carelessness, may directly affect the accuracy of diagnosis.
Depth convolutional neural networks (Deep convolutional neural network, DCNN) are a kind of engineerings
Habit technology, it is possible to prevente effectively from human factor, learns how to extract from markd mass data abundant with generation automatically
The visual signature of table.The technology uses backpropagation optimization algorithm, and machine is allowed to update its inner parameter, and study input picture arrives
The mapping relations of label.In recent years, DCNN substantially increases the performance of each task in computer vision.
2012, Krizhevsky et al.[1]DCNN is applied to image recognition for the first time, in ImageNet in 2012[2]Figure
Top-5 lower error rate is won the championship title to 15.3%, causes the upsurge of deep learning in picture classification match. 2015
Year, Simonyan et al.[3]Deepen the network number of plies to 16 and 19 layers, proposes VGG-16 and VGG-19, further reduced
The error rate of ImageNet image classification task.2016, He et al.[4]Using 152 layers of residual error network ResNet, by mistake
Rate is reduced to 3.57%, has been more than human eye.
DCNN not only has brilliant performance in image classification task, in the task of some structurings output, such as object
Detection[5-7], semantic segmentation[8,9]In similarly achieve brilliant effect.If applying DCNN in computer-aided diagnosis
In (computer aided diagnosis, CAD), then doctor can be assisted to make better medical diagnosis, early discovery is early controlled
It treats, improves therapeutic effect.
Existing DCNN method be principally dedicated to detect or identify single task role, few work can be realized simultaneously this two
The task of item tight association.The present invention provides a new multitask network frames end to end, it can sufficiently combine instruction
The characteristics of practicing image, feature abundant is extracted, while realizing the identification and detection of early cancer in endoscopic image.
Summary of the invention
For overcome the deficiencies in the prior art, the purpose of the present invention is to provide a kind of disappearing based on multitask neural network
Change road endoscopic image classification cancer detection system early, excluding human factor influences, and realizes the automatic diagnosis of digestive endoscope image.
The present invention provides the cancer detection system early of the digestive endoscope image classification based on multitask neural network, specific to wrap
It includes:
(1) feature extraction core network;
(2) digestive endoscope image classification branch;
(3) alimentary canal morning carninomatosis becomes region detection branch.
Further, in (1), feature extraction core network is in VGG-16[3]On the basis of construct and obtain, remove VGG-16
Last three full articulamentums after, core network include 13 convolutional layers, by two task sharings, the characteristic pattern of convolutional layer is in pond
Resolution ratio is gradually reduced under the action of changing layer, and convolutional layer is divided into 5 groups, respectively conv1_ by the position where the layer of pond
2,conv2_2,conv3_3,conv4_3,conv5_3.Network parameter is by ImageNet data set[2]The VGG- of upper pre-training
16 model initializations, to make full use of the ability of the profound feature of extraction learnt on ImageNet.
Further, in (2), input picture can be divided into abnormal and normal two class by digestive endoscope image classification branch;
It includes global average pond layer and 3 full articulamentums.Full connection is on the one hand reduced used here as the average pondization operation of the overall situation
The parameter amount of layer, still further aspect make neural network can handle the input picture of arbitrary dimension.Global pool is by conv5_
The two dimensional character figure of 3 the last one convolutional layer is mapped as vector, is input to full articulamentum, and first full articulamentum includes 4096
Neuron, second full articulamentum include 1024 neurons, and the full articulamentum of third includes 1 neuron, is swashed through sigmoid
After function living, the probability that output image is classified as abnormal image is obtained.
Further, in (3), alimentary canal morning carninomatosis becomes region detection branch, is Hou et al.[10]The obvious object of proposition
Network is detected, using lesion region as significant region, each pixel in image is divided into significant pixel or not significant pixel,
That is abnormal pixel or normal pixel, specific structure are as follows:
There are 4 pond layers in core network, adds 1 pond layer again after the conv5_3 of core network, totally 5 ponds
Layer can produce the characteristic pattern of 6 kinds of different scales, the corresponding side output of the characteristic pattern of every kind of scale, the side of different scale
Side output first passes around warp lamination and obtains characteristic pattern identical with input image resolution, then is connected by short by higher level
Characteristic pattern be connected to the characteristic pattern of lower level, allow the characteristic pattern of lower level to optimize the coarse characteristic pattern of higher level, warp
After the activation of sigmoid function, the lesion region testing result figure of 6 sides can produce altogether.It is more acurrate in order to further obtain
Testing result, the lesion region testing result figure of 6 sides, which can be connected, is input to a convolutional layer, through sigmoid
After function activation, the lesion region testing result figure of fusion is obtained.Value in lesion region testing result figure is bigger, indicates to correspond to
Pixel is more significant, and lesion may more occur.
Further, it is contemplated that general training concentrates abnormal image and normal picture quantity unequal, and there are certain samples
This unbalanced problem, loss function used in digestive endoscope image classification branch be Focal Loss [11], branch of classifying
Output image is classified as the Probability p of abnormal image, and therefore, the calculation of loss function is as follows:
Wherein y ∈ { 0,1 } is label, and 1 indicates abnormal, and 0 indicates normal.The form of loss function are as follows:
Loss=- ∑ αy(1-py)2log(py)
Wherein, αyUniformly it is set as 0.25.
Further, alimentary canal morning carninomatosis becomes the resolution ratio and input of the lesion detection result figure of region detection branch output
The resolution ratio of image is identical, and each pixel is divided into the probability of abnormal pixel in the value expression image in testing result figure.Consider
To the particularity of medical image, some sample lesion region occupied areas are larger in training set, some sample lesion region areas
It is smaller to there are serious imbalanced training sets even without lesion, therefore select the Focal Loss of α equilibrium[11]As damage
Lose function, calculation with classification branch used in loss function it is similar, but unlike training sample with pixel be list
Position, and αyIt can change with the quantity of all kinds of training samples, αyForm it is as follows:
Wherein, N indicates the quantity of pixel samples, NyIndicate the quantity of the pixel of y class.
Further, in present system, the training method of network model is as follows:
It is divided into alternately trained and joint training two parts, alternately when training, branch corresponding to fixed one of task
Parameter updates the loss function that remaining parameter in network is used to minimize another task, so that network only focuses on it every time
In a task;When joint training, the parameter of whole network is updated, while minimizing the loss function of two tasks, so that net
Network can learn to how coordinating two tasks.When training, sample includes at least abnormal image 1000 and opens, normal picture 1000
?.
Further, in present system, image classification method is as follows when test:
Image to be classified I is inputted, classification branch output I is the Probability p of abnormal image.Given threshold Tcla, as p > Tcla
When, then it is assumed that lesion has occurred in I, is abnormal image;Otherwise it is assumed that I is normal picture.
Further, in present system, irregularly to become method for detecting area as follows for early carninomatosis when test:
Image to be detected I is inputted, early carninomatosis becomes detection branches and exports lesion region testing result figure identical with I size,
Using the testing result figure M ∈ [0,1] of fusion as final output;Given threshold Tsal, lesion region is M (i, j) > TsalIt is right
The region answered, wherein i, j are the coordinate of pixel.
According to experimental result, in order to keep higher classification accuracy, it is proposed that threshold value TclaValue range 0.4 to 0.6
Between;In order to keep higher detection accuracy, it is proposed that threshold value TsalValue range is between 0.1 to 0.8.
Note that only needing network by a propagated forward after input test image I, classification and detection can be obtained simultaneously
As a result.
The beneficial effects of the present invention are: the present invention devises a multitask network end to end, for realizing simultaneously
The detection of the classification of digestive endoscope image and early cancer lesion region.Image to be tested need to only pass through a propagated forward
Classified and testing result, two task sharing core network parameters, effectively reduce calculation amount, improves diagnosis efficiency.
In addition, the present invention does not need manually to participate in classification and detection process, human factor is reduced, can be examined for doctor
It is disconnected that reference is provided, reduce morbidity and mortality.
Detailed description of the invention
Fig. 1 is network frame figure of the invention.
Fig. 2 is the detailed construction that lesion region detects network.
Fig. 3 is the Receiver operating curve (ROC) of endoscopic image classification.
Fig. 4 is the relational graph of accuracy rate (mIoU) and threshold value T of early cancer lesion detection.
Fig. 5 is the effect picture of the early cancer lesion detection of the present invention.
Specific embodiment
Embodiment of the present invention is described in detail below, but protection scope of the present invention is not limited to the implementation
Example.
Using the network structure in Fig. 1, with 1332 abnormal images, 1096 normal picture training multitask nerve nets
Network obtains automatic classification and detection model.
Specific implementation method is:
(1) before training, with the VGG-16 model initialization network parameter of pre-training, by the Image Adjusting in training set to uniting
One size 300 × 300;
(2) when training, random cropping image subtracts mean value to 224 × 224.If initial learning rate is 0.0001, decaying
Rate 0.9, every two periodic attenuation are primary.With the method for small lot stochastic gradient descent, loss function is minimized.The size criticized is set
It is 12.Over-fitting in order to prevent, the random some neurons for killing full articulamentum in classification branch;
Carry out alternately training first: the parameter of fixed image classification branch minimizes the loss function of lesion detection, allows net
Network learns lesion detection task, after convergence, then the parameter of fixed lesion detection branch, the loss function of image classification is minimized,
Allow e-learning image classification task;
Then it carries out joint training: updating parameter all in network, while minimizing image classification and lesion detection
Loss function, training allow e-learning how to coordinate the relationship between two tasks to restraining;
(3) when testing, image I is resized to 224 × 224, is input in trained model, model output is current
Image is classified as the Probability p of abnormal image and the lesion region testing result figure M of fusion.Set classification thresholds TclaIt is 0.5,
As p > 0.5, then it is assumed that I is abnormal image;Otherwise it is assumed that I is normal picture.Lesion detection threshold value T is setsalIt is 0.5, disease
The region of M (i, j) > 0.5 is in change region, and wherein i, j are the coordinate of pixel.
Fig. 3 is the ROC curve for evaluating classifying quality of the present invention, it can be seen that (AUC, maximum value are the area under ROC curve
1) reached 0.9955, illustrated that classifying quality of the invention is brilliant.
Fig. 4 is the relational graph of Detection accuracy of the present invention and threshold value T, it can be seen that when threshold value is between 0.1 to 0.8, inspection
The mIoU for surveying result is both greater than 0.7, and wherein highest accuracy rate has reached 0.789.
Fig. 5 is the example that the present invention detects that early carninomatosis becomes.Wherein Fig. 5 (a) is original image, in Fig. 5 (b) in black closed curve
Lesion region for true lesion region, in Fig. 5 (c) in blue closed curve for detection;As can be seen that system of the invention
It can accurately detect early carcinomatous change.
Bibliography
[1]Krizhevsky,A.,Sutskever,I.&Hinton,G.E.ImageNet classification with
deep convolutional neural networks.Advances in Neural Information Processing
Systems, 1097-1105(2012).
[2]Russakovsky,O.,Deng,J.,Su,H.et al.ImageNet Large Scale Visual
Recognition Challenge.International Journal of Computer Vision 115,211-252
(2015).
[3]Simonyan,K.&Zisserman A.Very deep convolutional networks for
large-scale image recognition.International Conference on Representation
Learning,(2014).
[4]He,K.,Zhang,X.,Ren,S.&Sun,J.Deep residual learning for image
recognition. IEEE Conference on Computer Vision and Pattern Recognition,770-
778(2016).
[5]Girshick,R.,Donahue,J.,Darrell,T.&Malik,J.Rich feature hierarchies
for accurate object detection and semantic segmentation.IEEE Conference on
Computer Vision and Pattern Recognition,580-587(2014).
[6]Girshick,R.Fast R-CNN.IEEE International Conference on Computer
Vision, 1440-1448(2015).
[7]Ren,S.,He,K.,Girshick,R.&Sun,J.Faster R-CNN:Towards real-time
object detection with region proposal networks.Neural Information Processing
Systems,(2015).
[8]Long,J.,Shelhamer,E.&Darrell,T.Fully convolutional networks for
semantic segmentation.IEEE International Conference on Computer Vision,3431-
3440(2015).
[9]Chen,L.,Papandreou,G.,Kokkinos,I.,Murphy,K.&Yuille,A.L.DeepLab:
Semantic image segmentation with deep convolutional nets,atrous convolution,
and fully connected CRFs.IEEE Transactions on Pattern Analysis and Machine
Intelligence 40,834-848 (2018).
[10]Hou,Q.,Cheng,M.,Hu,X.,Borji,A.,Tu,Z.&Torr,P.H.Deeply supervised
salient object detection with short connections.IEEE Conference on Computer
Vision and Pattern Recognition,3203-3212(2017).
[11]Lin,T.,Goyal,P.,Girshick,R.,He,K.&Dollar,P.Focal loss for dense
object detection.IEEE International Conference on Computer Vision,2999-3007
(2017)。
Claims (8)
1. a kind of digestive endoscope image classification based on multitask neural network cancer detection system early, which is characterized in that packet
It includes:
(1) feature extraction core network;
(2) digestive endoscope image classification branch;
(3) alimentary canal morning carninomatosis becomes region detection branch;
The feature extraction core network, constructs on the basis of VGG-16 and obtains, and removes the last three full connections of VGG-16
After layer, core network includes 13 convolutional layers, and the characteristic pattern of convolutional layer resolution ratio under the action of the layer of pond is gradually reduced, according to
Convolutional layer is divided into 5 groups, respectively conv1_2, conv2_2, conv3_3, conv4_3, conv5_ by the position where the layer of pond
3;
Digestive endoscope image classification branch, for input picture to be divided into abnormal and normal two class;It is average including the overall situation
Pond layer and 3 full articulamentums;The average pond layer of the overall situation by the two dimensional character figure of conv5_3 the last one convolutional layer be mapped as to
Amount is input to full articulamentum, and first full articulamentum includes 4096 neurons, and second full articulamentum includes 1024 nerves
Member, the full articulamentum of third includes 1 neuron, after sigmoid activation primitive, obtains input picture and is classified as Abnormal Map
The probability of picture;
The alimentary canal morning carninomatosis becomes region detection branch, is that the obvious object that Hou et al. is proposed detects network, by lesion region
As significant region, it is divided into significant pixel or not significant pixel, i.e. abnormal pixel or normal for each pixel in image
Pixel, specific structure are as follows:
There are 4 pond layers in core network, adds 1 pond layer again after the conv5_3 of core network, totally 5 pond layers, it can
To generate the characteristic pattern of 6 different scales, the corresponding side output of the characteristic pattern of each scale, the side output of different scale
It first passes around warp lamination and obtains characteristic pattern identical with input image resolution, then connected by short by the feature of higher level
Figure is connected to the characteristic pattern of lower level, allows the characteristic pattern of lower level to optimize the coarse characteristic pattern of higher level, through sigmoid
After function activation, the lesion region of raw 6 sides of common property detects figure;In order to further obtain more accurate testing result, by 6
The lesion region detection figure of side, which connects, is input to a convolutional layer, after the activation of sigmoid function, obtains fused
Lesion region detection figure.
2. system according to claim 1, which is characterized in that the parameter of core network is by pre- on ImageNet data set
Trained VGG-16 model initialization is classified two task sharings of branch and detection branches.
3. system according to claim 1, which is characterized in that loss function used in classification branch is Focal
Loss, the output of classification branch are the Probability p that image is classified as abnormal image, and therefore, the form of loss function is as follows:
Loss=- ∑ αy(1-py)2log(py)
Wherein, y ∈ { 0,1 } is label, and 1 indicates abnormal, and 0 indicates normal;αyIt is set as 0.25.
4. system according to claim 1,2 or 3, which is characterized in that the lesion region detection of lesion detection branch output
The resolution ratio of figure and the resolution ratio of input picture are identical, and lesion region, which detects the value in figure, indicates each pixel in former endoscopic image
It is divided into the probability of abnormal pixel, selects the Focal Loss of α equilibrium as loss function, αyForm it is as follows:
Wherein, N indicates the quantity of pixel samples, NyIndicate the quantity of the pixel of y class.
5. system according to claim 4, which is characterized in that the training method of network model is as follows:
It is divided into alternately trained and joint training two parts, alternately when training, the parameter of branch corresponding to fixed one of task,
Update the loss function that remaining parameter in network is used to minimize another task;When joint training, whole network is updated
Parameter, while minimizing the loss function of two tasks.
6. system according to claim 4, which is characterized in that image classification method is as follows when test:
Image to be classified I is inputted, classification branch output I is the Probability p of abnormal image;Given threshold Tcla, as p > TclaWhen, then
Think that lesion has occurred in I, is abnormal image;Otherwise it is assumed that I is normal picture.
7. system according to claim 4, which is characterized in that irregularly early carninomatosis becomes method for detecting area such as when test
Under:
Image to be detected I is inputted, early carninomatosis becomes detection branches and exports lesion region detection figure identical with I size, with fusion
Lesion region detection figure M ∈ [0,1] is as final output;The value that lesion region detects in figure is bigger, indicates respective pixel hair
The probability of sick change is bigger;Given threshold Tsal, lesion region is M (i, j) > TsalCorresponding region, wherein i, j are pixel
Coordinate.
8. system according to claim 6 or 7, which is characterized in that training sample includes at least abnormal image 1000 and opens, just
Normal image 1000 is opened.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810928147.5A CN109118485A (en) | 2018-08-13 | 2018-08-13 | Digestive endoscope image classification based on multitask neural network cancer detection system early |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810928147.5A CN109118485A (en) | 2018-08-13 | 2018-08-13 | Digestive endoscope image classification based on multitask neural network cancer detection system early |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109118485A true CN109118485A (en) | 2019-01-01 |
Family
ID=64853110
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810928147.5A Pending CN109118485A (en) | 2018-08-13 | 2018-08-13 | Digestive endoscope image classification based on multitask neural network cancer detection system early |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109118485A (en) |
Cited By (30)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109919212A (en) * | 2019-02-26 | 2019-06-21 | 中山大学肿瘤防治中心 | The multi-dimension testing method and device of tumour in digestive endoscope image |
CN109949275A (en) * | 2019-02-26 | 2019-06-28 | 中山大学肿瘤防治中心 | A kind of diagnostic method and device of superior gastrointestinal endoscope image |
CN110033042A (en) * | 2019-04-15 | 2019-07-19 | 青岛大学 | A kind of carcinoma of the rectum ring week incisxal edge MRI image automatic identifying method and system based on deep neural network |
CN110045699A (en) * | 2019-04-17 | 2019-07-23 | 北京中大科慧科技发展有限公司 | Data center's power manages system and method |
CN110136828A (en) * | 2019-05-16 | 2019-08-16 | 杭州健培科技有限公司 | A method of medical image multitask auxiliary diagnosis is realized based on deep learning |
CN110495847A (en) * | 2019-08-23 | 2019-11-26 | 重庆天如生物科技有限公司 | Alimentary canal morning cancer assistant diagnosis system and check device based on deep learning |
CN110717907A (en) * | 2019-10-06 | 2020-01-21 | 浙江大学 | Intelligent hand tumor detection method based on deep learning |
CN110763685A (en) * | 2019-10-22 | 2020-02-07 | 陕西源杰半导体技术有限公司 | Artificial intelligent detection method and device for DFB semiconductor laser chip surface defects |
CN110867249A (en) * | 2019-09-01 | 2020-03-06 | 厦门影诺医疗科技有限公司 | Intelligent real-time detection platform for lower alimentary canal uplift type lesion |
CN111047582A (en) * | 2019-12-17 | 2020-04-21 | 山东大学齐鲁医院 | Crohn's disease auxiliary diagnosis system under enteroscope based on degree of depth learning |
CN111222454A (en) * | 2020-01-03 | 2020-06-02 | 暗物智能科技(广州)有限公司 | Method and system for training multi-task target detection model and multi-task target detection |
CN111899229A (en) * | 2020-07-14 | 2020-11-06 | 武汉楚精灵医疗科技有限公司 | Advanced gastric cancer auxiliary diagnosis method based on deep learning multi-model fusion technology |
CN112102256A (en) * | 2020-08-22 | 2020-12-18 | 复旦大学 | Narrow-band endoscopic image-oriented cancer focus detection and diagnosis system for early esophageal squamous carcinoma |
CN112102234A (en) * | 2020-08-06 | 2020-12-18 | 复旦大学 | Ear sclerosis focus detection and diagnosis system based on target detection neural network |
CN112102332A (en) * | 2020-08-30 | 2020-12-18 | 复旦大学 | Cancer WSI segmentation method based on local classification neural network |
ES2799828A1 (en) * | 2019-06-17 | 2020-12-21 | Quibim S L | METHOD AND SYSTEM TO IDENTIFY ANOMALIES IN X-RAYS (Machine-translation by Google Translate, not legally binding) |
CN112308827A (en) * | 2020-10-23 | 2021-02-02 | 复旦大学 | Hair follicle detection method based on deep convolutional neural network |
CN112419246A (en) * | 2020-11-12 | 2021-02-26 | 复旦大学 | Depth detection network for quantifying esophageal mucosa IPCLs blood vessel morphological distribution |
CN112419248A (en) * | 2020-11-13 | 2021-02-26 | 复旦大学 | Ear sclerosis focus detection and diagnosis system based on small target detection neural network |
CN112466466A (en) * | 2021-01-27 | 2021-03-09 | 萱闱(北京)生物科技有限公司 | Digestive tract auxiliary detection method and device based on deep learning and computing equipment |
CN112884728A (en) * | 2021-02-04 | 2021-06-01 | 复旦大学附属中山医院 | Endoscope image lesion identification screening method based on deep learning convolutional neural network |
CN112884777A (en) * | 2021-01-22 | 2021-06-01 | 复旦大学 | Multi-modal collaborative esophageal cancer lesion image segmentation system based on self-sampling similarity |
WO2021139447A1 (en) * | 2020-09-30 | 2021-07-15 | 平安科技(深圳)有限公司 | Abnormal cervical cell detection apparatus and method |
CN113393456A (en) * | 2021-07-13 | 2021-09-14 | 湖南大学 | Automatic quality control method of early pregnancy fetus standard section based on multiple tasks |
CN113688788A (en) * | 2021-04-02 | 2021-11-23 | 四川大学 | Method and system for automatically identifying airway images on head side sheet based on deep learning |
CN114648509A (en) * | 2022-03-25 | 2022-06-21 | 中国医学科学院肿瘤医院 | Thyroid cancer detection system based on multi-classification task |
DE202023101183U1 (en) | 2023-03-12 | 2023-03-22 | Kusum Yadav | Artificial intelligence and IoT-based smart health system that uses image processing and deep learning to prevent and predict chronic gastrointestinal diseases |
WO2023060874A1 (en) * | 2021-10-12 | 2023-04-20 | 上海蜜度信息技术有限公司 | Picture classification and object detection synchronous processing method and system, storage medium, and terminal |
CN116740475A (en) * | 2023-08-15 | 2023-09-12 | 苏州凌影云诺医疗科技有限公司 | Digestive tract image recognition method and system based on state classification |
CN117558414A (en) * | 2023-11-23 | 2024-02-13 | 之江实验室 | System, electronic device and medium for predicting early recurrence of multi-tasking hepatocellular carcinoma |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107492095A (en) * | 2017-08-02 | 2017-12-19 | 西安电子科技大学 | Medical image pulmonary nodule detection method based on deep learning |
CN107808167A (en) * | 2017-10-27 | 2018-03-16 | 深圳市唯特视科技有限公司 | A kind of method that complete convolutional network based on deformable segment carries out target detection |
CN108154509A (en) * | 2018-01-12 | 2018-06-12 | 平安科技(深圳)有限公司 | Cancer recognition methods, device and storage medium |
CN108389207A (en) * | 2018-04-28 | 2018-08-10 | 上海视可电子科技有限公司 | A kind of the tooth disease diagnosing method, diagnostic device and intelligent image harvester |
-
2018
- 2018-08-13 CN CN201810928147.5A patent/CN109118485A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107492095A (en) * | 2017-08-02 | 2017-12-19 | 西安电子科技大学 | Medical image pulmonary nodule detection method based on deep learning |
CN107808167A (en) * | 2017-10-27 | 2018-03-16 | 深圳市唯特视科技有限公司 | A kind of method that complete convolutional network based on deformable segment carries out target detection |
CN108154509A (en) * | 2018-01-12 | 2018-06-12 | 平安科技(深圳)有限公司 | Cancer recognition methods, device and storage medium |
CN108389207A (en) * | 2018-04-28 | 2018-08-10 | 上海视可电子科技有限公司 | A kind of the tooth disease diagnosing method, diagnostic device and intelligent image harvester |
Non-Patent Citations (3)
Title |
---|
QIBIN HOU 等: ""Deeply supervised salient object detection with short connections"", 《ARXIV》 * |
TSUNG-YI LIN 等: ""Focal Loss for Dense Object Detection"", 《IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE》 * |
XULEI YANG 等: ""A Novel Multi-task Deep Learning Model for Skin Lesion Segmentation and Classification"", 《ARXIV》 * |
Cited By (40)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109919212A (en) * | 2019-02-26 | 2019-06-21 | 中山大学肿瘤防治中心 | The multi-dimension testing method and device of tumour in digestive endoscope image |
CN109949275A (en) * | 2019-02-26 | 2019-06-28 | 中山大学肿瘤防治中心 | A kind of diagnostic method and device of superior gastrointestinal endoscope image |
CN110033042A (en) * | 2019-04-15 | 2019-07-19 | 青岛大学 | A kind of carcinoma of the rectum ring week incisxal edge MRI image automatic identifying method and system based on deep neural network |
CN110045699A (en) * | 2019-04-17 | 2019-07-23 | 北京中大科慧科技发展有限公司 | Data center's power manages system and method |
CN110136828A (en) * | 2019-05-16 | 2019-08-16 | 杭州健培科技有限公司 | A method of medical image multitask auxiliary diagnosis is realized based on deep learning |
WO2020254700A1 (en) * | 2019-06-17 | 2020-12-24 | Quibim, S.L. | Method and system for identifying anomalies in x-rays |
ES2799828A1 (en) * | 2019-06-17 | 2020-12-21 | Quibim S L | METHOD AND SYSTEM TO IDENTIFY ANOMALIES IN X-RAYS (Machine-translation by Google Translate, not legally binding) |
CN110495847A (en) * | 2019-08-23 | 2019-11-26 | 重庆天如生物科技有限公司 | Alimentary canal morning cancer assistant diagnosis system and check device based on deep learning |
CN110495847B (en) * | 2019-08-23 | 2021-10-08 | 重庆天如生物科技有限公司 | Advanced learning-based auxiliary diagnosis system and examination device for early cancer of digestive tract |
CN110867249A (en) * | 2019-09-01 | 2020-03-06 | 厦门影诺医疗科技有限公司 | Intelligent real-time detection platform for lower alimentary canal uplift type lesion |
CN110717907A (en) * | 2019-10-06 | 2020-01-21 | 浙江大学 | Intelligent hand tumor detection method based on deep learning |
CN110763685B (en) * | 2019-10-22 | 2020-12-08 | 陕西源杰半导体技术有限公司 | Artificial intelligent detection method and device for DFB semiconductor laser chip surface defects |
CN110763685A (en) * | 2019-10-22 | 2020-02-07 | 陕西源杰半导体技术有限公司 | Artificial intelligent detection method and device for DFB semiconductor laser chip surface defects |
CN111047582A (en) * | 2019-12-17 | 2020-04-21 | 山东大学齐鲁医院 | Crohn's disease auxiliary diagnosis system under enteroscope based on degree of depth learning |
CN111222454A (en) * | 2020-01-03 | 2020-06-02 | 暗物智能科技(广州)有限公司 | Method and system for training multi-task target detection model and multi-task target detection |
CN111222454B (en) * | 2020-01-03 | 2023-04-07 | 暗物智能科技(广州)有限公司 | Method and system for training multi-task target detection model and multi-task target detection |
CN111899229A (en) * | 2020-07-14 | 2020-11-06 | 武汉楚精灵医疗科技有限公司 | Advanced gastric cancer auxiliary diagnosis method based on deep learning multi-model fusion technology |
CN112102234A (en) * | 2020-08-06 | 2020-12-18 | 复旦大学 | Ear sclerosis focus detection and diagnosis system based on target detection neural network |
CN112102256B (en) * | 2020-08-22 | 2022-04-12 | 复旦大学 | Narrow-band endoscopic image-oriented cancer focus detection and diagnosis system for early esophageal squamous carcinoma |
CN112102256A (en) * | 2020-08-22 | 2020-12-18 | 复旦大学 | Narrow-band endoscopic image-oriented cancer focus detection and diagnosis system for early esophageal squamous carcinoma |
CN112102332A (en) * | 2020-08-30 | 2020-12-18 | 复旦大学 | Cancer WSI segmentation method based on local classification neural network |
WO2021139447A1 (en) * | 2020-09-30 | 2021-07-15 | 平安科技(深圳)有限公司 | Abnormal cervical cell detection apparatus and method |
CN112308827A (en) * | 2020-10-23 | 2021-02-02 | 复旦大学 | Hair follicle detection method based on deep convolutional neural network |
CN112419246A (en) * | 2020-11-12 | 2021-02-26 | 复旦大学 | Depth detection network for quantifying esophageal mucosa IPCLs blood vessel morphological distribution |
CN112419248A (en) * | 2020-11-13 | 2021-02-26 | 复旦大学 | Ear sclerosis focus detection and diagnosis system based on small target detection neural network |
CN112419248B (en) * | 2020-11-13 | 2022-04-12 | 复旦大学 | Ear sclerosis focus detection and diagnosis system based on small target detection neural network |
CN112884777B (en) * | 2021-01-22 | 2022-04-12 | 复旦大学 | Multi-modal collaborative esophageal cancer lesion image segmentation system based on self-sampling similarity |
CN112884777A (en) * | 2021-01-22 | 2021-06-01 | 复旦大学 | Multi-modal collaborative esophageal cancer lesion image segmentation system based on self-sampling similarity |
CN112466466A (en) * | 2021-01-27 | 2021-03-09 | 萱闱(北京)生物科技有限公司 | Digestive tract auxiliary detection method and device based on deep learning and computing equipment |
CN112466466B (en) * | 2021-01-27 | 2021-05-18 | 萱闱(北京)生物科技有限公司 | Digestive tract auxiliary detection method and device based on deep learning and computing equipment |
CN112884728A (en) * | 2021-02-04 | 2021-06-01 | 复旦大学附属中山医院 | Endoscope image lesion identification screening method based on deep learning convolutional neural network |
CN113688788A (en) * | 2021-04-02 | 2021-11-23 | 四川大学 | Method and system for automatically identifying airway images on head side sheet based on deep learning |
CN113393456B (en) * | 2021-07-13 | 2022-04-19 | 湖南大学 | Automatic quality control method of early pregnancy fetus standard section based on multiple tasks |
CN113393456A (en) * | 2021-07-13 | 2021-09-14 | 湖南大学 | Automatic quality control method of early pregnancy fetus standard section based on multiple tasks |
WO2023060874A1 (en) * | 2021-10-12 | 2023-04-20 | 上海蜜度信息技术有限公司 | Picture classification and object detection synchronous processing method and system, storage medium, and terminal |
CN114648509A (en) * | 2022-03-25 | 2022-06-21 | 中国医学科学院肿瘤医院 | Thyroid cancer detection system based on multi-classification task |
DE202023101183U1 (en) | 2023-03-12 | 2023-03-22 | Kusum Yadav | Artificial intelligence and IoT-based smart health system that uses image processing and deep learning to prevent and predict chronic gastrointestinal diseases |
CN116740475A (en) * | 2023-08-15 | 2023-09-12 | 苏州凌影云诺医疗科技有限公司 | Digestive tract image recognition method and system based on state classification |
CN116740475B (en) * | 2023-08-15 | 2023-10-17 | 苏州凌影云诺医疗科技有限公司 | Digestive tract image recognition method and system based on state classification |
CN117558414A (en) * | 2023-11-23 | 2024-02-13 | 之江实验室 | System, electronic device and medium for predicting early recurrence of multi-tasking hepatocellular carcinoma |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109118485A (en) | Digestive endoscope image classification based on multitask neural network cancer detection system early | |
CN109376636B (en) | Capsule network-based eye fundus retina image classification method | |
Kaymak et al. | Automated age-related macular degeneration and diabetic macular edema detection on oct images using deep learning | |
Xiancheng et al. | Retina blood vessel segmentation using a U-net based Convolutional neural network | |
CN107437092A (en) | The sorting algorithm of retina OCT image based on Three dimensional convolution neutral net | |
CN106780466A (en) | A kind of cervical cell image-recognizing method based on convolutional neural networks | |
CN110288597B (en) | Attention mechanism-based wireless capsule endoscope video saliency detection method | |
CN108446729A (en) | Egg embryo classification method based on convolutional neural networks | |
CN112102256B (en) | Narrow-band endoscopic image-oriented cancer focus detection and diagnosis system for early esophageal squamous carcinoma | |
CN110647875B (en) | Method for segmenting and identifying model structure of blood cells and blood cell identification method | |
CN106980852B (en) | Based on Corner Detection and the medicine identifying system matched and its recognition methods | |
CN109670510A (en) | A kind of gastroscopic biopsy pathological data screening system and method based on deep learning | |
CN107066934A (en) | Tumor stomach cell image recognition decision maker, method and tumor stomach section identification decision equipment | |
CN108921825A (en) | The method and device of the facial skin points shape defect of detection based on deep learning | |
Ghosh et al. | Effective deep learning for semantic segmentation based bleeding zone detection in capsule endoscopy images | |
CN114998210B (en) | Retinopathy of prematurity detecting system based on deep learning target detection | |
CN108009592A (en) | A kind of diabetic retinal classification of images method | |
Chen et al. | Cell nuclei detection and segmentation for computational pathology using deep learning | |
CN112528782B (en) | Underwater fish target detection method and device | |
CN107274406A (en) | A kind of method and device of detection sensitizing range | |
Junjun et al. | Diabetic retinopathy detection based on deep convolutional neural networks for localization of discriminative regions | |
CN110188767A (en) | Keratonosus image sequence feature extraction and classifying method and device based on deep neural network | |
CN110288574A (en) | A kind of adjuvant Ultrasonographic Diagnosis hepatoncus system and method | |
Nazki et al. | MultiPathGAN: Structure preserving stain normalization using unsupervised multi-domain adversarial network with perception loss | |
CN109241963A (en) | Blutpunkte intelligent identification Method in capsule gastroscope image based on Adaboost machine learning |
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: 20190101 |
|
WD01 | Invention patent application deemed withdrawn after publication |