CN107680678A - Based on multiple dimensioned convolutional neural networks Thyroid ultrasound image tubercle auto-check system - Google Patents
Based on multiple dimensioned convolutional neural networks Thyroid ultrasound image tubercle auto-check system Download PDFInfo
- Publication number
- CN107680678A CN107680678A CN201710975555.1A CN201710975555A CN107680678A CN 107680678 A CN107680678 A CN 107680678A CN 201710975555 A CN201710975555 A CN 201710975555A CN 107680678 A CN107680678 A CN 107680678A
- Authority
- CN
- China
- Prior art keywords
- tubercle
- thyroid
- image
- feature
- multiple dimensioned
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- 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
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
-
- 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
- G06T2207/10—Image acquisition modality
- G06T2207/10132—Ultrasound image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Biomedical Technology (AREA)
- Software Systems (AREA)
- Bioinformatics & Computational Biology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Evolutionary Biology (AREA)
- Multimedia (AREA)
- Medical Informatics (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Radiology & Medical Imaging (AREA)
- Quality & Reliability (AREA)
- Ultra Sonic Daignosis Equipment (AREA)
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
The present invention provides one kind and is based on multiple dimensioned convolutional neural networks Thyroid ultrasound image tubercle auto-check system, including:Thyroid nodule is by thick tubercle sort module, thyroid nodule region automatic detection module and thyroid nodule essence sort module to essence;The feature of the size of different sensing regions is extracted by the convolutional neural networks of multiple dimensioned Fusion Features, thyroid nodule is automatically positioned so as to combine the context semantic feature of local and global information extraction tubercle.The present invention passes through the multiple dimensioned feature extraction by the thick neutral net to essence and the multiple dimensioned essence classification AlexNet by designing pyramid structure, the position of focus and the probability of good pernicious generation can accurately be predicted, doctor can be aided in carry out the diagnosis of thyroid gland focus, improve the objectivity of diagnosis, it is good with real-time, the characteristics of accuracy rate is high.
Description
Technical field
The present invention relates to based on multiple dimensioned convolutional neural networks Thyroid ultrasound image tubercle auto-check system, belong to deep
Degree study and medical assistance field.
Background technology
In the 1940s, ultrasonic imaging technique is begun in clinical practice, in 1962, start to carry out thyroid gland
Diagnosis.Thyroid ultrasound imaging technique can provide on thyroid hoc scenario for doctor, and it is first to have become modern doctor
The diagnosis of technique means of choosing.
Convolutional neural networks technology plays the research history for having had more than 30 years so far, volume since last century the eighties
Product neutral net is initially applied to the numeral identification of handwritten form, always is in identifications such as computer vision field, face, objects
One very challenging study hotspot.In recent years, with internet development, the acquisition of mass data collection is no longer one
Individual problem, 12 years, Alex obtained natural image Imagenet using convolutional neural networks with the advantage more than second place 20%
The champion of match, hereafter convolutional neural networks started the new epoch, from object identification, recognition of face, object positioning, semanteme point
The field such as cut and ranking list has all been refreshed with overwhelming advantage.
Observation analysis of the identification of thyroid excusing from death image before this always dependent on experienced doctor diagnoses, and not only consumes
When effort, and because doctor's resource in China is unbalanced, it is difficult to accomplish the lifting of integral level.By means of convolutional neural networks
For the advantage of natural image identification, there is high-quality medical resource to be converted into algorithm, new image is carried out the positioning of focus with
Identification, is applied in the identification of Thyroid ultrasound.
In addition, the automatic diagnosis of Thyroid ultrasound to improve doctor operating efficiency play the role of it is important, have it is wide
Application prospect and huge economic value.However, because excusing from death image has, contrast is low at present, the unconspicuous spy in border
Sign, vision difference is small, and the research to thyroid ultrasonoscopy is also faced with lot of challenges, because these phenomenons contain more behind
For the physics and visual signature mechanism of complexity, compellent simulated effect is realized, it is necessary to which multi-crossed disciplines are theoretical and soft or hard
The highly effective algorithm that part combines is configured as supporting.
The content of the invention
Present invention solves the technical problem that it is:Overcome the deficiency of the diagnosis of existing Thyroid ultrasound image, there is provided one
Kind is based on multiple dimensioned convolutional neural networks Thyroid ultrasound image tubercle auto-check system, realizes the automatic of thyroid nodule
Positioning and identification, the positioning of tubercle and good pernicious judgement can be effectively carried out to Thyroid ultrasound image.
The technical solution adopted by the present invention is:Examined automatically based on multiple dimensioned convolutional neural networks Thyroid ultrasound image tubercle
Disconnected system, it is characterised in that including:Thyroid nodule is by thick tubercle sort module, thyroid nodule region automatic detection to essence
Module and thyroid nodule essence sort module;Wherein:
Thyroid nodule is by the thick tubercle sort module to essence:Including by the thick Thyroid ultrasound image tubercle to essence oneself
Dynamic positioning and classification, described be automatically positioned include two stages with classification, and first stage is the rough sort stage, by number
Data preprocess process, using the detection algorithm based on deep neural network VGG, extract first by by multiple dimensioned convolutional Neural institute
The semantic feature of description, positioning and preliminary rough sort are carried out to possible thyroid nodule, and provide candidate frame;Second stage
For the ultrasonoscopy block in candidate frame, nature is trained on ImageNet data sets using the Alex networks further lifted
The parameter model of the disaggregated model of image is trained on the image data set of thyroid nodule and transfer learning;After training
By model parameter extraction characteristic vector, the probable value of each class is obtained by softmax functions, to provide final tubercle
Position and good pernicious judged result;
Thyroid nodule region automatic detection module:Including training and test phase, in the training stage, pass through basic network
Multiple dimensioned convolution feature is extracted with the multiple dimensioned full convolutional network newly increased, the convolution kernel of multiple yardsticks is utilized in detection process
The feature and context semantic information of multiple yardsticks of input picture are extracted, so as to obtain characteristic response figure, it is pre- to include tubercle
The coordinate information of frame is surveyed, while carries out classification based training, it is good pernicious with predicted position and tentative prediction;Meanwhile design is based on grid
Candidate region selection mechanism, to image block carry out feature extraction;Entered respectively by two object functions after image block extraction
Row constraint is the softmax functions of classification respectively, for the position of tubercle, enters row constraint using the L2 norms of recurrence, simultaneously
To nodule position information, the network parameter trained is preserved;In test phase, the image uploaded first to user is located in advance
Reason, is then input in the parameter model trained;
Thyroid ultrasound image essence sort module:The position of the candidate frame obtained according to automatic detection module, to original
Ultrasonoscopy intercept the image block of relevant position, i.e., the overall random interception of image is had the subgraph of fixed size, and handle
Image block is input in smart Classification Neural, and the neural sorter network of essence classification extracts the property of multilayer feature using convolution operation
Matter, with reference to the unfixed feature of Thyroid ultrasound image tubercle size, pyramid is introduced in the structure of smart Classification Neural
Space structure, while multiple dimensioned pyramid convolution feature is extracted, pass through on the basis of multiple dimensioned pyramid convolution feature
The good pernicious probability of softmax Function Estimations, provides the smart classification results for belonging to good Malignant Nodules.
The thyroid nodule ultrasonoscopy is as follows by the specific implementation process of the thick tubercle sort module to essence:
(1) data prediction is carried out to the Thyroid ultrasound image of acquired original by data preprocessing method and data increases
By force, pretreatment includes carrying out data size replacement, and data strengthen to ultrasonoscopy Mirroring Mapping, random piecemeal, overall diagram
As being divided into different image-regions, to increase the amount of training data;
(2) by by thick to two stages of essence, in the rough sort stage, for detect tubercle by multiple dimensioned convolutional Neural
The semantic feature described by end-to-end automatic detection network model on VGG network foundations, while by detecting thyroid nodule
Judge that its is good pernicious while region, obtain the candidate frame of thyroid nodule and the result of rough sort to possible thyroid nodule
Carry out positioning and preliminary rough sort;
(3) in smart sorting phase, after data prediction and data enhancing are carried out to original labeled data, according to candidate
Ultrasonoscopy block in frame, the figure in candidate frame is further classified by the AlexNet neutral nets of redesign
The good pernicious judged result determined, utilizes the Alex networks further lifted, it is proposed that for thyroid ultrasonic tubercle
Multiple dimensioned pyramid convolutional neural networks, and on ImageNet data sets train natural image disaggregated model parameter
Model is trained on the image data set of thyroid nodule.By model parameter extraction characteristic vector after training, pass through
Softmax functions obtain the probable value of each class, to provide the position of final tubercle and good pernicious judged result.
Thyroid nodule region automatic detection module is implemented as follows:
In the training stage,
(1) the multiple dimensioned convolution kernel feature of extraction logical first, specifically utilizes net based on VGG-16 convolutional neural networks
Network feature extraction, the full convolutional layer of three yardsticks is then added, the wherein size of convolution kernel is 1,3,5,7 further to extract
The feature of different scale and the context semantic information in space, so as to obtain characteristic response figure, include the seat of tubercle prediction block
Information is marked, while carries out classification based training, it is good pernicious with predicted position and tentative prediction;
(2) the candidate region selection mechanism based on grid is designed, after carrying out feature extraction based on general image, is being rolled up entirely
Being designed after lamination increases multiple fixed proportion sized images blocks, and same image averaging is divided into difference according to tile size
The small image block of ratio, and carry out feature extraction further directed to the image block in frame;
(3) the softmax functions that row constraint is classification respectively are entered by two object functions respectively after image block extraction,
Input is the Feature maps that probe frame passes through after convolutional layer, and output is into good pernicious probability;Meanwhile for tubercle
Position, enter row constraint using the L2 norms of recurrence, while obtain nodule position information, preserve the network parameter trained;
(4) pre-processed in test phase, the image uploaded first to user, be then input to the parameter mould trained
In type.
The Thyroid ultrasound image essence sort module is implemented as follows:
(1) data prediction and data are carried out to original labeled data first strengthens, wherein, data prediction includes pair
Data carry out size normalization, and the normalization of distribution, its distribution is belonged to normal distribution, average 0, variance 1;
(2) and then by the Alexnet of first five layer the feature that feature extraction tentatively extracts Thyroid ultrasound image is carried out;
(3) in layer 6, it is proposed that for the multiple dimensioned pyramid convolutional neural networks of thyroid ultrasonic tubercle,
After layer 5 input, all passages are split into three passages, extract the feature of the size of different convolution kernels side by side, point
The spatial pyramid structure of prestige prestige 1,3,5,7, finally merges into same passage;
(4) after merging, row constraint is entered by using the object function softmax of forecast classification, and in ImageNet numbers
It is trained according to the parameter model for the disaggregated model that natural image is trained on collection on the image data set of thyroid nodule.Simultaneously
Gradient decline is carried out using back-propagation algorithm, every a kind of probable value is obtained, to provide the position of final tubercle and good evil
The judged result of property.
The principle of the present invention is:
(1) by carrying out automatic detection and classification to smart multiple dimensioned mode by thick, calculated using based on SSD method
The candidate frame of thyroid tubercle.In order to take into account thyroid multiple dimensioned and polymorphic feature.
(2) present invention devises the multi-level network structure for judging network structure completion rough sort and essence classification.
(3) in order to carry out the automatic detection of thyroid nodule, the present invention is by based on SSD automatic detection networks, it is proposed that
A kind of automatic detection convolutional neural networks mapped by multiple features.This method is calculated by using convolutional neural networks
Feature Mapping, whole mapping graph is carried out to the decomposition of multiple yardsticks, the image block after decomposition classified, to predict first shape
The position of gland tubercle, and the tubercle in image block is automatically classified.
(4) in order to further be classified to thyroid tubercle, the present invention passes through the pyramidal feature in design space
UNE structure is diagnosed automatically to the thyroid nodule of automatic detection in (2).
The present invention being a little compared with prior art:
(1) it is proposed by the present invention based on by thick automatic detection and identification method to essence, it on the one hand can carry out quick standard
True detection.On the other hand the result for providing and accurately identifying can be layered.And the feature of multiple dimensioned spatial pyramid is carried out
UNE structure.
(2) existing target automatic detection algorithm is contrasted, it is proposed by the present invention to be based on the multiple dimensioned polymorphic spy of thyroid gland
Sign mapping blending algorithm, can adapt to the polymorphic automatic detection and rough sort of thyroid nodule.
(3) it is proposed by the present invention based on the pyramidal automatic classification of thyroid nodule multiscale space, can be to thyroid gland
Different scale and the tubercle of form classify automatically.
Brief description of the drawings
Fig. 1 is being formed based on multiple dimensioned convolutional neural networks Thyroid ultrasound image tubercle auto-check system for the present invention
Block diagram;
Fig. 2 is based on multiple dimensioned convolutional neural networks Thyroid ultrasound image tubercle auto-check system flow for the present invention
Figure;
Fig. 3 is by the thick tubercle sort module implementation process figure to essence;
Fig. 4 is thyroid nodule region automatic detection module implementation process figure;
Fig. 5 is thyroidine sort module implementation process figure;
Fig. 6 is the output result of automatic recognition system, thyroid nodule detection and classification results schematic diagram.
Embodiment
The present invention is further illustrated with reference to other accompanying drawings and embodiment.
As shown in Figure 1, 2, it is of the invention to be diagnosed automatically based on multiple dimensioned convolutional neural networks Thyroid ultrasound image tubercle
System is included by thick tubercle sort module, thyroid nodule region automatic detection module and thyroidine sort module to essence.
Fig. 1 is based on multiple dimensioned convolutional neural networks Thyroid ultrasound image tubercle auto-check system, including three modules:
Thyroid nodule region automatic detection module, thyroid nodule by slightly to smart sort module and Thyroid ultrasound essence sort module,
Wherein automatic detection module key technology is multiple dimensioned convolutional layer fusion, and the key technology of second module is full images candidate knot
Save the function constraint that position returns and classified;3rd module is the design of Analysis On Multi-scale Features pyramid structure.
Fig. 2 is to be based on multiple dimensioned convolutional neural networks Thyroid ultrasound image tubercle auto-check system flow chart, first
Carry out knuckle areas automatic detection, then by by the thick taxonomy model progress tubercle to essence be automatically positioned and rough sort, most
Rough sort result carries out precise classification afterwards.
As shown in figure 3, realized by the thick thyroid nodule sort module to essence as follows:
(1) data prediction and data enhancing are carried out to the Thyroid ultrasound image of acquired original first.Pretreatment includes
The size that size is reset to 300*300 fixation is carried out to image.Data strengthen to ultrasonoscopy Mirroring Mapping, random piecemeal,
General image is divided into different image-regions;
(2) module includes, in the rough sort stage, inputting ultrasonoscopy to be sorted, more chis to smart two stages by thick
It is respectively that 3,5,7 grade convolution kernels (size is consistent with VGG-16) are rolled up to spend convolutional Neural VGG networks to carry out 16 convolution sizes
Product, the semantic feature described by end-to-end automatic detection network model, its Feature maps is calculated by convolution and pond layer,
Three thyroid gland convolutional layers are added on the basis of basic network, for the layer newly added, pass through 32,64,128 volumes respectively
Product core is predicted.For the characteristic layer that a size is w*h, c passage, it is predicted using 5*5 convolution kernel, is entirely being schemed
The probability bivector of the position of tubercle is predicted on picture, the element of the bivector is good pernicious score respectively.Finally, it is defeated
Go out the result of the rough sort of tubercle;
(3) in smart assorting process, the ultrasonoscopy block region in candidate frame, the AlexNet redesigned is configured
Neutral net carries out the good pernicious judged result that is determined of further classify to the image in candidate frame, utilizes and is further lifted
Alex networks trained on ImageNet data sets natural image disaggregated model parameter model thyroid nodule figure
As being trained on data set and transfer learning.By model parameter extraction characteristic vector after training, pass through softmax functions
Every a kind of probable value is obtained, to provide the position of final tubercle and good pernicious judged result.
As shown in figure 4, automatic detection module realization in thyroid nodule region is as follows:
(1) the Thyroid ultrasound image, it is necessary to global is extracted for the unconspicuous thyroid feature of ultrasound in border first
Context semantic information auxiliary judgment.Relatively large convolution yardstick is devised herein and carries out amalgamation of global information, is used first
VGGNet carries out tuning and extraction Feature map, and the segmentation of image block is carried out on Feature map, and uses fixed size
The candidate frame of size be scanned, be predicted.The size of candidate frame now applies to thyroid tubercle.Tubercle
Ratio typically 4:1 and 1:Between 5.The yardstick of candidate frame is also in the range of this to adapt to the detection of thyroid tubercle.
(2) and then after last pooling layer of network above-mentioned candidate frame is added.Acted in the form of convolution
To feature map.On each image block, the classification that the region belongs to is predicted by network structure.Each image block
There is the prediction result of the candidate frame of the size of 10 ratios.The characteristics of image of multiple yardsticks can be thus predicted.
(3) in Feature maps L nodule candidate frame of each position prediction.For each candidate frame, 2 classes are predicted
Other score, and 4 deviants relative to candidate frame set in advance, (2+4) * L fallout predictors are so needed, w*h's
(2+4) * L*w*h predicted values will be produced on Feature maps.
(4) finally, the ratio per a kind of sample of loss function is controlled to carry out.The ratio taken is 3:1, pernicious knot
The weight of section account for 75%.To lift the importance of classification.It is capable of the classification pernicious sample of more accuracy rate.
Lcla=0.75lp+0。25ln
Lall=Lloc+Lcla
Wherein, LclaThe loss function of presentation class, the present invention is using Softmaxloss.LlocIt is the recurrence of position
Loss function, take is L2 norms to the present invention.The backpropagation for calculating L2 norms carries out the renewal of model parameter.By with
The constraint of upper two formula obtains the rough sort result of thyroid nodule and the position of tubercle.
As shown in figure 5, smart Classification Neural module realization is as follows:
(1) scale size of the image block in the candidate frame obtained in automatic detection module is reset to 224*224 first,
Calculate image and the convolution kernel of initialization carries out sliding window scanning, pond is carried out on obtained new Feature maps pictures
Operation, half of the image down sampling to original size;
(2) Feature maps activation primitive layer ReLu is calculated, is controlled in the range of [0, just infinite], is calculated
The result of corresponding multiplication between Feature maps and Relu activation primitive elements, the small feature extraction of filter value mainly influence
Semantic category another characteristic;
(3) divide by 6 layers of convolution and pond and ReLu operations, layer 6s of the Feature maps in AlexNet
Convolution kernel for three passages is parallel, and convolution kernel size is respectively 3,5,7, calculates Feature maps with after convolution, obtaining
New Feature maps, then carry out multi-channel feature merging, finally by the training of full articulamentum obtain characteristic vector with
Carry out category classification;
(4) gradient is calculated by back-propagation algorithm and stochastic gradient descent method, carries out parameter renewal;
(5) by the model parameter obtained in (4), feature extraction is carried out to test image, then passes through softmax functions
Carry out good pernicious prediction, obtain Two-dimensional Probabilistic vector, to it is good it is pernicious judge, export good pernicious result.
As shown in fig. 6, thyroid nodule automatic detection and the displaying of rough sort module results.
Thyroid nodule automatic detection result is the position of tubercle and the output of candidate frame, wherein dark and light color frame
For output result, dark color is pernicious, and light color is benign.Present system position is accurate, and can navigate to doctor and mark manually
The tubercle of the spill tag of note.
Shown in the following form of accuracy rate
Title | Accuracy rate (variance) |
AlexNet | 94.7% (1.1) |
GoogleNet | 95.2% (2.2) |
AlexNet+ pyramid structures | 94.5% (3) |
GoogleNet+ pyramid structures | 96.2% (2) |
The present invention | 97.4% (2) |
Above table shows the of the invention and algorithm of existing main flow comparative result, and first row is the name of algorithm, the
Two row are corresponding results, and variance is represented inside bracket.Accuracy rate is the average tested three times.In order to verify the present invention in data
Stability and generalization on collection, carried out 3 experiments, data set have 4000 samples, 2000 positive samples and 2000
Negative sample, randomly choose and be trained on 80% training set, 20% is tested, and the accuracy rate for showing the present invention is
97.4%, it is above other systems.
Claims (4)
1. it is based on multiple dimensioned convolutional neural networks Thyroid ultrasound image tubercle auto-check system, it is characterised in that:Including first
Shape gland tubercle is by thick to the tubercle sort module of essence, thyroid nodule region automatic detection module and thyroid nodule essence classification mould
Block;Wherein:
Thyroid nodule is by the thick tubercle sort module to essence:Including by the automatic fixed of the thick Thyroid ultrasound image tubercle to essence
Position and classification, described be automatically positioned include two stages with classification, and first stage is the rough sort stage, pre- by data
Processing procedure, using the detection algorithm based on deep neural network VGG, extract first by as described by multiple dimensioned convolutional Neural
Semantic feature, positioning and preliminary rough sort are carried out to possible thyroid nodule, and provide candidate frame;Second stage is root
According to the ultrasonoscopy block in candidate frame, natural image is trained on ImageNet data sets using the Alex networks further lifted
The parameter model of disaggregated model be trained on the image data set of thyroid nodule and transfer learning;Pass through after training
Model parameter extraction characteristic vector, the probable value of each class is obtained by softmax functions, to provide the position of final tubercle
With good pernicious judged result;
Thyroid nodule region automatic detection module:Including training and test phase, in the training stage, by basic network and newly
Increased multiple dimensioned full convolutional network extracts multiple dimensioned convolution feature, is extracted in detection process using the convolution kernel of multiple yardsticks
The feature and context semantic information of multiple yardsticks of input picture, so as to obtain characteristic response figure, include tubercle prediction block
Coordinate information, while carry out classification based training, it is good pernicious with predicted position and tentative prediction;Meanwhile design the time based on grid
Favored area selection mechanism, feature extraction is carried out to image block;Carried out about by two object functions respectively after image block extraction
Beam is the softmax functions of classification respectively, for the position of tubercle, enters row constraint using the L2 norms of recurrence, while tied
Positional information is saved, preserves the network parameter trained;In test phase, the image uploaded first to user pre-processes, so
It is input to afterwards in the parameter model trained;
Thyroid ultrasound image essence sort module:The position of the candidate frame obtained according to automatic detection module, to original ultrasound
Image intercept the image block of relevant position, i.e., overall random subgraph of the interception with fixed size of image, and image
Block is input in smart Classification Neural, and the neural sorter network of essence classification utilizes the property of convolution operation extraction multilayer feature, ginseng
The unfixed feature of Thyroid ultrasound image tubercle size is examined, pyramid space knot is introduced in the structure of smart Classification Neural
Structure, while multiple dimensioned pyramid convolution feature is extracted, pass through softmax functions on the basis of multiple dimensioned pyramid convolution feature
Estimate good pernicious probability, provide the smart classification results for belonging to good Malignant Nodules.
2. according to claim 1 diagnosed automatically based on multiple dimensioned convolutional neural networks Thyroid ultrasound image tubercle is
System, it is characterised in that:The thyroid nodule ultrasonoscopy is as follows by the specific implementation process of the thick tubercle sort module to essence:
(1) carrying out data prediction and data to the Thyroid ultrasound image of acquired original by data preprocessing method strengthens, in advance
Processing includes carrying out data size replacement, and data strengthen to ultrasonoscopy Mirroring Mapping, random piecemeal, general image split
For different image-regions, to increase the amount of training data;
(2) by by thick to two stages of essence, in the rough sort stage, for detect tubercle by multiple dimensioned convolutional Neural VGG nets
The semantic feature described by end-to-end automatic detection network model on the basis of network, while by detecting thyroid nodule region
Judge that its is good pernicious simultaneously, obtain the candidate frame of thyroid nodule and the result of rough sort is determined possible thyroid nodule
Position and preliminary rough sort;
(3) in smart sorting phase, after data prediction and data enhancing are carried out to original labeled data, according in candidate frame
Ultrasonoscopy block, further classification is carried out to the figure in candidate frame by the AlexNet neutral nets of redesign and is obtained
The good pernicious judged result determined, utilizes the Alex networks further lifted, it is proposed that for the more of thyroid ultrasonic tubercle
The pyramid convolutional neural networks of yardstick, and on ImageNet data sets train natural image disaggregated model parameter model
It is trained on the image data set of thyroid nodule;By model parameter extraction characteristic vector after training, pass through
Softmax functions obtain the probable value of each class, to provide the position of final tubercle and good pernicious judged result.
3. according to claim 1 diagnosed automatically based on multiple dimensioned convolutional neural networks Thyroid ultrasound image tubercle is
System, it is characterised in that:Thyroid nodule region automatic detection module is implemented as follows:
In the training stage,
(1) first by extracting multiple dimensioned convolution kernel feature, network based on VGG-16 convolutional neural networks is specifically utilized
Feature extraction, the full convolutional layer of three yardsticks is then added, the wherein size of convolution kernel is 3,5,7 further to extract difference
The feature of yardstick and the context semantic information in space, so as to obtain characteristic response figure, include the coordinate letter of tubercle prediction block
Breath, while classification based training is carried out, it is good pernicious with predicted position and tentative prediction;
(2) the candidate region selection mechanism based on grid is designed, after carrying out feature extraction based on general image, in full convolutional layer
Design increases multiple fixed proportion sized images blocks afterwards, and same image averaging is divided into different proportion according to tile size
Small image block, and further directed in frame image block carry out feature extraction;
(3) the softmax functions that row constraint is classification respectively, input are entered by two object functions respectively after image block extraction
It is the Feature maps that probe frame passes through after convolutional layer, output is into good pernicious probability;Meanwhile for the position of tubercle
Put, enter row constraint using the L2 norms of recurrence, while obtain nodule position information, preserve the network parameter trained;
(4) pre-processed in test phase, the image uploaded first to user, be then input to the parameter model trained
In.
4. according to claim 1 diagnosed automatically based on multiple dimensioned convolutional neural networks Thyroid ultrasound image tubercle is
System, it is characterised in that:The Thyroid ultrasound image essence sort module is implemented as follows:
(1) data prediction and data are carried out to original labeled data first strengthens, wherein, data prediction is included to data
Size normalization, and the normalization of distribution are carried out, its distribution is belonged to normal distribution, average 0, variance 1;
(2) and then by the Alexnet of first five layer the feature that feature extraction tentatively extracts Thyroid ultrasound image is carried out;
(3) in layer 6, it is proposed that for the multiple dimensioned pyramid convolutional neural networks of thyroid ultrasonic tubercle, the 5th
After layer inputs, all passages are split into three passages, the feature of the size of different convolution kernels is extracted side by side, is divided into 1,
3,5,7 spatial pyramid structures, finally merge into same passage;
(4) after merging, row constraint is entered by using the object function softmax of forecast classification, and in ImageNet data sets
The parameter model of the disaggregated model of upper training natural image is trained on the image data set of thyroid nodule, is utilized simultaneously
Back-propagation algorithm carries out gradient decline, obtains every a kind of probable value, to provide the position of final tubercle and good pernicious
Judged result.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710975555.1A CN107680678B (en) | 2017-10-18 | 2017-10-18 | Thyroid ultrasound image nodule diagnosis system based on multi-scale convolution neural network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710975555.1A CN107680678B (en) | 2017-10-18 | 2017-10-18 | Thyroid ultrasound image nodule diagnosis system based on multi-scale convolution neural network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN107680678A true CN107680678A (en) | 2018-02-09 |
CN107680678B CN107680678B (en) | 2020-12-01 |
Family
ID=61139726
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710975555.1A Active CN107680678B (en) | 2017-10-18 | 2017-10-18 | Thyroid ultrasound image nodule diagnosis system based on multi-scale convolution neural network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107680678B (en) |
Cited By (67)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108510495A (en) * | 2018-04-09 | 2018-09-07 | 沈阳东软医疗***有限公司 | A kind of lung image data processing method based on artificial intelligence, apparatus and system |
CN108648192A (en) * | 2018-05-17 | 2018-10-12 | 杭州依图医疗技术有限公司 | A kind of method and device of detection tubercle |
CN108664971A (en) * | 2018-05-22 | 2018-10-16 | 中国科学技术大学 | Pulmonary nodule detection method based on 2D convolutional neural networks |
CN108710919A (en) * | 2018-05-25 | 2018-10-26 | 东南大学 | A kind of crack automation delineation method based on multi-scale feature fusion deep learning |
CN108734694A (en) * | 2018-04-09 | 2018-11-02 | 华南农业大学 | Thyroid tumors ultrasonoscopy automatic identifying method based on faster r-cnn |
CN108742679A (en) * | 2018-06-29 | 2018-11-06 | 上海联影医疗科技有限公司 | Nodule detection device and method |
CN108830891A (en) * | 2018-06-05 | 2018-11-16 | 成都精工华耀科技有限公司 | A kind of rail splice fastener loosening detection method |
CN108875906A (en) * | 2018-04-20 | 2018-11-23 | 华侨大学 | A kind of multiple dimensioned convolutional neural networks learning method gradually to add up |
CN108968991A (en) * | 2018-05-08 | 2018-12-11 | 平安科技(深圳)有限公司 | Hand bone X-ray bone age assessment method, apparatus, computer equipment and storage medium |
CN109003282A (en) * | 2018-07-27 | 2018-12-14 | 京东方科技集团股份有限公司 | A kind of method, apparatus and computer storage medium of image procossing |
CN109063710A (en) * | 2018-08-09 | 2018-12-21 | 成都信息工程大学 | Based on the pyramidal 3D CNN nasopharyngeal carcinoma dividing method of Analysis On Multi-scale Features |
CN109087327A (en) * | 2018-07-13 | 2018-12-25 | 天津大学 | A kind of thyroid nodule ultrasonic image division method cascading full convolutional neural networks |
CN109166107A (en) * | 2018-04-28 | 2019-01-08 | 北京市商汤科技开发有限公司 | A kind of medical image cutting method and device, electronic equipment and storage medium |
CN109363699A (en) * | 2018-10-16 | 2019-02-22 | 杭州依图医疗技术有限公司 | A kind of method and device of breast image lesion identification |
CN109447169A (en) * | 2018-11-02 | 2019-03-08 | 北京旷视科技有限公司 | The training method of image processing method and its model, device and electronic system |
CN109447940A (en) * | 2018-08-28 | 2019-03-08 | 天津医科大学肿瘤医院 | Convolutional neural networks training method, ultrasound image recognition positioning method and system |
CN109472801A (en) * | 2018-11-22 | 2019-03-15 | 廖祥 | It is a kind of for multiple dimensioned neuromorphic detection and dividing method |
CN109472791A (en) * | 2018-10-31 | 2019-03-15 | 深圳大学 | Ultrasonic image division method and computer equipment |
CN109583483A (en) * | 2018-11-13 | 2019-04-05 | 中国科学院计算技术研究所 | A kind of object detection method and system based on convolutional neural networks |
CN109584211A (en) * | 2018-10-31 | 2019-04-05 | 南开大学 | A kind of vision automatic testing method of animal oocyte polar body |
CN109615016A (en) * | 2018-12-20 | 2019-04-12 | 北京理工大学 | A kind of object detection method of the convolutional neural networks based on pyramid input gain |
CN109616195A (en) * | 2018-11-28 | 2019-04-12 | 武汉大学人民医院(湖北省人民医院) | The real-time assistant diagnosis system of mediastinum endoscopic ultrasonography image and method based on deep learning |
CN109685143A (en) * | 2018-12-26 | 2019-04-26 | 上海市第十人民医院 | A kind of thyroid gland technetium sweeps the identification model construction method and device of image |
CN109740485A (en) * | 2018-12-27 | 2019-05-10 | 中国水利水电科学研究院 | Reservoir or dyke recognition methods based on spectrum analysis and depth convolutional neural networks |
CN109754403A (en) * | 2018-11-29 | 2019-05-14 | 中国科学院深圳先进技术研究院 | Tumour automatic division method and system in a kind of CT image |
CN110060235A (en) * | 2019-03-27 | 2019-07-26 | 天津大学 | A kind of thyroid nodule ultrasonic image division method based on deep learning |
CN110060774A (en) * | 2019-04-29 | 2019-07-26 | 赵蕾 | A kind of thyroid nodule recognition methods based on production confrontation network |
CN110070546A (en) * | 2019-04-18 | 2019-07-30 | 山东师范大学 | A kind of multiple target based on deep learning jeopardizes the automatic division method of organ, apparatus and system |
CN110110748A (en) * | 2019-03-29 | 2019-08-09 | 广州思德医疗科技有限公司 | A kind of recognition methods of original image and device |
CN110164550A (en) * | 2019-05-22 | 2019-08-23 | 杭州电子科技大学 | A kind of congenital heart disease aided diagnosis method based on multi-angle of view conspiracy relation |
CN110211116A (en) * | 2019-06-03 | 2019-09-06 | 东北大学 | A kind of Thyroid ultrasound image tubercle analysis method based on deep learning network and shallow-layer Texture Feature Fusion |
CN110210543A (en) * | 2019-05-24 | 2019-09-06 | 上海联影智能医疗科技有限公司 | Image classification system, method, equipment and storage medium |
CN110232386A (en) * | 2019-05-09 | 2019-09-13 | 杭州深睿博联科技有限公司 | Based on the pyramidal Lung neoplasm classification method of local feature and device |
CN110235172A (en) * | 2018-06-07 | 2019-09-13 | 深圳迈瑞生物医疗电子股份有限公司 | Image analysis method and ultrasonic image equipment based on ultrasonic image equipment |
CN110288574A (en) * | 2019-06-13 | 2019-09-27 | 南通市传染病防治院(南通市第三人民医院) | A kind of adjuvant Ultrasonographic Diagnosis hepatoncus system and method |
CN110309841A (en) * | 2018-09-28 | 2019-10-08 | 浙江农林大学 | A kind of hickory nut common insect pests recognition methods based on deep learning |
CN110363104A (en) * | 2019-06-24 | 2019-10-22 | 中国科学技术大学 | A kind of detection method of diesel oil black smoke vehicle |
CN110379509A (en) * | 2019-07-23 | 2019-10-25 | 安徽磐众信息科技有限公司 | A kind of Breast Nodules aided diagnosis method and system based on DSSD |
CN110491480A (en) * | 2019-05-22 | 2019-11-22 | 腾讯科技(深圳)有限公司 | A kind of medical image processing method, device, electromedical equipment and storage medium |
CN110516688A (en) * | 2019-08-30 | 2019-11-29 | 北京推想科技有限公司 | The extracting method and system of Lung neoplasm attributive character information |
CN110751651A (en) * | 2019-09-27 | 2020-02-04 | 西安电子科技大学 | MRI pancreas image segmentation method based on multi-scale migration learning |
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 |
CN110880035A (en) * | 2019-11-14 | 2020-03-13 | 北京推想科技有限公司 | Convolutional neural network training method and device and nodule sign identification method and device |
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 |
CN111160413A (en) * | 2019-12-12 | 2020-05-15 | 天津大学 | Thyroid nodule classification method based on multi-scale feature fusion |
CN111179275A (en) * | 2019-12-31 | 2020-05-19 | 电子科技大学 | Medical ultrasonic image segmentation method |
CN111227864A (en) * | 2020-01-12 | 2020-06-05 | 刘涛 | Method and apparatus for lesion detection using ultrasound image using computer vision |
CN111341443A (en) * | 2020-03-09 | 2020-06-26 | 西安中科长青医疗科技研究院有限公司 | Ultrasonic thyroid nodule intelligent evaluation method based on deep learning |
CN111652846A (en) * | 2020-04-30 | 2020-09-11 | 成都数之联科技有限公司 | Semiconductor defect identification method based on characteristic pyramid convolution neural network |
CN111680687A (en) * | 2020-06-09 | 2020-09-18 | 江西理工大学 | Depth fusion model applied to mammary X-ray image anomaly identification and classification method thereof |
CN111724343A (en) * | 2020-05-18 | 2020-09-29 | 天津大学 | Thyroid nodule ultrasonic image data set enhancing method based on antagonistic learning |
CN112437948A (en) * | 2018-07-31 | 2021-03-02 | 奥林巴斯株式会社 | Image diagnosis support system and image diagnosis support device |
CN112508958A (en) * | 2020-12-16 | 2021-03-16 | 桂林电子科技大学 | Lightweight multi-scale biomedical image segmentation method |
CN112529894A (en) * | 2020-12-22 | 2021-03-19 | 徐州医科大学 | Thyroid nodule diagnosis method based on deep learning network |
CN112687391A (en) * | 2021-03-15 | 2021-04-20 | 四川大学 | Construction method of intelligent identification and omission-prevention system for breast ultrasonic image focus |
US10993653B1 (en) | 2018-07-13 | 2021-05-04 | Johnson Thomas | Machine learning based non-invasive diagnosis of thyroid disease |
CN112820399A (en) * | 2021-01-26 | 2021-05-18 | 北京小白世纪网络科技有限公司 | Method and device for automatically diagnosing benign and malignant thyroid nodules |
CN112861876A (en) * | 2021-01-25 | 2021-05-28 | 北京小白世纪网络科技有限公司 | Automatic liver cancer ultrasonic image identification method and device based on convolutional neural network |
CN112903813A (en) * | 2021-01-14 | 2021-06-04 | 北京安铁软件技术有限公司 | Railway track ultrasonic automatic flaw detection method |
CN112927808A (en) * | 2021-03-01 | 2021-06-08 | 北京小白世纪网络科技有限公司 | Thyroid ultrasound image-based nodule grading system and method |
CN112927217A (en) * | 2021-03-23 | 2021-06-08 | 内蒙古大学 | Thyroid nodule invasiveness prediction method based on target detection |
WO2021137755A1 (en) * | 2019-12-30 | 2021-07-08 | Medo Dx Pte. Ltd | Apparatus and method for automated analyses of ultrasound images |
CN113223716A (en) * | 2021-05-20 | 2021-08-06 | 复旦大学附属中山医院 | Method for predicting benign and malignant neck lymph nodes before ablation of minimal thyroid carcinoma |
CN113256704A (en) * | 2021-03-26 | 2021-08-13 | 上海师范大学 | Grain length and width measuring method |
WO2021169452A1 (en) * | 2020-09-27 | 2021-09-02 | 平安科技(深圳)有限公司 | Thyroid nodule image processing method and apparatus, and computer device |
CN114219807A (en) * | 2022-02-22 | 2022-03-22 | 成都爱迦飞诗特科技有限公司 | Mammary gland ultrasonic examination image grading method, device, equipment and storage medium |
CN114419016A (en) * | 2022-01-25 | 2022-04-29 | 华润武钢总医院 | Ultrasonic super-resolution method and system for differential diagnosis of benign and malignant thyroid nodules |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113537357A (en) * | 2021-07-19 | 2021-10-22 | 烟台毓璜顶医院 | Thyroid cancer CT image classification system based on depth residual error network |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103984958A (en) * | 2014-05-07 | 2014-08-13 | 深圳大学 | Method and system for segmenting cervical caner cells |
CN105981041A (en) * | 2014-05-29 | 2016-09-28 | 北京旷视科技有限公司 | Facial landmark localization using coarse-to-fine cascaded neural networks |
CN106056595A (en) * | 2015-11-30 | 2016-10-26 | 浙江德尚韵兴图像科技有限公司 | Method for automatically identifying whether thyroid nodule is benign or malignant based on deep convolutional neural network |
CN106504232A (en) * | 2016-10-14 | 2017-03-15 | 北京网医智捷科技有限公司 | A kind of pulmonary nodule automatic testing method based on 3D convolutional neural networks |
CN106780448A (en) * | 2016-12-05 | 2017-05-31 | 清华大学 | A kind of pernicious sorting technique of ultrasonic Benign Thyroid Nodules based on transfer learning Yu Fusion Features |
US20170243053A1 (en) * | 2016-02-18 | 2017-08-24 | Pinscreen, Inc. | Real-time facial segmentation and performance capture from rgb input |
CN107180426A (en) * | 2017-06-06 | 2017-09-19 | 西北工业大学 | Area of computer aided Lung neoplasm sorting technique based on transportable multiple-model integration |
-
2017
- 2017-10-18 CN CN201710975555.1A patent/CN107680678B/en active Active
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103984958A (en) * | 2014-05-07 | 2014-08-13 | 深圳大学 | Method and system for segmenting cervical caner cells |
CN105981041A (en) * | 2014-05-29 | 2016-09-28 | 北京旷视科技有限公司 | Facial landmark localization using coarse-to-fine cascaded neural networks |
CN106056595A (en) * | 2015-11-30 | 2016-10-26 | 浙江德尚韵兴图像科技有限公司 | Method for automatically identifying whether thyroid nodule is benign or malignant based on deep convolutional neural network |
US20170243053A1 (en) * | 2016-02-18 | 2017-08-24 | Pinscreen, Inc. | Real-time facial segmentation and performance capture from rgb input |
CN106504232A (en) * | 2016-10-14 | 2017-03-15 | 北京网医智捷科技有限公司 | A kind of pulmonary nodule automatic testing method based on 3D convolutional neural networks |
CN106780448A (en) * | 2016-12-05 | 2017-05-31 | 清华大学 | A kind of pernicious sorting technique of ultrasonic Benign Thyroid Nodules based on transfer learning Yu Fusion Features |
CN107180426A (en) * | 2017-06-06 | 2017-09-19 | 西北工业大学 | Area of computer aided Lung neoplasm sorting technique based on transportable multiple-model integration |
Non-Patent Citations (4)
Title |
---|
HAOBIN DOU,XIHONG WU: "Coarse-to-Fine Trained Multi-Scale Convolutional Neural Networks for Image Classification", 《2015 INTERNATIONAL JOINT CONFERENCE ON NEURAL NETWORKS》 * |
JINLIAN MA,FA WU,JIANG ZHU,DONG XU,DEXING KONG: "A pre-trained convolutional neural network based method for thyroid nodule diagnosis", 《ULTRASONICS》 * |
JINLIAN MA,FAWU,TIAN"AN JIANG,JIANG ZHU: "Cascade convolutional neural networks for automatic detection of thyroid nodules in ultrasound images", 《THE INTERNATIONAL JOURNAL OF MEDICAL PHYSICS RESEARCH AND PRACTICE》 * |
张永义: "基于深度神经网络的场景分类方法研究", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
Cited By (101)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108734694A (en) * | 2018-04-09 | 2018-11-02 | 华南农业大学 | Thyroid tumors ultrasonoscopy automatic identifying method based on faster r-cnn |
CN108510495A (en) * | 2018-04-09 | 2018-09-07 | 沈阳东软医疗***有限公司 | A kind of lung image data processing method based on artificial intelligence, apparatus and system |
CN108875906B (en) * | 2018-04-20 | 2019-06-04 | 华侨大学 | A kind of multiple dimensioned convolutional neural networks learning method gradually to add up |
CN108875906A (en) * | 2018-04-20 | 2018-11-23 | 华侨大学 | A kind of multiple dimensioned convolutional neural networks learning method gradually to add up |
CN109166107A (en) * | 2018-04-28 | 2019-01-08 | 北京市商汤科技开发有限公司 | A kind of medical image cutting method and device, electronic equipment and storage medium |
CN108968991A (en) * | 2018-05-08 | 2018-12-11 | 平安科技(深圳)有限公司 | Hand bone X-ray bone age assessment method, apparatus, computer equipment and storage medium |
CN108648192A (en) * | 2018-05-17 | 2018-10-12 | 杭州依图医疗技术有限公司 | A kind of method and device of detection tubercle |
CN108664971B (en) * | 2018-05-22 | 2021-12-14 | 中国科学技术大学 | Pulmonary nodule detection method based on 2D convolutional neural network |
CN108664971A (en) * | 2018-05-22 | 2018-10-16 | 中国科学技术大学 | Pulmonary nodule detection method based on 2D convolutional neural networks |
CN108710919A (en) * | 2018-05-25 | 2018-10-26 | 东南大学 | A kind of crack automation delineation method based on multi-scale feature fusion deep learning |
CN108830891A (en) * | 2018-06-05 | 2018-11-16 | 成都精工华耀科技有限公司 | A kind of rail splice fastener loosening detection method |
CN108830891B (en) * | 2018-06-05 | 2022-01-18 | 成都精工华耀科技有限公司 | Method for detecting looseness of steel rail fishplate fastener |
CN113222966A (en) * | 2018-06-07 | 2021-08-06 | 深圳迈瑞生物医疗电子股份有限公司 | Image analysis method based on ultrasonic imaging equipment and ultrasonic imaging equipment |
CN113222966B (en) * | 2018-06-07 | 2023-01-10 | 深圳迈瑞生物医疗电子股份有限公司 | Image analysis method based on ultrasonic imaging equipment and ultrasonic imaging equipment |
CN110235172A (en) * | 2018-06-07 | 2019-09-13 | 深圳迈瑞生物医疗电子股份有限公司 | Image analysis method and ultrasonic image equipment based on ultrasonic image equipment |
CN110235172B (en) * | 2018-06-07 | 2021-07-20 | 深圳迈瑞生物医疗电子股份有限公司 | Image analysis method based on ultrasonic imaging equipment and ultrasonic imaging equipment |
CN108742679A (en) * | 2018-06-29 | 2018-11-06 | 上海联影医疗科技有限公司 | Nodule detection device and method |
CN109087327B (en) * | 2018-07-13 | 2021-07-06 | 天津大学 | Thyroid nodule ultrasonic image segmentation method of cascaded full convolution neural network |
US10993653B1 (en) | 2018-07-13 | 2021-05-04 | Johnson Thomas | Machine learning based non-invasive diagnosis of thyroid disease |
CN109087327A (en) * | 2018-07-13 | 2018-12-25 | 天津大学 | A kind of thyroid nodule ultrasonic image division method cascading full convolutional neural networks |
CN109003282B (en) * | 2018-07-27 | 2022-04-29 | 京东方科技集团股份有限公司 | Image processing method and device and computer storage medium |
CN109003282A (en) * | 2018-07-27 | 2018-12-14 | 京东方科技集团股份有限公司 | A kind of method, apparatus and computer storage medium of image procossing |
CN112437948A (en) * | 2018-07-31 | 2021-03-02 | 奥林巴斯株式会社 | Image diagnosis support system and image diagnosis support device |
CN109063710A (en) * | 2018-08-09 | 2018-12-21 | 成都信息工程大学 | Based on the pyramidal 3D CNN nasopharyngeal carcinoma dividing method of Analysis On Multi-scale Features |
CN109063710B (en) * | 2018-08-09 | 2022-08-16 | 成都信息工程大学 | 3D CNN nasopharyngeal carcinoma segmentation method based on multi-scale feature pyramid |
CN109447940B (en) * | 2018-08-28 | 2021-09-28 | 天津医科大学肿瘤医院 | Convolutional neural network training method, ultrasonic image identification and positioning method and system |
CN109447940A (en) * | 2018-08-28 | 2019-03-08 | 天津医科大学肿瘤医院 | Convolutional neural networks training method, ultrasound image recognition positioning method and system |
CN110309841A (en) * | 2018-09-28 | 2019-10-08 | 浙江农林大学 | A kind of hickory nut common insect pests recognition methods based on deep learning |
CN109363699A (en) * | 2018-10-16 | 2019-02-22 | 杭州依图医疗技术有限公司 | A kind of method and device of breast image lesion identification |
CN109472791A (en) * | 2018-10-31 | 2019-03-15 | 深圳大学 | Ultrasonic image division method and computer equipment |
CN109472791B (en) * | 2018-10-31 | 2020-10-27 | 深圳大学 | Ultrasonic image segmentation method and computer equipment |
CN109584211A (en) * | 2018-10-31 | 2019-04-05 | 南开大学 | A kind of vision automatic testing method of animal oocyte polar body |
CN109447169B (en) * | 2018-11-02 | 2020-10-27 | 北京旷视科技有限公司 | Image processing method, training method and device of model thereof and electronic system |
CN109447169A (en) * | 2018-11-02 | 2019-03-08 | 北京旷视科技有限公司 | The training method of image processing method and its model, device and electronic system |
CN109583483B (en) * | 2018-11-13 | 2020-12-11 | 中国科学院计算技术研究所 | Target detection method and system based on convolutional neural network |
CN109583483A (en) * | 2018-11-13 | 2019-04-05 | 中国科学院计算技术研究所 | A kind of object detection method and system based on convolutional neural networks |
CN109472801A (en) * | 2018-11-22 | 2019-03-15 | 廖祥 | It is a kind of for multiple dimensioned neuromorphic detection and dividing method |
CN109616195A (en) * | 2018-11-28 | 2019-04-12 | 武汉大学人民医院(湖北省人民医院) | The real-time assistant diagnosis system of mediastinum endoscopic ultrasonography image and method based on deep learning |
CN109754403A (en) * | 2018-11-29 | 2019-05-14 | 中国科学院深圳先进技术研究院 | Tumour automatic division method and system in a kind of CT image |
CN109615016A (en) * | 2018-12-20 | 2019-04-12 | 北京理工大学 | A kind of object detection method of the convolutional neural networks based on pyramid input gain |
CN109615016B (en) * | 2018-12-20 | 2021-06-22 | 北京理工大学 | Target detection method of convolutional neural network based on pyramid input gain |
CN109685143A (en) * | 2018-12-26 | 2019-04-26 | 上海市第十人民医院 | A kind of thyroid gland technetium sweeps the identification model construction method and device of image |
CN109740485A (en) * | 2018-12-27 | 2019-05-10 | 中国水利水电科学研究院 | Reservoir or dyke recognition methods based on spectrum analysis and depth convolutional neural networks |
CN110060235A (en) * | 2019-03-27 | 2019-07-26 | 天津大学 | A kind of thyroid nodule ultrasonic image division method based on deep learning |
CN110110748B (en) * | 2019-03-29 | 2021-08-17 | 广州思德医疗科技有限公司 | Original picture identification method and device |
CN110110748A (en) * | 2019-03-29 | 2019-08-09 | 广州思德医疗科技有限公司 | A kind of recognition methods of original image and device |
CN110070546B (en) * | 2019-04-18 | 2021-08-27 | 山东师范大学 | Automatic multi-target organ-at-risk segmentation method, device and system based on deep learning |
CN110070546A (en) * | 2019-04-18 | 2019-07-30 | 山东师范大学 | A kind of multiple target based on deep learning jeopardizes the automatic division method of organ, apparatus and system |
CN110060774B (en) * | 2019-04-29 | 2022-12-09 | 赵蕾 | Thyroid nodule identification method based on generative confrontation network |
CN110060774A (en) * | 2019-04-29 | 2019-07-26 | 赵蕾 | A kind of thyroid nodule recognition methods based on production confrontation network |
CN110232386A (en) * | 2019-05-09 | 2019-09-13 | 杭州深睿博联科技有限公司 | Based on the pyramidal Lung neoplasm classification method of local feature and device |
CN110232386B (en) * | 2019-05-09 | 2022-04-05 | 杭州深睿博联科技有限公司 | Pulmonary nodule classification method and device based on local feature pyramid |
CN110164550A (en) * | 2019-05-22 | 2019-08-23 | 杭州电子科技大学 | A kind of congenital heart disease aided diagnosis method based on multi-angle of view conspiracy relation |
CN110491480A (en) * | 2019-05-22 | 2019-11-22 | 腾讯科技(深圳)有限公司 | A kind of medical image processing method, device, electromedical equipment and storage medium |
CN110164550B (en) * | 2019-05-22 | 2021-07-09 | 杭州电子科技大学 | Congenital heart disease auxiliary diagnosis method based on multi-view cooperative relationship |
CN110210543B (en) * | 2019-05-24 | 2021-08-17 | 上海联影智能医疗科技有限公司 | Image classification system, method, apparatus and storage medium |
CN110210543A (en) * | 2019-05-24 | 2019-09-06 | 上海联影智能医疗科技有限公司 | Image classification system, method, equipment and storage medium |
CN110211116A (en) * | 2019-06-03 | 2019-09-06 | 东北大学 | A kind of Thyroid ultrasound image tubercle analysis method based on deep learning network and shallow-layer Texture Feature Fusion |
CN110288574A (en) * | 2019-06-13 | 2019-09-27 | 南通市传染病防治院(南通市第三人民医院) | A kind of adjuvant Ultrasonographic Diagnosis hepatoncus system and method |
CN110363104A (en) * | 2019-06-24 | 2019-10-22 | 中国科学技术大学 | A kind of detection method of diesel oil black smoke vehicle |
CN110363104B (en) * | 2019-06-24 | 2022-07-15 | 中国科学技术大学 | Detection method of diesel black smoke vehicle |
CN110379509A (en) * | 2019-07-23 | 2019-10-25 | 安徽磐众信息科技有限公司 | A kind of Breast Nodules aided diagnosis method and system based on DSSD |
CN110516688A (en) * | 2019-08-30 | 2019-11-29 | 北京推想科技有限公司 | The extracting method and system of Lung neoplasm attributive character information |
CN110751651A (en) * | 2019-09-27 | 2020-02-04 | 西安电子科技大学 | MRI pancreas image segmentation method based on multi-scale migration learning |
CN110751651B (en) * | 2019-09-27 | 2022-03-04 | 西安电子科技大学 | MRI pancreas image segmentation method based on multi-scale migration learning |
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 |
CN110880035A (en) * | 2019-11-14 | 2020-03-13 | 北京推想科技有限公司 | Convolutional neural network training method and device and nodule sign identification method and device |
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 |
CN111160413B (en) * | 2019-12-12 | 2023-11-17 | 天津大学 | Thyroid nodule classification method based on multi-scale feature fusion |
CN111160413A (en) * | 2019-12-12 | 2020-05-15 | 天津大学 | Thyroid nodule classification method based on multi-scale feature fusion |
WO2021137755A1 (en) * | 2019-12-30 | 2021-07-08 | Medo Dx Pte. Ltd | Apparatus and method for automated analyses of ultrasound images |
US11721022B2 (en) | 2019-12-30 | 2023-08-08 | Exo Imaging, Inc. | Apparatus and method for automated analyses of ultrasound images |
CN111179275B (en) * | 2019-12-31 | 2023-04-25 | 电子科技大学 | Medical ultrasonic image segmentation method |
CN111179275A (en) * | 2019-12-31 | 2020-05-19 | 电子科技大学 | Medical ultrasonic image segmentation method |
CN111227864B (en) * | 2020-01-12 | 2023-06-09 | 刘涛 | Device for detecting focus by using ultrasonic image and computer vision |
CN111227864A (en) * | 2020-01-12 | 2020-06-05 | 刘涛 | Method and apparatus for lesion detection using ultrasound image using computer vision |
CN111341443A (en) * | 2020-03-09 | 2020-06-26 | 西安中科长青医疗科技研究院有限公司 | Ultrasonic thyroid nodule intelligent evaluation method based on deep learning |
CN111652846A (en) * | 2020-04-30 | 2020-09-11 | 成都数之联科技有限公司 | Semiconductor defect identification method based on characteristic pyramid convolution neural network |
CN111652846B (en) * | 2020-04-30 | 2022-08-16 | 成都数之联科技股份有限公司 | Semiconductor defect identification method based on characteristic pyramid convolution neural network |
CN111724343A (en) * | 2020-05-18 | 2020-09-29 | 天津大学 | Thyroid nodule ultrasonic image data set enhancing method based on antagonistic learning |
CN111680687A (en) * | 2020-06-09 | 2020-09-18 | 江西理工大学 | Depth fusion model applied to mammary X-ray image anomaly identification and classification method thereof |
CN111680687B (en) * | 2020-06-09 | 2022-05-10 | 江西理工大学 | Depth fusion classification method applied to mammary X-ray image anomaly identification |
WO2021169452A1 (en) * | 2020-09-27 | 2021-09-02 | 平安科技(深圳)有限公司 | Thyroid nodule image processing method and apparatus, and computer device |
CN112508958A (en) * | 2020-12-16 | 2021-03-16 | 桂林电子科技大学 | Lightweight multi-scale biomedical image segmentation method |
CN112508958B (en) * | 2020-12-16 | 2022-07-19 | 桂林电子科技大学 | Lightweight multi-scale biomedical image segmentation method |
CN112529894A (en) * | 2020-12-22 | 2021-03-19 | 徐州医科大学 | Thyroid nodule diagnosis method based on deep learning network |
CN112903813B (en) * | 2021-01-14 | 2023-03-14 | 北京安铁软件技术有限公司 | Railway track ultrasonic automatic flaw detection method |
CN112903813A (en) * | 2021-01-14 | 2021-06-04 | 北京安铁软件技术有限公司 | Railway track ultrasonic automatic flaw detection method |
CN112861876A (en) * | 2021-01-25 | 2021-05-28 | 北京小白世纪网络科技有限公司 | Automatic liver cancer ultrasonic image identification method and device based on convolutional neural network |
CN112820399A (en) * | 2021-01-26 | 2021-05-18 | 北京小白世纪网络科技有限公司 | Method and device for automatically diagnosing benign and malignant thyroid nodules |
CN112927808A (en) * | 2021-03-01 | 2021-06-08 | 北京小白世纪网络科技有限公司 | Thyroid ultrasound image-based nodule grading system and method |
CN112687391A (en) * | 2021-03-15 | 2021-04-20 | 四川大学 | Construction method of intelligent identification and omission-prevention system for breast ultrasonic image focus |
CN112927217B (en) * | 2021-03-23 | 2022-05-03 | 内蒙古大学 | Thyroid nodule invasiveness prediction method based on target detection |
CN112927217A (en) * | 2021-03-23 | 2021-06-08 | 内蒙古大学 | Thyroid nodule invasiveness prediction method based on target detection |
CN113256704A (en) * | 2021-03-26 | 2021-08-13 | 上海师范大学 | Grain length and width measuring method |
CN113256704B (en) * | 2021-03-26 | 2024-04-05 | 上海师范大学 | Grain length and width measuring method |
CN113223716A (en) * | 2021-05-20 | 2021-08-06 | 复旦大学附属中山医院 | Method for predicting benign and malignant neck lymph nodes before ablation of minimal thyroid carcinoma |
CN114419016A (en) * | 2022-01-25 | 2022-04-29 | 华润武钢总医院 | Ultrasonic super-resolution method and system for differential diagnosis of benign and malignant thyroid nodules |
CN114419016B (en) * | 2022-01-25 | 2022-11-04 | 华润武钢总医院 | Ultrasonic super-resolution method and system for thyroid nodule benign and malignant differential diagnosis |
CN114219807B (en) * | 2022-02-22 | 2022-07-12 | 成都爱迦飞诗特科技有限公司 | Mammary gland ultrasonic examination image grading method, device, equipment and storage medium |
CN114219807A (en) * | 2022-02-22 | 2022-03-22 | 成都爱迦飞诗特科技有限公司 | Mammary gland ultrasonic examination image grading method, device, equipment and storage medium |
Also Published As
Publication number | Publication date |
---|---|
CN107680678B (en) | 2020-12-01 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107680678A (en) | Based on multiple dimensioned convolutional neural networks Thyroid ultrasound image tubercle auto-check system | |
CN109447940B (en) | Convolutional neural network training method, ultrasonic image identification and positioning method and system | |
CN108257135A (en) | The assistant diagnosis system of medical image features is understood based on deep learning method | |
CN107977671A (en) | A kind of tongue picture sorting technique based on multitask convolutional neural networks | |
Huang et al. | A multidirectional and multiscale morphological index for automatic building extraction from multispectral GeoEye-1 imagery | |
CN108765408A (en) | Build the method in cancer pathology image virtual case library and the multiple dimensioned cancer detection system based on convolutional neural networks | |
CN110942446A (en) | Pulmonary nodule automatic detection method based on CT image | |
CN107280697A (en) | Lung neoplasm grading determination method and system based on deep learning and data fusion | |
CN107346420A (en) | Text detection localization method under a kind of natural scene based on deep learning | |
CN110018524A (en) | A kind of X-ray safety check contraband recognition methods of view-based access control model-attribute | |
CN106096654A (en) | A kind of cell atypia automatic grading method tactful based on degree of depth study and combination | |
CN106462771A (en) | 3D image significance detection method | |
CN106529499A (en) | Fourier descriptor and gait energy image fusion feature-based gait identification method | |
CN106408030A (en) | SAR image classification method based on middle lamella semantic attribute and convolution neural network | |
CN110021425A (en) | A kind of relatively detector and its construction method and cervical cancer cell detection method | |
CN108053412B (en) | Method for determining optimal segmentation result of remote sensing image by fusing multiple quantitative evaluation indexes | |
Ren et al. | The accuracy of LCZ maps generated by the world urban database and access portal tools (WUDAPT) method: A case study of Hong Kong | |
CN101551854B (en) | A processing system of unbalanced medical image and processing method thereof | |
CN101055631A (en) | Space data fuzzy evidence weight analysis method | |
CN109102498A (en) | A kind of method of cluster type nucleus segmentation in cervical smear image | |
Cao et al. | An automatic breast cancer grading method in histopathological images based on pixel-, object-, and semantic-level features | |
CN109522831A (en) | A kind of vehicle real-time detection method based on micro- convolutional neural networks | |
Zhang et al. | Impervious surface extraction from high-resolution satellite image using pixel-and object-based hybrid analysis | |
CN114240874A (en) | Bone age assessment method and device based on deep convolutional neural network and feature fusion and computer readable storage medium | |
CN109671055A (en) | Pulmonary nodule detection method and device |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |