CN109087327A - A kind of thyroid nodule ultrasonic image division method cascading full convolutional neural networks - Google Patents
A kind of thyroid nodule ultrasonic image division method cascading full convolutional neural networks Download PDFInfo
- Publication number
- CN109087327A CN109087327A CN201810772176.7A CN201810772176A CN109087327A CN 109087327 A CN109087327 A CN 109087327A CN 201810772176 A CN201810772176 A CN 201810772176A CN 109087327 A CN109087327 A CN 109087327A
- Authority
- CN
- China
- Prior art keywords
- sampling
- neural networks
- layer
- convolutional neural
- full convolutional
- 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
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/181—Segmentation; Edge detection involving edge growing; involving edge linking
Abstract
The invention discloses a kind of thyroid nodule ultrasonic image division methods for cascading full convolutional neural networks, it include: simple full convolutional neural networks of the building based on U-Net, the ultrasound image in Thyroid ultrasound data is split according to simple full convolutional neural networks, is therefrom partitioned into area-of-interest;The further feature for extracting area-of-interest as down-sampling layer using VGG19-FCN network, realizes the automatic semantic segmentation to thyroid nodule with this;Wherein, the simple full convolutional neural networks include: five convolutional layers for down-sampling and five up-sampling layers for up-sampling;Wherein, first five time convolution conv is made of the convolutional layer and a pond layer of two 3x3, and each convolutional layer uses ReLU as activation primitive, and rear five up-samplings layer is warp lamination.The present invention provides high-precision nodule image for the pernicious identification of Benign Thyroid Nodules, to play better booster action in terms of medical diagnosis.
Description
Technical field
The present invention relates to technical field of image processing and medicine auxiliary diagnosis fields more particularly to a kind of image, semantic to divide
Method is in particular to partitioned into the method for the tubercle part of Thyroid ultrasound image automatically based on full convolutional neural networks.
Background technique
Thyroid nodule is a kind of now generally existing disease, has investigation to point out the generation of the thyroid nodule in crowd
Rate nearly 50%, but the thyroid nodule of only 4%-8% can be accessible in physical palpation.Thyroid nodule has good, pernicious
Point, pernicious incidence be 5%-10%.Early detection lesion has its good pernicious, clinical treatment of identification and surgical selection important
Meaning.Thyroid nodule ultrasonic examination based on ultrasonic imaging technique is test mode common at present, however diagnosis is super
The result of sound thyroid gland image suffers from the shadow of the factors such as the imaging mechanism of medical imaging devices, acquisition condition, display equipment
It rings and easily causes mistaken diagnosis or fail to pinpoint a disease in diagnosis.Therefore, realize that the diagnosis of thyroid gland visual aids is very necessary using computer.But it is intrinsic
Image-forming mechanism ultrasonic thyroid nodule picture quality that clinical acquisitions are arrived it is poor, cause auxiliary diagnosis accuracy and
Automation is affected, and the mode of the common semi-automatic segmentation thyroid nodule based on active contour is divided accurately at present
Property is lower, will cause biggish interference to the good pernicious classification of neural network.
Medical image segmentation is always the hot spot application in image, semantic segmentation field.Traditional medical image method is mainly
Method based on level set.2014, this method was applied to heart nuclear magnetic resonance (Nuclear by Chunming Li et al. people
Magnetic Resonance, NMR) image segmentation and body of gland dye image.However Thyroid ultrasound image due to contrast is low,
Make this method super in application Thyreoidine there are speckle echo, thyroid nodule obscurity boundary and the features such as having calcification point shade
Accuracy rate is lower when dividing tubercle in acoustic image.In recent years, with the rapid development of deep learning, convolutional neural networks are in image
All considerably beyond conventional method in the accuracy rate and efficiency in segmentation field.Jinlian Ma et al. is for the first time by convolutional neural networks
Applied to the nodule segmentation of Thyroid ultrasound image, and achieve higher accuracy rate.Classical convolutional neural networks
(Convolutional Neural Network, CNN) obtains the feature of regular length using full articulamentum after convolutional layer
Vector is classified, and full articulamentum is then replaced with convolution by the full convolutional neural networks (FCN) that Jonathan Long et al. is proposed
Layer, and the characteristic pattern of the last one convolutional layer is up-sampled using warp lamination, so that it is restored to original image size, thus
One prediction is generated to each pixel, while also remaining the spatial information in original image, finally in the characteristic pattern of up-sampling
It is upper to be classified pixel-by-pixel.FCN is applied to osteosarcoma CT image segmentation by Huang Lin et al..The stacking of LeiBi et al. is rolled up entirely
Product neural network improves the performance of medical image segmentation, takes on chest X-ray, cardiac ultrasound images and histology picture
Obtained good effect.Ronneberger obtains ISBI cell tracking based on the FCN U-net model proposed
The champion of challenge 2015.Obviously full convolutional neural networks have huge potentiality in medical image segmentation field.
Summary of the invention
Based on problem above, it is a primary object of the present invention to overcome deficiency in the prior art, it is complete to provide a kind of cascade
The thyroid nodule ultrasonic image division method of convolutional neural networks provides high-precision for the pernicious identification of Benign Thyroid Nodules
Nodule image, to play better booster action in terms of medical diagnosis.
A kind of thyroid nodule ultrasonic image division method cascading full convolutional neural networks, the method includes following steps
It is rapid:
The simple full convolutional neural networks based on U-Net are constructed, according to simple full convolutional neural networks to Thyroid ultrasound
Ultrasound image in data is split, and is therefrom partitioned into area-of-interest;
The further feature for being extracted area-of-interest as down-sampling layer using VGG19-FCN network, is realized with this to first shape
The automatic semantic segmentation of gland tubercle;
Wherein, the simple full convolutional neural networks include:
Five convolutional layers for down-sampling and five up-sampling layers for up-sampling;
Wherein, first five time convolution conv is made of the convolutional layer and a pond layer of two 3x3, and each convolutional layer makes
Use ReLU as activation primitive, rear five up-samplings layer is warp lamination;
The simple full convolutional neural networks carry out down-sampling to input picture using five convolution sum ponds and extract deep layer
Feature, for carrying out the pixel classifications of neural network to different classes of pixel;Again by the characteristic pattern of each pond down-sampling according to
The secondary characteristic pattern with up-sampling carries out interpolation, then supplements during up-sampling characteristic pattern is gradually reverted to original image resolution ratio
Shallow-layer detailed information.
The full articulamentum of last three layers of the VGG19-FCN network is 7 × 7 × 4096,1 × 1 × 4096 and 1 × 1 × 2
Convolutional layer, being is two classification problems an of pixel due to thyroid nodule segmentation, and the last layer convolution depth is 2,.
The VGG19-FCN network passes through 3 deconvolution since the 5th layer gradually to up-sampling 3 times, wherein preceding divide twice
Interpolation is not carried out with pond layer pool4 and pool5 to supplement the detailed information lost during down-sampling, is adopted in last time
Sample reverts to original image resolution ratio, obtains mask image, draws out the edge of tubercle in the roi by mask image.
Input of the output of the simple full convolutional neural networks as the VGG19-FCN network, described simple complete
Convolutional neural networks to training set test set make segmentation after through screening be exactly the VGG19-FCN network training set and survey
Examination collection.
The beneficial effect of the technical scheme provided by the present invention is that:
1, the present invention passes through first part using full convolutional neural networks are cascaded to thyroid nodule progress semantic segmentation
Neural network extracts ROI, carries out thyroid nodule segmentation to ROI by the neural network of second part;
2, the present invention can not only be partitioned into thyroid nodule automatically, and segmentation precision is higher, can be used as doctor couple
The reference of the good pernicious diagnosis of tubercle, pernicious also important in inhibiting good to machine recognition tubercle.
Detailed description of the invention
Fig. 1 is to divide thyroid nodule flow chart based on full convolutional neural networks;
Fig. 2 is the schematic network structure of the full convolutional neural networks of automatic segmentation thyroid nodule;
Fig. 3 is the schematic diagram of Thyroid ultrasound image original image;
Fig. 4 is the schematic diagram using the UNET ROI region being partitioned into;Fig. 5 is the schematic diagram for the tubercle part that expert draws;
Fig. 6 is the schematic diagram for being partitioned into knuckle areas automatically using full convolutional neural networks.
Specific embodiment
To make the object, technical solutions and advantages of the present invention clearer, embodiment of the present invention is made below further
Ground detailed description.
Embodiment 1
A kind of thyroid nodule ultrasonic image division method cascading full convolutional neural networks, referring to Fig. 1, this method includes
Following steps:
101: simple full convolutional neural networks of the building based on U-Net, according to simple full convolutional neural networks to thyroid gland
Ultrasound image in ultrasound data is split, and is therefrom partitioned into area-of-interest;
102: extracting the further feature of area-of-interest as down-sampling layer using VGG19-FCN network, realized pair with this
The automatic semantic segmentation of thyroid nodule;
Wherein, the simple full convolutional neural networks in step 102 include:
Five convolutional layers for down-sampling and five up-sampling layers for up-sampling;
Wherein, first five time convolution conv is made of the convolutional layer and a pond layer of two 3x3, and each convolutional layer makes
Use ReLU as activation primitive, rear five up-samplings layer is warp lamination;
Further, simple full convolutional neural networks carry out down-sampling extraction to input picture using five convolution sum ponds
Further feature, for carrying out the pixel classifications of neural network to different classes of pixel;Again by the feature of each pond down-sampling
Figure successively in up-sampling characteristic pattern carry out interpolation, then will up-sampling characteristic pattern gradually revert to original image resolution ratio during
Supplement shallow-layer detailed information.
Wherein, last three layers of the VGG19 network in step 102 full articulamentum are 7 × 7 × 4096,1 × 1 × 4096 and 1
× 1 × 2 convolutional layer is since thyroid nodule segmentation is two classification problems an of pixel, and the last layer convolution depth is
2,.
The VGG19-FCN network passes through 3 deconvolution since the 5th layer gradually to up-sampling 3 times, wherein preceding divide twice
Interpolation is not carried out with pond layer pool4 and pool5 to supplement the detailed information lost during down-sampling, is adopted in last time
Sample reverts to original image resolution ratio, obtains mask image, draws out the edge of tubercle in the roi by mask image.
Input of the output of above-mentioned simple full convolutional neural networks as VGG19-FCN network, in simple full convolutional Neural
Network to training set test set make segmentation after through screening be exactly the VGG19-FCN network training set and test set.
In conclusion the embodiment of the invention provides a kind of thyroid nodule ultrasound images for cascading full convolutional neural networks
Dividing method provides high-precision nodule image by this method for the pernicious identification of Benign Thyroid Nodules, to examine in medicine
Disconnected aspect plays better booster action.
Embodiment 2
To achieve the above object, the technical thought of the embodiment of the present invention is: reading Thyroid ultrasound data, passes through simple U
The full convolutional neural networks of type remove the background area in ultrasound image, building training, test sample data set, and building depth is complete
Convolutional neural networks, i.e. FCN, are trained data set, obtain being suitble to the semantic segmentation of ultrasound image as a result, specific steps packet
It includes:
201: reading Thyroid ultrasound data, remove ultrasound image by the simple full convolutional neural networks based on U-Net
Background area:
Wherein, ultrasound image is made of region of interest (ROI) and background area, and ROI includes important diagnostic message, and is carried on the back
The a large amount of highlighted letters and symbol that scene area includes, easily cause interference when neural network extracts knuckle areas.It is above-mentioned to solve
Problem, the embodiment of the present invention carry out ROI extraction to ultrasound image first.
ROI is rectangular area, is located substantially at the near center location of whole ultrasound image, and area accounting is larger.It is different
Though the diasonograph of model take the ROI of ultrasound image length-width ratio, tone, in terms of have any different, and background area
The difference in domain be it is similar, such difference is advantageous the study of neural network.The embodiment of the present invention proposes that one kind is based on
The simple full convolutional neural networks of U-Net are split ROI, and network structure is as shown in table 1:
The simple full convolutional neural networks of table 1
From table 1, it can be seen that simply full convolutional neural networks structure includes: that five convolution (are used to down adopt to the U-Net
Sample) and five up-sampling layers (for up-sampling).Wherein, first five time convolution (conv) by two 3x3 convolutional layer
(conv1_1, conv1_2, conv4_1 and conv4_2 etc., such as: 3 × 3 × 64,3 × 3 × 128 etc.) and a pond layer
(pool1, pool2 and pool5 etc., such as: it 112 × 112 × 128 etc.) forms, each convolutional layer (conv) uses ReLU
As activation primitive, rear five up-sampling layers (decon) are warp lamination.
Down-sampling is carried out to input picture using five convolution sum ponds and extracts further feature, for different classes of picture
Element carries out the pixel classifications of neural network;The characteristic pattern of each pond down-sampling is successively carried out with the characteristic pattern in up-sampling again
Interpolation, then shallow-layer detailed information is supplemented during up-sampling characteristic pattern is gradually reverted to original image resolution ratio.
The embodiment of the present invention carries out down-sampling to input picture using five convolution sum ponds and extracts further feature.Convolution fortune
The essence of calculation is to do dot product in the regional area of filter (convolution kernel) and input data, as shown in Equation 1:
Wherein, y [m, n] is the local feature figure matrix of output;X [m, n] is electric-wave filter matrix;H [m, n] is input picture
Regional area matrix;X [i, j] is the electric-wave filter matrix in current location;M is electric-wave filter matrix width;N is electric-wave filter matrix
Highly;I is the width subscript of current location, and value range is 0 to m;J is the height subscript of current location, and value range arrives for 0
n。
Convolution kernel slides on the original image, obtains the characteristic pattern of original image.The depth of characteristic pattern corresponds to convolution kernel
Number indicates original image inhomogeneity another characteristic.(wherein, convolution receptive field is that this field is public for convolution kernel size and convolution receptive field
The technical term known, the embodiment of the present invention do not repeat them here this) quantity of size and parameter is positively correlated, and big convolution kernel can be with
More fully feature is extracted, but also containing more parameters simultaneously leads to the training speed and computational efficiency drop of model
It is low.
To solve the above problems, the embodiment of the present invention replaces 5 × 5 convolution kernel using two continuous 3 × 3 convolution kernels
(referring to table 1) reduces the complexity of convolution kernel in the case where guaranteeing that convolution receptive field size is constant, promotes computational efficiency.
The essence in pond is sampling, for the characteristic pattern of input, a kind of algorithm is selected to compress it, maximum value pond
It is common one of pond algorithm, as shown in formula (2):
Y=Max (a, b, c, d) (2)
Wherein, a, b, c, d are four positions of the corresponding regional area of pond matrix of 2 × 2 sizes.
Pondization can effectively reduce number of parameters, to reduce while increasing the convolution receptive field of subsequent convolutional layer
The complexity of model.When micro-displacement occurs in neighborhood for the pixel of input picture, the output of pond layer remains to remain unchanged, because
This pondization also has certain disturbance rejection effect, and the robustness of network can be improved.
After 5 continuous down-samplings (i.e. five convolution sum ponds), the resolution ratio of characteristic pattern is lower with respect to original image, can not
The classification of pixel is restored in original image, the embodiment of the present invention is gradually gone back characteristic pattern by 5 up-samplings (i.e. up-sampling layer)
Original is to original image resolution ratio to solve the above problems.In reduction process, the embodiment of the present invention is by the feature of each pond down-sampling
Figure successively carries out interpolation, the detailed information lost during supplement pondization with up-sampling characteristic pattern.
202: thyroid nodule in ROI is split.
The ROI region that first part's network pruning goes out only includes a small amount of interference and thyroid gland areas.In Thyroid ultrasound figure
Tubercle is not of uniform size as in, and most obscurity boundary.Therefore the segmentation at thyroid nodule edge has certain challenge.
In face of the above problem, the embodiment of the present invention chooses the deep layer that the deeper VGG19 of depth extracts ROI as down-sampling layer
Feature, and respectively corresponded in upper sampling process with down-sampling pond layer and do interpolation acquisition shallow-layer feature, Thyreoidine is used in building
The full convolutional neural networks of ultrasound image, network structure are as shown in table 2:
Full convolutional neural networks of 2 VGG19 of table as down-sampling layer
From table 2, it can be seen that the full convolutional neural networks structure of the depth includes: preceding convolution twice by two convolutional layers
(conv1_1, conv1_2, conv2_1 and conv2_2) and pond layer (pool1 and pool2) composition, next
Cubic convolution by four convolutional layers (conv3_1, conv3_2, conv3_3, conv3_4, conv5_1 and conv5_4 etc.)
It is formed with a pond layer (pool3, pool4, pool5), each convolutional layer selects ReLU as activation primitive;Finally three times
Up-sampling is warp lamination (deconv1, deconv2 and deconv3).
Down-sampling based on VGG19 model is divided into five convolution, wherein preceding convolution twice uses 23 × 3 convolution kernels respectively
Instead of 5 × 5 convolution kernel, cubic convolution replaces 11 × 11 convolution kernels using 43 × 3 convolution kernels later.With the convolution number of plies
Increase, in the case where guaranteeing down-sampling convolution receptive field, required parameter is less, and model complexity is greatly reduced.Multilayer
Convolution also brings more activation primitives, and decision function is made more to have resolving ability, more to different classes of separating capacity
By force.
Last three layers of VGG19 full articulamentum are replaced with 7 × 7 × 4096,1 × 1 × 4096 and 1 × 1 × 2 by present networks
Convolutional layer, the last layer convolution depth are 2 be since whether two classification problems that thyroid nodule segmentation is a pixel (are
Tubercle pixel).Present networks by 3 deconvolution since the 5th layer ing gradually to up-sampling 3 times, wherein it is preceding twice respectively with pool4
Interpolation is carried out with pool5 to supplement the detailed information lost during down-sampling, and last time up-sampling reverts to original image point
Resolution obtains mask image, draws out the edge of tubercle in the roi by mask image.
Wherein, the connection of first part and second part neural network is using the output of a upper neural network as next
The input of a neural network, a upper neural network training set test set is made segmentation after by screening be exactly next mind
Training set and test set through network.
Embodiment 3
The scheme in Examples 1 and 2 is further introduced below with reference to Fig. 1-Fig. 3, described below:
1) ultrasound image data for reading thyroid nodule, can be various types of picture formats:
2) all pictures in training set are first read in, the label data image including ultrasound image Yu labeled ROI trains
The model (i.e. first part's neural network) of automatic segmentation ROI based on UNET;
3) using the ROI being partitioned into and the lower thyroid nodule label data image marked of expert's guidance as input, training
The model for being divided thyroid nodule automatically based on VGG19-FCN out is constituted ROI parted pattern and nodule segmentation Cascade certainly
Then the medical assist of dynamic segmentation thyroid nodule ultrasound image is read in test set data and is tested.Use this method
When automatic segmentation thyroid nodule, it is only necessary to read in all Thyroid ultrasound images to be divided.
4) carry out ROI extraction to Thyroid ultrasound image: (i.e. original image is as schemed for the Thyroid ultrasound image that step 1) is read
Shown in 3) it is input in ROI extraction model, obtained mask is found maximal margin, according to most by output masking in original image
Big edge is split original image, obtain it is as shown in Figure 4 only include thyroid gland part ROI image.
5) selecting step 4) processed ROI image 3700 opens, wherein 3000 are used as training set, another 700 as surveying
Examination collection.Tubercle part and non-nodules part are intercepted out by expert, then pass through the full convolution depth using VGG19 as down-sampling layer
Neural network (referring to fig. 2) trains the model divided automatically and preservation, the ROI image input tubercle point that step 4) is extracted
It cuts in model, obtained mask is found into maximal margin in original image, in the roi goes out nodule segmentation according to maximal margin
Come.
As a result show: Fig. 4 is the ROI image for only including thyroid gland part, and Fig. 5 is the thyroid nodule portion of expert's label
Point, Fig. 6 is the thyroid nodule partial schematic diagram that this method is divided automatically.As can be seen that ROI parted pattern can be by the portion ROI
Complete parttion is divided to come out, the tubercle of nodule segmentation model segmentation divides similarity very high with the tuberal part that expert marks, only at edge
Details have difference.
Finally, the embodiment of the present invention uses the mode of cascade neural network, the full convolutional Neural net of simple UNET is used first
Network is partitioned into ROI region, then uses VGG19-FCN (FCN i.e. using VGG19 as down-sampling layer) to carry out in ROI region again
Semantic segmentation substantially increases the accuracy rate of nodule segmentation;The embodiment of the present invention is using the full convolutional neural networks of cascade to first shape
Gland ultrasound image is split, and is had to medical diagnosis processes such as the position judgement of thyroid nodule, good pernicious identifications heuristic
Help.
The embodiment of the present invention to the model of each device in addition to doing specified otherwise, the model of other devices with no restrictions,
As long as the device of above-mentioned function can be completed.
It will be appreciated by those skilled in the art that attached drawing is the schematic diagram of a preferred embodiment, the embodiments of the present invention
Serial number is for illustration only, does not represent the advantages or disadvantages of the embodiments.
The foregoing is merely presently preferred embodiments of the present invention, is not intended to limit the invention, it is all in spirit of the invention and
Within principle, any modification, equivalent replacement, improvement and so on be should all be included in the protection scope of the present invention.
Claims (3)
1. a kind of thyroid nodule ultrasonic image division method for cascading full convolutional neural networks, which is characterized in that the method
The following steps are included:
The simple full convolutional neural networks based on U-Net are constructed, according to simple full convolutional neural networks to Thyroid ultrasound data
In ultrasound image carry out semantic segmentation, be therefrom partitioned into area-of-interest;
The further feature for being extracted area-of-interest as down-sampling layer using VGG19 network, is realized with this to thyroid nodule
Automatic semantic segmentation;
Wherein, the simple full convolutional neural networks include:
Five convolutional layers for down-sampling and five up-sampling layers for up-sampling;
Wherein, first five time convolution conv is made of the convolutional layer and a pond layer of two 3x3, and each convolutional layer uses
For ReLU as activation primitive, rear five up-samplings layer is warp lamination;
The simple full convolutional neural networks carry out down-sampling to input picture using five convolution sum ponds and extract further feature,
For carrying out the pixel classifications of neural network to different classes of pixel;Again by the characteristic pattern of each pond down-sampling successively with it is upper
Characteristic pattern in sampling carries out interpolation, then will up-sampling characteristic pattern gradually revert to original image resolution ratio during to supplement shallow-layer thin
Save information.
2. a kind of thyroid nodule ultrasonic image division method for cascading full convolutional neural networks according to claim 1,
It is characterized in that,
The convolution that the full articulamentum of last three layers of the VGG19-FCN network is 7 × 7 × 4096,1 × 1 × 4096 and 1 × 1 × 2
Layer, being is two classification problems an of pixel due to thyroid nodule segmentation, and the last layer convolution depth is 2,.
The VGG19-FCN network by 3 deconvolution since the 5th layer ing gradually to up-sampling 3 times, wherein it is preceding twice respectively with
Pond layer pool4 and pool5 carry out interpolation to supplement the detailed information lost during down-sampling, and last time up-sampling is also
Original arrives original image resolution ratio, obtains mask image, draws out the edge of tubercle in the roi by mask image.
3. a kind of thyroid nodule Ultrasound Image Segmentation side for cascading full convolutional neural networks according to claim 1 or 2
Method, which is characterized in that
Input of the output of the simple full convolutional neural networks as the VGG19-FCN network, in the simple full convolution
Neural network to training set test set make segmentation after through screening be exactly the VGG19-FCN network training set and test
Collection.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810772176.7A CN109087327B (en) | 2018-07-13 | 2018-07-13 | Thyroid nodule ultrasonic image segmentation method of cascaded full convolution neural network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810772176.7A CN109087327B (en) | 2018-07-13 | 2018-07-13 | Thyroid nodule ultrasonic image segmentation method of cascaded full convolution neural network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109087327A true CN109087327A (en) | 2018-12-25 |
CN109087327B CN109087327B (en) | 2021-07-06 |
Family
ID=64837883
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810772176.7A Active CN109087327B (en) | 2018-07-13 | 2018-07-13 | Thyroid nodule ultrasonic image segmentation method of cascaded full convolution neural network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109087327B (en) |
Cited By (32)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109919098A (en) * | 2019-03-08 | 2019-06-21 | 广州视源电子科技股份有限公司 | The recognition methods of target object and device |
CN109949318A (en) * | 2019-03-07 | 2019-06-28 | 西安电子科技大学 | Full convolutional neural networks epileptic focus dividing method based on multi-modal 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 |
CN110074813A (en) * | 2019-04-26 | 2019-08-02 | 深圳大学 | A kind of ultrasonic image reconstruction method and system |
CN110110808A (en) * | 2019-05-16 | 2019-08-09 | 京东方科技集团股份有限公司 | A kind of pair of image carries out the method, apparatus and computer readable medium of target mark |
CN110136809A (en) * | 2019-05-22 | 2019-08-16 | 腾讯科技(深圳)有限公司 | A kind of medical image processing method, device, electromedical equipment and storage medium |
CN110211134A (en) * | 2019-05-30 | 2019-09-06 | 上海商汤智能科技有限公司 | A kind of image partition method and device, electronic equipment and storage medium |
CN110232378A (en) * | 2019-05-30 | 2019-09-13 | 苏宁易购集团股份有限公司 | A kind of image interest point detecting method, system and readable storage medium storing program for executing |
CN110610487A (en) * | 2019-08-29 | 2019-12-24 | 上海杏脉信息科技有限公司 | Method and device for measuring increase of heart shadow |
CN110706793A (en) * | 2019-09-25 | 2020-01-17 | 天津大学 | Attention mechanism-based thyroid nodule semi-supervised segmentation method |
CN110992306A (en) * | 2019-11-04 | 2020-04-10 | 国网河北省电力有限公司检修分公司 | Method and device for segmenting deteriorated insulator in infrared image based on deep learning |
CN111047598A (en) * | 2019-11-04 | 2020-04-21 | 华北电力大学(保定) | Deep learning-based ultraviolet discharge light spot segmentation method and device for power transmission and transformation equipment |
CN111105410A (en) * | 2019-12-27 | 2020-05-05 | 中国人民解放军陆军军医大学第二附属医院 | Hematopoietic tissue proportion determining device and method based on bone marrow biopsy image |
CN111161273A (en) * | 2019-12-31 | 2020-05-15 | 电子科技大学 | Medical ultrasonic image segmentation method based on deep learning |
CN111210439A (en) * | 2019-12-26 | 2020-05-29 | 中国地质大学(武汉) | Semantic segmentation method and device by suppressing uninteresting information and storage device |
CN111223092A (en) * | 2020-02-28 | 2020-06-02 | 长沙大端信息科技有限公司 | Automatic quality control system and detection method for ultrasonic sectional images of fetus |
CN111242956A (en) * | 2020-01-09 | 2020-06-05 | 西北工业大学 | U-Net-based ultrasonic fetal heart and fetal lung deep learning joint segmentation method |
CN111292324A (en) * | 2020-03-20 | 2020-06-16 | 电子科技大学 | Multi-target identification method and system for brachial plexus ultrasonic image |
CN111311561A (en) * | 2020-02-10 | 2020-06-19 | 浙江未来技术研究院(嘉兴) | Automatic operation area photometry method and device based on microsurgery imaging system |
WO2020173163A1 (en) * | 2019-02-26 | 2020-09-03 | 电信科学技术研究院有限公司 | Image segmentation method and apparatus |
CN111626994A (en) * | 2020-05-18 | 2020-09-04 | 江苏远望仪器集团有限公司 | Equipment fault defect diagnosis method based on improved U-Net neural network |
CN111724397A (en) * | 2020-06-18 | 2020-09-29 | 上海应用技术大学 | Automatic segmentation method for bleeding area of craniocerebral CT image |
CN111738338A (en) * | 2020-06-23 | 2020-10-02 | 征图新视(江苏)科技股份有限公司 | Defect detection method applied to motor coil based on cascaded expansion FCN network |
CN111798455A (en) * | 2019-09-25 | 2020-10-20 | 天津大学 | Thyroid nodule real-time segmentation method based on full convolution dense cavity network |
CN112085028A (en) * | 2020-08-31 | 2020-12-15 | 浙江工业大学 | Tooth panoramic semantic segmentation method based on feature map disturbance and boundary supervision |
CN112085729A (en) * | 2020-09-18 | 2020-12-15 | 无锡祥生医疗科技股份有限公司 | Pleural line region extraction method, storage medium, and ultrasound diagnostic apparatus |
CN112529894A (en) * | 2020-12-22 | 2021-03-19 | 徐州医科大学 | Thyroid nodule diagnosis method based on deep learning network |
CN113034373A (en) * | 2019-12-24 | 2021-06-25 | 无锡祥生医疗科技股份有限公司 | Ultrasonic image denoising neural network, denoising method and ultrasonic diagnostic equipment |
CN113223005A (en) * | 2021-05-11 | 2021-08-06 | 天津大学 | Thyroid nodule automatic segmentation and grading intelligent system |
CN113256605A (en) * | 2021-06-15 | 2021-08-13 | 四川大学 | Breast cancer image identification and classification method based on deep neural network |
CN114119458A (en) * | 2021-09-14 | 2022-03-01 | 福州大学 | Thyroid medullary cancer ultrasonic image identification method based on clinical priori knowledge guidance |
CN114782676A (en) * | 2022-04-02 | 2022-07-22 | 北京广播电视台 | Method and system for extracting region of interest of video |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106778664A (en) * | 2016-12-29 | 2017-05-31 | 天津中科智能识别产业技术研究院有限公司 | The dividing method and its device of iris region in a kind of iris image |
CN107680678A (en) * | 2017-10-18 | 2018-02-09 | 北京航空航天大学 | Based on multiple dimensioned convolutional neural networks Thyroid ultrasound image tubercle auto-check system |
-
2018
- 2018-07-13 CN CN201810772176.7A patent/CN109087327B/en active Active
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106778664A (en) * | 2016-12-29 | 2017-05-31 | 天津中科智能识别产业技术研究院有限公司 | The dividing method and its device of iris region in a kind of iris image |
CN107680678A (en) * | 2017-10-18 | 2018-02-09 | 北京航空航天大学 | Based on multiple dimensioned convolutional neural networks Thyroid ultrasound image tubercle auto-check system |
Non-Patent Citations (1)
Title |
---|
廖欣等: "基于深度卷积神经网络的宫颈细胞病理智能辅助诊断方法", 《液晶与显示》 * |
Cited By (48)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2020173163A1 (en) * | 2019-02-26 | 2020-09-03 | 电信科学技术研究院有限公司 | Image segmentation method and apparatus |
CN109949318A (en) * | 2019-03-07 | 2019-06-28 | 西安电子科技大学 | Full convolutional neural networks epileptic focus dividing method based on multi-modal image |
CN109949318B (en) * | 2019-03-07 | 2023-11-14 | 西安电子科技大学 | Full convolution neural network epileptic focus segmentation method based on multi-modal image |
CN109919098A (en) * | 2019-03-08 | 2019-06-21 | 广州视源电子科技股份有限公司 | The recognition methods of target object and device |
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 |
CN110074813A (en) * | 2019-04-26 | 2019-08-02 | 深圳大学 | A kind of ultrasonic image reconstruction method and system |
WO2020228490A1 (en) * | 2019-05-16 | 2020-11-19 | 京东方科技集团股份有限公司 | Method and device for performing target labelling on image, and computer recording medium |
CN110110808B (en) * | 2019-05-16 | 2022-04-15 | 京东方科技集团股份有限公司 | Method and device for performing target labeling on image and computer recording medium |
US11735316B2 (en) | 2019-05-16 | 2023-08-22 | Beijing Boe Technology Development Co., Ltd. | Method and apparatus of labeling target in image, and computer recording medium |
CN110110808A (en) * | 2019-05-16 | 2019-08-09 | 京东方科技集团股份有限公司 | A kind of pair of image carries out the method, apparatus and computer readable medium of target mark |
CN110136809A (en) * | 2019-05-22 | 2019-08-16 | 腾讯科技(深圳)有限公司 | A kind of medical image processing method, device, electromedical equipment and storage medium |
CN110232378A (en) * | 2019-05-30 | 2019-09-13 | 苏宁易购集团股份有限公司 | A kind of image interest point detecting method, system and readable storage medium storing program for executing |
CN110211134A (en) * | 2019-05-30 | 2019-09-06 | 上海商汤智能科技有限公司 | A kind of image partition method and device, electronic equipment and storage medium |
CN110211134B (en) * | 2019-05-30 | 2021-11-05 | 上海商汤智能科技有限公司 | Image segmentation method and device, electronic equipment and storage medium |
CN110610487A (en) * | 2019-08-29 | 2019-12-24 | 上海杏脉信息科技有限公司 | Method and device for measuring increase of heart shadow |
CN111798455B (en) * | 2019-09-25 | 2023-07-04 | 天津大学 | Thyroid nodule real-time segmentation method based on full convolution dense cavity network |
CN110706793A (en) * | 2019-09-25 | 2020-01-17 | 天津大学 | Attention mechanism-based thyroid nodule semi-supervised segmentation method |
CN111798455A (en) * | 2019-09-25 | 2020-10-20 | 天津大学 | Thyroid nodule real-time segmentation method based on full convolution dense cavity network |
CN110992306B (en) * | 2019-11-04 | 2022-04-22 | 国网河北省电力有限公司检修分公司 | Method and device for segmenting deteriorated insulator in infrared image based on deep learning |
CN110992306A (en) * | 2019-11-04 | 2020-04-10 | 国网河北省电力有限公司检修分公司 | Method and device for segmenting deteriorated insulator in infrared image based on deep learning |
CN111047598A (en) * | 2019-11-04 | 2020-04-21 | 华北电力大学(保定) | Deep learning-based ultraviolet discharge light spot segmentation method and device for power transmission and transformation equipment |
CN111047598B (en) * | 2019-11-04 | 2023-09-01 | 华北电力大学(保定) | Deep learning-based ultraviolet discharge light spot segmentation method and device for power transmission and transformation equipment |
CN113034373A (en) * | 2019-12-24 | 2021-06-25 | 无锡祥生医疗科技股份有限公司 | Ultrasonic image denoising neural network, denoising method and ultrasonic diagnostic equipment |
CN111210439B (en) * | 2019-12-26 | 2022-06-24 | 中国地质大学(武汉) | Semantic segmentation method and device by suppressing uninteresting information and storage device |
CN111210439A (en) * | 2019-12-26 | 2020-05-29 | 中国地质大学(武汉) | Semantic segmentation method and device by suppressing uninteresting information and storage device |
CN111105410A (en) * | 2019-12-27 | 2020-05-05 | 中国人民解放军陆军军医大学第二附属医院 | Hematopoietic tissue proportion determining device and method based on bone marrow biopsy image |
CN111161273B (en) * | 2019-12-31 | 2023-03-21 | 电子科技大学 | Medical ultrasonic image segmentation method based on deep learning |
CN111161273A (en) * | 2019-12-31 | 2020-05-15 | 电子科技大学 | Medical ultrasonic image segmentation method based on deep learning |
CN111242956A (en) * | 2020-01-09 | 2020-06-05 | 西北工业大学 | U-Net-based ultrasonic fetal heart and fetal lung deep learning joint segmentation method |
CN111311561A (en) * | 2020-02-10 | 2020-06-19 | 浙江未来技术研究院(嘉兴) | Automatic operation area photometry method and device based on microsurgery imaging system |
CN111311561B (en) * | 2020-02-10 | 2023-10-10 | 浙江未来技术研究院(嘉兴) | Automatic operation area photometry method and device based on microsurgery imaging system |
CN111223092A (en) * | 2020-02-28 | 2020-06-02 | 长沙大端信息科技有限公司 | Automatic quality control system and detection method for ultrasonic sectional images of fetus |
CN111292324B (en) * | 2020-03-20 | 2022-03-01 | 电子科技大学 | Multi-target identification method and system for brachial plexus ultrasonic image |
CN111292324A (en) * | 2020-03-20 | 2020-06-16 | 电子科技大学 | Multi-target identification method and system for brachial plexus ultrasonic image |
CN111626994A (en) * | 2020-05-18 | 2020-09-04 | 江苏远望仪器集团有限公司 | Equipment fault defect diagnosis method based on improved U-Net neural network |
CN111724397B (en) * | 2020-06-18 | 2024-04-16 | 上海应用技术大学 | Automatic segmentation method for craniocerebral CT image bleeding area |
CN111724397A (en) * | 2020-06-18 | 2020-09-29 | 上海应用技术大学 | Automatic segmentation method for bleeding area of craniocerebral CT image |
CN111738338A (en) * | 2020-06-23 | 2020-10-02 | 征图新视(江苏)科技股份有限公司 | Defect detection method applied to motor coil based on cascaded expansion FCN network |
CN112085028B (en) * | 2020-08-31 | 2024-03-12 | 浙江工业大学 | Tooth full-scene semantic segmentation method based on feature map disturbance and boundary supervision |
CN112085028A (en) * | 2020-08-31 | 2020-12-15 | 浙江工业大学 | Tooth panoramic semantic segmentation method based on feature map disturbance and boundary supervision |
CN112085729A (en) * | 2020-09-18 | 2020-12-15 | 无锡祥生医疗科技股份有限公司 | Pleural line region extraction method, storage medium, and ultrasound diagnostic apparatus |
CN112529894A (en) * | 2020-12-22 | 2021-03-19 | 徐州医科大学 | Thyroid nodule diagnosis method based on deep learning network |
CN113223005A (en) * | 2021-05-11 | 2021-08-06 | 天津大学 | Thyroid nodule automatic segmentation and grading intelligent system |
CN113256605B (en) * | 2021-06-15 | 2021-11-02 | 四川大学 | Breast cancer image identification and classification method based on deep neural network |
CN113256605A (en) * | 2021-06-15 | 2021-08-13 | 四川大学 | Breast cancer image identification and classification method based on deep neural network |
CN114119458A (en) * | 2021-09-14 | 2022-03-01 | 福州大学 | Thyroid medullary cancer ultrasonic image identification method based on clinical priori knowledge guidance |
CN114782676B (en) * | 2022-04-02 | 2023-01-06 | 北京广播电视台 | Method and system for extracting region of interest of video |
CN114782676A (en) * | 2022-04-02 | 2022-07-22 | 北京广播电视台 | Method and system for extracting region of interest of video |
Also Published As
Publication number | Publication date |
---|---|
CN109087327B (en) | 2021-07-06 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109087327A (en) | A kind of thyroid nodule ultrasonic image division method cascading full convolutional neural networks | |
CN106056595B (en) | Based on the pernicious assistant diagnosis system of depth convolutional neural networks automatic identification Benign Thyroid Nodules | |
CN112150428B (en) | Medical image segmentation method based on deep learning | |
US7876947B2 (en) | System and method for detecting tagged material using alpha matting | |
CN108961229A (en) | Cardiovascular OCT image based on deep learning easily loses plaque detection method and system | |
CN111179237B (en) | Liver and liver tumor image segmentation method and device | |
CN110310289A (en) | Lung tissue's image partition method based on deep learning | |
CN110599499B (en) | MRI image heart structure segmentation method based on multipath convolutional neural network | |
CN110188792A (en) | The characteristics of image acquisition methods of prostate MRI 3-D image | |
CN112488976B (en) | Multi-modal medical image fusion method based on DARTS network | |
CN113506310B (en) | Medical image processing method and device, electronic equipment and storage medium | |
Tirpude et al. | Automated detection and extraction of brain tumor from MRI images | |
CN109902682A (en) | A kind of mammary gland x line image detection method based on residual error convolutional neural networks | |
CN114693933A (en) | Medical image segmentation device based on generation of confrontation network and multi-scale feature fusion | |
Sun et al. | A novel gastric ulcer differentiation system using convolutional neural networks | |
CN113223004A (en) | Liver image segmentation method based on deep learning | |
JP2024027078A (en) | Multi-scale whole slide pathological feature fusion extraction method, system, electronic equipment and storage medium | |
CN111091575A (en) | Medical image segmentation method based on reinforcement learning method | |
JP2023517058A (en) | Automatic detection of tumors based on image processing | |
Somasundaram et al. | Brain extraction method for T1-weighted magnetic resonance scans | |
Sun et al. | ASU-Net: U-shape adaptive scale network for mass segmentation in mammograms | |
CN116664590A (en) | Automatic segmentation method and device based on dynamic contrast enhancement magnetic resonance image | |
CN105956587A (en) | Method for automatically extracting meniscus from knee-joint magnetic resonance image sequence based on shape constraint | |
Azli et al. | Ultrasound image segmentation using a combination of edge enhancement and kirsch’s template method for detecting follicles in ovaries | |
CN113139627B (en) | Mediastinal lump identification method, system 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 |