CN110321968A - A kind of ultrasound image sorter - Google Patents
A kind of ultrasound image sorter Download PDFInfo
- Publication number
- CN110321968A CN110321968A CN201910624643.6A CN201910624643A CN110321968A CN 110321968 A CN110321968 A CN 110321968A CN 201910624643 A CN201910624643 A CN 201910624643A CN 110321968 A CN110321968 A CN 110321968A
- Authority
- CN
- China
- Prior art keywords
- image
- ultrasound image
- unit
- sample
- layer
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
- Ultra Sonic Daignosis Equipment (AREA)
Abstract
The embodiment of the invention discloses a kind of ultrasound image sorter, converting unit formats the ultrasound image of acquisition, obtains target ultrasound image；Target ultrasound image is carried out residual noise reduction, obtains multiple subgraphs by residual noise reduction unit.Residual noise reduction is carried out to image, can be multiple subgraphs by Ultrasound Image Segmentation, allow ultrasound image to be divided into smaller elementary area and handled, effectively improve the accuracy of image analysis.And compared with traditional Convolution Analysis, the mode of residual noise reduction can reduce data processing amount, to improve the treatment effeciency of image classification.Determination unit utilizes trained capsule network, analyzes multiple subgraphs, determines image category belonging to ultrasound image.Capsule network is by carrying out clustering to subgraph, it can be deduced that the relevance between ultrasound image and each image category, to accurately evaluate image category belonging to ultrasound image.
Description
Technical field
The present invention relates to ultrasound image processing technology fields, more particularly to a kind of ultrasound image sorter.
Background technique
Thyroid papillary carcinoma is one of the most common type cancer, and the morbidity performance of thyroid papillary carcinoma can be divided into shape
Irregularly, the symptoms such as clear, the even, calcification of Echo heterogenicity are owed on boundary.Ultrasonic examination is preferred imageological examination.Medical image data
It is higher to the job requirement of doctor with complexity, diversity.In the case where needing to diagnose a large amount of ultrasound images, doctor's
Diagnosis efficiency often declines with the increase of ultrasound image quantity.Since the fatigue of doctor will increase its False Rate, so knot
It closes to obtain a result after the relevant technologies such as artificial intelligence pre-process and gives doctor's judgement again, the work of doctor can be mitigated significantly
Amount improves working efficiency.
Existing medical image sorting technique is to realize its function with convolutional neural networks for basic module.Due to medical shadow
A series of problems, such as complexity of picture is higher, and ultrasound image is there are cell overlap, edge blurry, and blood vessel interferes, only uses convolution
Neural network often will cause a large amount of false positive phenomenon as classifier, and low accuracy rate is right during actual medical judges
The help of doctor is little
It is those skilled in the art's problem to be solved as it can be seen that how to promote the accuracy of ultrasound image classification.
Summary of the invention
The purpose of the embodiment of the present invention is that providing a kind of ultrasound image sorter, the standard of ultrasound image classification can be promoted
True property.
In order to solve the above technical problems, the embodiment of the present invention provides a kind of ultrasound image sorter, including converting unit,
Residual noise reduction unit and determination unit；
The converting unit obtains target ultrasound image for formatting to the ultrasound image of acquisition；
The residual noise reduction unit obtains multiple subgraphs for the target ultrasound image to be carried out residual noise reduction；
The determination unit is analyzed the multiple subgraph, is determined for utilizing trained capsule network
Image category belonging to the ultrasound image.
Optionally, the determination unit includes extracting subelement, cluster subelement and classification subelement；
The extraction subelement will be described primary special for extracting the primary features of each subgraph, and according to default dimension
Sign is converted into feature vector；
The cluster subelement carries out clustering processing to described eigenvector, obtains defeated for utilizing dynamic routing algorithm
Result out；
The classification subelement is determined belonging to the ultrasound image for the norm value according to the output result
Image category.
Optionally, the converting unit includes format conversion subunit and size conversion subunit；
The format conversion subunit, the ultrasound image for will acquire are converted into jpg image；
The size conversion subunit is big for the jpg image to be fixed as presetted pixel using bilinear interpolation
Small target ultrasound image.
Optionally, it is directed to the training process of the capsule network, described device includes extraction unit, cluster cell, meter
Calculate unit, judging unit, adjustment unit and output unit；
The extraction unit, for extracting the sample primary features of each sample image, and according to default dimension, by the sample
This primary features is converted into sampling feature vectors；Wherein, each sample image has its corresponding real image class label；
The cluster cell, for utilizing dynamic routing algorithm, to the spy of sample corresponding to the same target sample image
It levies vector and carries out clustering processing, obtain sample output result；
The computing unit, for calculating the corresponding norm value of the sample output result；And by the maximum norm of value
It is worth the sample norm value as the target sample image；
The judging unit, for judging whether the sample norm value matches with the real image class label；If it is not,
Then trigger the adjustment unit；If so, triggering the output unit；
The adjustment unit, for the deviation according to sample output valve and the real image class label, described in adjustment
The weighted value of capsule network；And the step of the sample primary features for extracting each sample image is returned to after adjusting weighted value
Suddenly；
The output unit is trained for terminating, and exports trained capsule network.
Optionally, the cluster cell is specifically used for obtaining sample output result according to following formula:
In formula,
Wherein, vjIndicate the output result of jth layer capsule layer；sjIndicate the vector of input jth layer capsule layer；Make it
The output valve of capsule layer guarantees in section [0,1]；cijWhat is indicated is adjacent capsules layer i.e. i, j layers of relating value；Indicate be
Show that jth layer predicted vector exports by calculating i-th layer of capsule layer；bijThe i-th layer of glue epithecium jth layer capsule layer choosing indicated is selected
Probability, routing layer start execute when, initial value is set as 0；K is the total number of capsule in digital capsule layer, j ∈ [1,
k]；WijWhat is indicated is for learning and the weighted value of backpropagation；uiWhat is indicated is the output of i-th layer of capsule layer.
Optionally, the judgement is specifically used for calculating sample output valve and the real image class label according to following formula
Deviation Lc:
Lc=Tcmax(0,m+-||vc||)2+λ(1-Tc)max(0,||vc||-m-)2；
Wherein, c indicates the classification of image classification；TcIndicate the indicator function of c classification, T in the presence of the c that classifiesc=1, classify c
In the absence of Tc=0；m+Indicate top edge parameter；m-Indicate lower edge parameter；λ indicates scale parameter；vcIndicate that ultrasound image exists
Output result under c classification.
Optionally, described image classification include image in irregular shape, the clear image of obscure boundary, the even image of Echo heterogenicity,
Calcification image and normal picture.
It optionally, further include display unit；The display unit, for showing image class belonging to the ultrasound image
Not.
It optionally, further include storage unit；
The storage unit, for storing image category belonging to the ultrasound image.
It optionally, further include statistic unit；
The statistic unit, for counting total and every kind of image category institute of the ultrasound image obtained in preset time
The number of corresponding ultrasound image.
Converting unit formats the ultrasound image of acquisition it can be seen from above-mentioned technical proposal, obtains target
Ultrasound image；Target ultrasound image is carried out residual noise reduction, obtains multiple subgraphs by residual noise reduction unit.Image is carried out residual
Difference processing, can by Ultrasound Image Segmentation be multiple subgraphs, allow ultrasound image be divided into smaller elementary area into
Row processing, effectively improves the accuracy of image analysis.And compared with traditional Convolution Analysis, the mode of residual noise reduction can
To reduce data processing amount, to improve the treatment effeciency of image classification.Determination unit utilizes trained capsule network, right
Multiple subgraphs are analyzed, and determine image category belonging to ultrasound image.Capsule network is by clustering subgraph
Analysis, it can be deduced that the relevance between ultrasound image and each image category, to accurately evaluate ultrasound image
Affiliated image category.
Detailed description of the invention
In order to illustrate the embodiments of the present invention more clearly, attached drawing needed in the embodiment will be done simply below
It introduces, it should be apparent that, drawings in the following description are only some embodiments of the invention, for ordinary skill people
For member, without creative efforts, it is also possible to obtain other drawings based on these drawings.
Fig. 1 is a kind of structural schematic diagram of ultrasound image sorter provided in an embodiment of the present invention；
Fig. 2 is a kind of network model schematic diagram of residual error network and the capsule network integration provided in an embodiment of the present invention.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, rather than whole embodiments.Based on this
Embodiment in invention, those of ordinary skill in the art are without making creative work, obtained every other
Embodiment belongs to the scope of the present invention.
In order to enable those skilled in the art to better understand the solution of the present invention, with reference to the accompanying drawings and detailed description
The present invention is described in further detail.
Next, ultrasound image sorter provided by the embodiment of the present invention is discussed in detail.Fig. 1 is the embodiment of the present invention
A kind of structural schematic diagram of the ultrasound image sorter provided, which includes converting unit 11,12 and of residual noise reduction unit
Determination unit 13.
Converting unit 11 obtains target ultrasound image for formatting to the ultrasound image of acquisition.
It after getting ultrasound image, needs to pre-process ultrasound image, i.e., converts calculating for ultrasound image
The picture format that machine can identify can convert ultrasound image to jpg image in practical applications.
In view of difference can in order to realize the standardization processing of ultrasound image for the sizes of different ultrasound images
To utilize bilinear interpolation method, the ultrasound image after format is converted is fixed as the ultrasound image of presetted pixel size.
The value of presetted pixel size can be set according to actual needs, for example, presetted pixel size can be set as
78*78 pixel value.
By taking jpg image as an example, jpg image can be fixed as to the target ultrasound image of 78*78 pixel value.
Residual noise reduction unit 12 obtains multiple subgraphs for target ultrasound image to be carried out residual noise reduction.
12 main function of residual noise reduction unit is that target ultrasound image is divided into smaller elementary area to handle, from
And make analysis result more careful accurate.And compared with the processing mode of convolutional neural networks, target ultrasound image passes through
Residual noise reduction reduces the calculation amount of image analysis, improves the treatment effeciency of image classification.
Residual noise reduction unit 12 may include having multiple residual error modules, each residual error module is 1* by convolution kernel size
1, convolution kernel size is 3*3, and port number is 32 compositions.Correspondingly, target ultrasound image progress residual noise reduction, available 32
A subgraph.
In practical applications, residual noise reduction can be named according to the number of the included residual error module of residual noise reduction unit 12
Unit 12, is named as ResNet1 comprising a residual error module, is named as ResNet2 comprising two residual error modules, and so on,
ResNetN is named as comprising N number of residual error module.
Determination unit 13 analyzes multiple subgraphs, determines ultrasonic figure for utilizing trained capsule network
As affiliated image category.
Ultrasound image sorter provided in an embodiment of the present invention is suitable for point of the ultrasound image of thyroid papillary carcinoma
Class.It is showed according to the morbidity of thyroid papillary carcinoma, can be divided into that in irregular shape, obscure boundary is clear, the uniform calcium of Echo heterogenicity
The symptoms such as change.Therefore, in capsule network training, the image category of setting may include image in irregular shape, obscure boundary
The even image of clear image, Echo heterogenicity, calcification image and normal picture.
Every ultrasound image has its corresponding multiple subgraph, is by same Zhang Chaosheng in embodiments of the present invention
Subgraph corresponding to image is input in trained capsule network as one group of image, so that it is determined that going out this ultrasound image
Affiliated image category.
According to the different layer functions of capsule network, determination unit 13 can be divided into extract subelement, cluster subelement and
Sort out subelement.
Subelement is extracted to convert primary features to for extracting the primary features of each subgraph, and according to default dimension
Feature vector.
Primary features reflect the image distribution feature of subgraph.The subgraph of different images classification is extracted primary special
Sign is different.
The primary features extracted belong to scalar, for the ease of subsequent processing analysis, need to convert vector for scalar
Form.In embodiments of the present invention, by the way that primary features are carried out Spatial Dimension conversion, on the basis of primary features dimension again
A new dimension is extended, to obtain and the one-to-one feature vector of each primary features.
For example, a primary features of convolutional layer output are [8,24], then converted by the Spatial Dimension of initial capsule layer
Afterwards, can become [8,8,3], the number of numerical value does not become in matrix, a dimension is only increased, so as to so that this feature
Vector characterizes more parameters, to improve the accuracy of identification, prepares for next layer of input.
Subelement is clustered, for utilizing dynamic routing algorithm, clustering processing is carried out to feature vector, obtains output result.
Clustering processing is carried out to feature vector to refer to clustering feature vector corresponding to same ultrasound image
Processing.
In conjunction with above-mentioned introduction, image category may include having 5 classifications, correspondingly, obtained output result includes 5
Export result.This 5 output results reflect the relevance of ultrasound image Yu this 5 classifications respectively.
Sort out subelement, for the norm value according to output result, determines image category belonging to ultrasound image.
The number of element can be 16 in each output result, and further seek L2 norm to this 5 output results,
The norm value (namely length) of each output result is found out, what the maximum output result of value represented is exactly image maximum probability
That classification.In embodiments of the present invention, the probability that entity occurs, L2 norm are measured by the size of output vector L2 norm
Value is bigger, and the probability that corresponding classification occurs is bigger.The corresponding image category of the maximum norm value of value is ultrasound image
Affiliated image category.
Capsule network includes input layer, convolutional layer, initial capsule layer (PrimaryCaps), digital capsule layer
(DigitCaps) and output layer.
In embodiments of the present invention, cooperating by residual error network and capsule network, realizes the classification of ultrasound image.
As shown in Fig. 2, the leftmost side represents ultrasound image in Fig. 2, ResNet is represented for residual error network and the network model of the capsule network integration
Residual error network, PrimaryCaps represent initial capsule layer, and DigitCaps represents digital capsule layer.Ultrasound image passes through residual error net
Network processing, available multiple subgraphs, PrimaryCaps are used to be presented this multiple subgraph in the form of feature vector,
DigitCaps analyzes feature vector using dynamic routing algorithm, obtains output result.
In embodiments of the present invention, capsule network is trained by sample image, continuously adjusts capsule network
Weighted value, so that the accuracy of capsule network can satisfy preset requirement, to obtain trained capsule network.
Trained capsule network is the basis for carrying out image category analysis.Next, by training to capsule network
Cheng Zhankai is introduced.
It is directed to the training process of capsule network, ultrasound image sorter includes extraction unit, cluster cell, calculates list
Member, judging unit, adjustment unit and output unit.
Extraction unit, it is for extracting the sample primary features of each sample image, and according to default dimension, sample is primary special
Sign is converted into sampling feature vectors；Wherein, each sample image has its corresponding real image class label.
Cluster cell, for utilize dynamic routing algorithm, to sample characteristics corresponding to the same target sample image to
Amount carries out clustering processing, obtains sample output result.
Cluster cell is specifically used for obtaining sample output result according to following formula:
In formula,
Wherein, vjIndicate the output result of jth layer capsule layer；sjIndicate the vector of input jth layer capsule layer；Make it
The output valve of capsule layer guarantees in section [0,1]；cijWhat is indicated is adjacent capsules layer i.e. i, j layers of relating value；Indicate be
Show that jth layer predicted vector exports by calculating i-th layer of capsule layer；bijThe i-th layer of glue epithecium jth layer capsule layer choosing indicated is selected
Probability, routing layer start execute when, initial value is set as 0；K is the total number of capsule in digital capsule layer, j ∈ [1,
k]；WijWhat is indicated is for learning and the weighted value of backpropagation；uiWhat is indicated is the output of i-th layer of capsule layer.
Computing unit, for calculating the corresponding norm value of sample output result；And using the maximum norm value of value as mesh
Mark the sample norm value of sample image.
Whether judging unit matches for judgement sample norm value with real image class label.
It may include sample norm value and real image class label that whether sample norm value matches with real image class label
Value it is identical, also may include the deviation of sample norm value and real image class label within a preset range.
When sample norm value and real image class label mismatch, then illustrate the classification assessment accuracy of capsule network not
Reach requirement, adjustment unit can be triggered at this time.
Adjustment unit adjusts the power of capsule network for the deviation according to sample output valve and real image class label
Weight values；And the step of sample primary features for extracting each sample image are returned after adjusting weighted value.
It, can be with the image category of pre-set ultrasound image in the training process of capsule network.
For the ease of distinguishing a variety of image categories, classification marker can be set for every kind of image category, for example, using 0_
00001 indicates that first image in irregular shape, 0_00002 indicate second image in irregular shape, and so on.?
When capsule network training, classification in irregular shape includes 444 sample images in total.1_00001 indicates clear the of obscure boundary
One image, in capsule network training, the clear class of obscure boundary amounts to 499 sample images.2_00001 indicates that Echo heterogenicity is even
First image, in capsule network training, the even class of Echo heterogenicity amounts to 454 sample images.3_00001 indicates calcification
First image, in capsule network training, calcification class amounts to 515 sample images.4_00001 indicates normal first figure
Picture, in capsule network training, which amounts to 468 sample images.
Pond layer in traditional convolutional network can lose some information during dimensionality reduction, so as to cause accuracy of identification
Decline.The accuracy of identification of image category can be effectively promoted using the dynamic routing principle in capsule network, principle is main
The degree of association that upper and lower capsule layer is updated by judging weighted value accepts or rejects this layer ratio shared in learning process, if prediction
Result it is close with true value, then increase the relating value of this layer, if prediction result differed with true value farther out, reduce the layer
Relating value.
In embodiments of the present invention, judgement is specifically used for calculating sample output valve and real image classification according to following formula
The deviation L of valuec:
Lc=Tcmax(0,m+-||vc||)2+λ(1-Tc)max(0,||vc||-m-)2；
Wherein, c indicates the classification of image classification；TcIndicate the indicator function of c classification, T in the presence of the c that classifiesc=1, classify c
In the absence of Tc=0；m+Indicate top edge parameter；m-Indicate lower edge parameter；λ indicates scale parameter；vcIndicate that ultrasound image exists
Output result under c classification.
When sample norm value is matched with real image class label, then the classification assessment accuracy of capsule network is illustrated
Reach requirement, can trigger output unit at this time terminates to train, and exports trained capsule network.
Converting unit formats the ultrasound image of acquisition it can be seen from above-mentioned technical proposal, obtains target
Ultrasound image；Target ultrasound image is carried out residual noise reduction, obtains multiple subgraphs by residual noise reduction unit.Image is carried out residual
Difference processing, can by Ultrasound Image Segmentation be multiple subgraphs, allow ultrasound image be divided into smaller elementary area into
Row processing, effectively improves the accuracy of image analysis.And compared with traditional Convolution Analysis, the mode of residual noise reduction can
To reduce data processing amount, to improve the treatment effeciency of image classification.Determination unit utilizes trained capsule network, right
Multiple subgraphs are analyzed, and determine image category belonging to ultrasound image.Capsule network is by clustering subgraph
Analysis, it can be deduced that the relevance between ultrasound image and each image category, to accurately evaluate ultrasound image
Affiliated image category.
In embodiments of the present invention, the classification assessment result of ultrasound image, Ke Yishe are got information about for the ease of doctor
Set display unit.Image category belonging to ultrasound image is shown by the display unit.
In practical applications, it is called for the ease of the subsequent image category assessment result to ultrasound image, in ultrasound
Storage unit can be set in image classification device.
Storage unit, for storing image category belonging to ultrasound image.
For example, its corresponding image category of ultrasound image can be stored in the form of corresponding relationship list, with
Facilitate and subsequent checks calling.
For the ease of understanding the probability that different images classification occurs, it is single that statistics can be set in ultrasound image sorter
Member.Statistic unit surpasses corresponding to total and every kind of image category of the ultrasound image obtained in preset time for counting
The number of acoustic image.
A kind of ultrasound image sorter is provided for the embodiments of the invention above to be described in detail.In specification
Each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments,
The same or similar parts in each embodiment may refer to each other.For the device disclosed in the embodiment, due to itself and implementation
Method disclosed in example is corresponding, so being described relatively simple, reference may be made to the description of the method.It should be pointed out that
For those skilled in the art, without departing from the principle of the present invention, can also to the present invention into
Row some improvements and modifications, these improvements and modifications also fall within the scope of protection of the claims of the present invention.
Professional further appreciates that, unit described in conjunction with the examples disclosed in the embodiments of the present disclosure
And algorithm steps, can be realized with electronic hardware, computer software, or a combination of the two, in order to clearly demonstrate hardware and
The interchangeability of software generally describes each exemplary composition and step according to function in the above description.These
Function is implemented in hardware or software actually, the specific application and design constraint depending on technical solution.Profession
Technical staff can use different methods to achieve the described function each specific application, but this realization is not answered
Think beyond the scope of this invention.
The step of method described in conjunction with the examples disclosed in this document or algorithm, can directly be held with hardware, processor
The combination of capable software module or the two is implemented.Software module can be placed in random access memory (RAM), memory, read-only deposit
Reservoir (ROM), electrically programmable ROM, electrically erasable ROM, register, hard disk, moveable magnetic disc, CD-ROM or technology
In any other form of storage medium well known in field.
Claims (10)
1. a kind of ultrasound image sorter, which is characterized in that including converting unit, residual noise reduction unit and determination unit；
The converting unit obtains target ultrasound image for formatting to the ultrasound image of acquisition；
The residual noise reduction unit obtains multiple subgraphs for the target ultrasound image to be carried out residual noise reduction；
The determination unit is analyzed the multiple subgraph, is determined described for utilizing trained capsule network
Image category belonging to ultrasound image.
2. the apparatus according to claim 1, which is characterized in that the determination unit includes extracting subelement, cluster son list
Member and classification subelement；
The extraction subelement turns the primary features for extracting the primary features of each subgraph, and according to default dimension
Turn to feature vector；
The cluster subelement carries out clustering processing to described eigenvector for utilizing dynamic routing algorithm, obtains output knot
Fruit；
The classification subelement determines image belonging to the ultrasound image for the norm value according to the output result
Classification.
3. the apparatus according to claim 1, which is characterized in that the converting unit includes format conversion subunit and size
Conversion subunit；
The format conversion subunit, the ultrasound image for will acquire are converted into jpg image；
The size conversion subunit, for the jpg image to be fixed as presetted pixel size using bilinear interpolation
Target ultrasound image.
4. the apparatus according to claim 1, which is characterized in that be directed to the training process of the capsule network, the dress
It sets including extraction unit, cluster cell, computing unit, judging unit, adjustment unit and output unit；
The extraction unit will be at the beginning of the sample for extracting the sample primary features of each sample image, and according to default dimension
Grade feature is converted into sampling feature vectors；Wherein, each sample image has its corresponding real image class label；
The cluster cell, for utilize dynamic routing algorithm, to sample characteristics corresponding to the same target sample image to
Amount carries out clustering processing, obtains sample output result；
The computing unit, for calculating the corresponding norm value of the sample output result；And the maximum norm value of value is made
For the sample norm value of the target sample image；
The judging unit, for judging whether the sample norm value matches with the real image class label；If it is not, then touching
Send out adjustment unit described；If so, triggering the output unit；
The adjustment unit adjusts the capsule for the deviation according to sample output valve and the real image class label
The weighted value of network；And the step of sample primary features for extracting each sample image are returned after adjusting weighted value；
The output unit is trained for terminating, and exports trained capsule network.
5. device according to claim 4, which is characterized in that the cluster cell is specifically used for obtaining according to following formula
Result is exported to sample:
In formula,
Wherein, vjIndicate the output result of jth layer capsule layer；sjIndicate the vector of input jth layer capsule layer；Make its capsule
The output valve of layer guarantees in section [0,1]；cijWhat is indicated is adjacent capsules layer i.e. i, j layers of relating value；What is indicated is to pass through
It calculates i-th layer of capsule layer and show that jth layer predicted vector exports；bijThe i-th layer of glue epithecium jth layer capsule layer choosing indicated is selected general
Rate, when routing layer starts to execute, initial value is set as 0；K is the total number of capsule in digital capsule layer, j ∈ [1, k]；Wij
What is indicated is for learning and the weighted value of backpropagation；uiWhat is indicated is the output of i-th layer of capsule layer.
6. device according to claim 5, which is characterized in that the judgement is specifically used for calculating sample according to following formula
The deviation L of output valve and the real image class labelc:
Lc=Tcmax(0,m+-||vc||)2+λ(1-Tc)max(0,||vc||-m-)2；
Wherein, c indicates the classification of image classification；TcIndicate the indicator function of c classification, T in the presence of the c that classifiesc=1, classification c is not deposited
When Tc=0；m+Indicate top edge parameter；m-Indicate lower edge parameter；λ indicates scale parameter；vcIndicate ultrasound image in c class
Output result under not.
7. device described in -6 any one according to claim 1, which is characterized in that described image classification includes in irregular shape
The clear image of image, obscure boundary, the even image of Echo heterogenicity, calcification image and normal picture.
8. device described in -6 any one according to claim 1, which is characterized in that further include display unit；The displaying is single
Member, for showing image category belonging to the ultrasound image.
9. device described in -6 any one according to claim 1, which is characterized in that further include storage unit；
The storage unit, for storing image category belonging to the ultrasound image.
10. device described in -6 any one according to claim 1, which is characterized in that further include statistic unit；
The statistic unit, for counting corresponding to total and every kind of image category of the ultrasound image obtained in preset time
Ultrasound image number.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910624643.6A CN110321968B (en) | 2019-07-11 | 2019-07-11 | Ultrasonic image classification device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910624643.6A CN110321968B (en) | 2019-07-11 | 2019-07-11 | Ultrasonic image classification device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110321968A true CN110321968A (en) | 2019-10-11 |
| CN110321968B CN110321968B (en) | 2023-05-05 |
Family
ID=68121960
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910624643.6A Active CN110321968B (en) | 2019-07-11 | 2019-07-11 | Ultrasonic image classification device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110321968B (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111080168A (en) * | 2019-12-30 | 2020-04-28 | 国网江苏省电力有限公司信息通信分公司 | Power communication network equipment reliability evaluation method based on capsule network |
| CN111723840A (en) * | 2020-05-08 | 2020-09-29 | 天津大学 | Clustering and style migration method for ultrasonic images |
| CN112233106A (en) * | 2020-10-29 | 2021-01-15 | 电子科技大学中山学院 | Thyroid cancer ultrasonic image analysis method based on residual capsule network |
| CN113450325A (en) * | 2021-06-28 | 2021-09-28 | 什维新智医疗科技(上海)有限公司 | Thyroid nodule benign and malignant recognition device |
| CN115019305A (en) * | 2022-08-08 | 2022-09-06 | 成都西交智汇大数据科技有限公司 | Method, device and equipment for identifying root tip cells and readable storage medium |
| CN117351012A (en) * | 2023-12-04 | 2024-01-05 | 天津医科大学第二医院 | Fetal image recognition method and system based on deep learning |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180012107A1 (en) * | 2015-12-11 | 2018-01-11 | Tencent Technology (Shenzhen) Company Limited | Image classification method, electronic device, and storage medium |
| CN108830826A (en) * | 2018-04-28 | 2018-11-16 | 四川大学 | A kind of system and method detecting Lung neoplasm |
| CN109300107A (en) * | 2018-07-24 | 2019-02-01 | 深圳先进技术研究院 | Patch processing method, device and the calculating equipment of magnetic resonance vascular wall imaging |
| CN109840560A (en) * | 2019-01-25 | 2019-06-04 | 西安电子科技大学 | Based on the image classification method for incorporating cluster in capsule network |
-
2019
- 2019-07-11 CN CN201910624643.6A patent/CN110321968B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180012107A1 (en) * | 2015-12-11 | 2018-01-11 | Tencent Technology (Shenzhen) Company Limited | Image classification method, electronic device, and storage medium |
| CN108830826A (en) * | 2018-04-28 | 2018-11-16 | 四川大学 | A kind of system and method detecting Lung neoplasm |
| CN109300107A (en) * | 2018-07-24 | 2019-02-01 | 深圳先进技术研究院 | Patch processing method, device and the calculating equipment of magnetic resonance vascular wall imaging |
| CN109840560A (en) * | 2019-01-25 | 2019-06-04 | 西安电子科技大学 | Based on the image classification method for incorporating cluster in capsule network |
Non-Patent Citations (1)
| Title |
|---|
| 李筱鹤: "非约束条件下人脸表情识别问题的研究", 《中国博士学位论文全文数据库 信息科技辑》 * |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111080168A (en) * | 2019-12-30 | 2020-04-28 | 国网江苏省电力有限公司信息通信分公司 | Power communication network equipment reliability evaluation method based on capsule network |
| CN111723840A (en) * | 2020-05-08 | 2020-09-29 | 天津大学 | Clustering and style migration method for ultrasonic images |
| CN112233106A (en) * | 2020-10-29 | 2021-01-15 | 电子科技大学中山学院 | Thyroid cancer ultrasonic image analysis method based on residual capsule network |
| CN113450325A (en) * | 2021-06-28 | 2021-09-28 | 什维新智医疗科技(上海)有限公司 | Thyroid nodule benign and malignant recognition device |
| CN115019305A (en) * | 2022-08-08 | 2022-09-06 | 成都西交智汇大数据科技有限公司 | Method, device and equipment for identifying root tip cells and readable storage medium |
| CN115019305B (en) * | 2022-08-08 | 2022-11-11 | 成都西交智汇大数据科技有限公司 | Method, device and equipment for identifying root tip cells and readable storage medium |
| CN117351012A (en) * | 2023-12-04 | 2024-01-05 | 天津医科大学第二医院 | Fetal image recognition method and system based on deep learning |
| CN117351012B (en) * | 2023-12-04 | 2024-03-12 | 天津医科大学第二医院 | Fetal image recognition method and system based on deep learning |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110321968B (en) | 2023-05-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110321968A (en) | A kind of ultrasound image sorter | |
| CN101713776B (en) | Neural network-based method for identifying and classifying visible components in urine | |
| CN102842032B (en) | Method for recognizing pornography images on mobile Internet based on multi-mode combinational strategy | |
| Omonigho et al. | Breast cancer: tumor detection in mammogram images using modified alexnet deep convolution neural network | |
| CN108520518A (en) | A kind of thyroid tumors Ultrasound Image Recognition Method and its device | |
| US20050251347A1 (en) | Automatic visual recognition of biological particles | |
| CN108364293A (en) | A kind of on-line training thyroid tumors Ultrasound Image Recognition Method and its device | |
| CN107066934A (en) | Tumor stomach cell image recognition decision maker, method and tumor stomach section identification decision equipment | |
| CN109363699A (en) | A kind of method and device of breast image lesion identification | |
| Chen et al. | Local feature based mammographic tissue pattern modelling and breast density classification | |
| CN108229674A (en) | The training method and device of cluster neural network, clustering method and device | |
| CN106874489A (en) | A kind of Lung neoplasm image block search method and device based on convolutional neural networks | |
| CN110517253A (en) | The method of the good pernicious classification of Lung neoplasm based on 3D multiple target feature learning | |
| CN109086723A (en) | A kind of method, apparatus and equipment of the Face datection based on transfer learning | |
| CN109817339A (en) | Patient's group technology and device based on big data | |
| CN108664838A (en) | Based on the monitoring scene pedestrian detection method end to end for improving RPN depth networks | |
| CN107729926A (en) | A kind of data amplification method based on higher dimensional space conversion, mechanical recognition system | |
| CN109871901A (en) | A kind of unbalanced data classification method based on mixing sampling and machine learning | |
| CN109614869A (en) | A kind of pathological image classification method based on multi-scale compress rewards and punishments network | |
| CN109363697A (en) | A kind of method and device of breast image lesion identification | |
| WO2021027152A1 (en) | Image synthesis method based on conditional generative adversarial network, and related device | |
| CN104809472A (en) | SVM-based food classifying and recognizing method | |
| Gupta et al. | Partially-independent framework for breast cancer histopathological image classification | |
| CN109461144A (en) | A kind of method and device of breast image identification | |
| CN115601751A (en) | Fundus image semantic segmentation method based on domain generalization |
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 |