CN111046951A - Medical image classification method - Google Patents

Medical image classification method Download PDF

Info

Publication number
CN111046951A
CN111046951A CN201911271721.5A CN201911271721A CN111046951A CN 111046951 A CN111046951 A CN 111046951A CN 201911271721 A CN201911271721 A CN 201911271721A CN 111046951 A CN111046951 A CN 111046951A
Authority
CN
China
Prior art keywords
matrix
medical images
samples
graph
hidden 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.)
Withdrawn
Application number
CN201911271721.5A
Other languages
Chinese (zh)
Inventor
李泽琦
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Anhui Weiaoman Robot Co ltd
Original Assignee
Anhui Weiaoman Robot Co ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Anhui Weiaoman Robot Co ltd filed Critical Anhui Weiaoman Robot Co ltd
Priority to CN201911271721.5A priority Critical patent/CN111046951A/en
Publication of CN111046951A publication Critical patent/CN111046951A/en
Withdrawn legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H30/00—ICT specially adapted for the handling or processing of medical images
    • G16H30/40—ICT specially adapted for the handling or processing of medical images for processing medical images, e.g. editing

Abstract

The invention relates to a medical image classification method, which comprises the following steps: 1) collecting a large number of medical images, extracting features to obtain a sample set, manually labeling and initializing; 2) randomly generating an input weight vector and an input bias of a hidden layer mapping function; 3) generating a hidden layer output function; 4) generating a hidden layer output matrix; 5) constructing a graph Laplace matrix; 6) and predicting by using the trained model. The invention has the following advantages: 1) only a small amount of labels are needed to be marked for the medical images by experts; 2) data distribution and prior connection information can be fully utilized to realize high-accuracy classification under a small amount of data; 3) the model training is efficient without the need to resort to large numbers of expensive high-speed computers.

Description

Medical image classification method
Technical Field
The invention belongs to the technical field of computers, and particularly relates to a medical image classification method.
Background
Medical images are increasingly used in clinical diagnosis and treatment, and how to use a large number of medical images to assist doctors in diagnosis and treatment of diseases is a problem of extensive research in the industry at present. An excellent medical image classification method must be carefully classified according to the types of diseases and the types of donors so as to perform efficient retrieval and information analysis and mining at any time. The traditional medical image mainly adopts the methods of manual identification and character classification. However, with the increase of medical images, especially the differences of race, gender, age, etc. involved therein, the difficulty of manual identification is getting greater and the workload is getting greater. Therefore, it is a future trend to introduce increasingly sophisticated computer image recognition technology to replace manual work to accomplish the above work.
Disclosure of Invention
The invention overcomes the defects of the prior art and provides a medical image classification method, which comprises the following steps:
step 1, collecting a large number of medical images
Figure BDA0002314387040000011
Using pairs of automatic encoders
Figure BDA0002314387040000012
Training to obtain
Figure BDA0002314387040000013
A set of feature vectors, i.e. a set of samples
Figure BDA0002314387040000014
To pair
Figure BDA0002314387040000015
Marking to obtain corresponding category label
Figure BDA0002314387040000016
Wherein the content of the first and second substances,
Figure BDA0002314387040000017
is a d-dimensional column vector and is,
Figure BDA0002314387040000018
in the case of a real number,
Figure BDA0002314387040000019
for the number of labeled samples and n for the number of all samples,
Figure BDA00023143870400000110
number of unlabeled samples;
initialization: the following parameters were manually set: model complexity coefficient gammaA>0, smooth conformity coefficient γI>0, ligation fusionCoefficient η ∈ (0,1), loss coefficient C>0, number of hidden layer nodes N>0；
Step 2, randomly generating an input weight vector of a hidden layer mapping function
Figure BDA00023143870400000111
And is offset from the input
Figure BDA00023143870400000112
The following were used:
randomly generating N w to obtain w1,...,wN(ii) a Randomly generating N b to obtain b1,...,bN；
Step 3, generating a hidden layer output function:
h(x)＝[G(w1,b1,x),…,G(WN,bN,x)]T
wherein G (w, b, x) is an activation function, and x represents a sample;
step 4, generating a hidden layer output matrix:
Figure BDA00023143870400000113
step 5, constructing a graph Laplace matrix:
step 501, constructing a Laplace matrix L of the feature similarity graphG：
LG＝DG-WG
Wherein, WGIs a feature similarity matrix whose i row and j column elements [ W ]G]ijComprises the following steps:
Figure BDA0002314387040000021
wherein x isiAnd xjIn order to be a sample of the sample,
Figure BDA0002314387040000022
σ>0 is the Gaussian kernel width; dGIs WGA degree matrix of (c);
step 502,Constructing a Bilink Laplace matrix Lm：
Lm＝Dm-Wm
Wherein, WmFor a must-join graph matrix, when xiAnd xjWhen it is a homogeneous sample, WmIth row and jth column element [ W ]m]ijWhen x is 1iAnd xjIf the same type of sample is unknown, [ W ]m]ij＝0；DmIs WmA degree matrix of (c);
step 503, construct the Laplace matrix L of the must-break graphc：
Lc＝Dc-Wc
Wherein, WcTo break the graph matrix, when xiAnd xjWhen it is a heterogeneous sample, WcIth row and jth column element [ W ]c]ijWhen x is 1iAnd xjIf the heterogeneous sample is unknown, [ W ]c]ij＝0；DcIs WcA degree matrix of (c);
and 6, predicting the category of the medical image by using the following models:
Figure BDA0002314387040000023
wherein, I is a unit array,
Figure BDA0002314387040000024
is a diagonal matrix in front of
Figure BDA0002314387040000025
Each diagonal element is 1, and the other diagonal elements are 0;
Figure BDA0002314387040000026
the number of marked samples, u the number of unmarked samples,
Figure BDA0002314387040000027
wherein, the activation function G (w, b, x) involved in step 3 is:
Figure BDA0002314387040000028
or:
Figure BDA0002314387040000029
wherein N is>d、
Figure BDA00023143870400000210
η∈(0.5,1)。
Wherein the auto-encoder comprises at least one convolutional layer and one pooling layer.
Compared with the prior art, the invention has the advantages that: 1) only a small amount of labels are needed to be marked for the medical images by experts; 2) data distribution and prior connection information can be fully utilized to realize high accuracy under a small amount of data; 3) the model training is efficient without the need to resort to large numbers of expensive high-speed computers.
Drawings
FIG. 1 is a flow chart of the method of the present invention;
Detailed Description
The invention is further described below with reference to examples, but the scope of the invention is not limited thereto.
As shown in fig. 1, the present invention is specifically implemented as follows:
step 1, collecting a large number of medical images
Figure BDA0002314387040000031
Using pairs of automatic encoders
Figure BDA0002314387040000032
Training to obtain
Figure BDA0002314387040000033
A set of feature vectors, i.e. a set of samples
Figure BDA0002314387040000034
To pair
Figure BDA0002314387040000035
Marking to obtain corresponding category label
Figure BDA0002314387040000036
Wherein the content of the first and second substances,
Figure BDA0002314387040000037
is a d-dimensional column vector and is,
Figure BDA0002314387040000038
in the case of a real number,
Figure BDA0002314387040000039
for the number of labeled samples and n for the number of all samples,
Figure BDA00023143870400000310
number of unlabeled samples;
initialization: the following parameters were manually set: model complexity coefficient gammaA>0, smooth conformity coefficient γI>0, connectivity fusion coefficient η ∈ (0,1), loss coefficient C>0, number of hidden layer nodes N>0；
Step 2, randomly generating an input weight vector of a hidden layer mapping function
Figure BDA00023143870400000311
And is offset from the input
Figure BDA00023143870400000312
The following were used:
randomly generating N w to obtain w1,...,wN(ii) a Randomly generating N b to obtain b1,...,bN；
Step 3, generating a hidden layer output function:
h(x)＝[G(w1,b1,x),…,G(WN,bN,x)]T
wherein G (w, b, x) is an activation function, and x represents a sample;
step 4, generating a hidden layer output matrix:
Figure BDA00023143870400000313
step 5, constructing a graph Laplace matrix:
step 501, constructing a Laplace matrix L of the feature similarity graphG：
LG＝DG-WG
Wherein, WGIs a feature similarity matrix whose i row and j column elements [ W ]G]ijComprises the following steps:
Figure BDA00023143870400000314
wherein x isiAnd xjIn order to be a sample of the sample,
Figure BDA00023143870400000315
σ>0 is the Gaussian kernel width; dGIs WGA degree matrix of (c);
step 502, construct the Laplace matrix L of the must-link graphm：
Lm＝Dm-Wm
Wherein, WmFor a must-join graph matrix, when xiAnd xjWhen it is a homogeneous sample, WmIth row and jth column element [ W ]m]ijWhen x is 1iAnd xjIf the same type of sample is unknown, [ W ]m]ij＝0；DmIs WmA degree matrix of (c);
step 503, construct the Laplace matrix L of the must-break graphc：
Lc＝Dc-Wc
Wherein, WcTo break the graph matrix, when xiAnd xjWhen it is a heterogeneous sample, WcIth row and jth column element [ W ]c]ijWhen x is 1iAnd xjIf the heterogeneous sample is unknown, [ W ]c]ij＝0；DcIs WcA degree matrix of (c);
and 6, predicting the category of the medical image by using the following models:
Figure BDA0002314387040000041
wherein, I is a unit array,
Figure BDA0002314387040000042
is a diagonal matrix in front of
Figure BDA0002314387040000048
Each diagonal element is 1, and the other diagonal elements are 0;
Figure BDA0002314387040000043
the number of marked samples, u the number of unmarked samples,
Figure BDA0002314387040000044
preferably, the activation function G (w, b, x) involved in step 3 is:
Figure BDA0002314387040000045
or:
Figure BDA0002314387040000046
preferably, N>d、
Figure BDA0002314387040000047
η∈(0.5,1)。
Preferably, the auto-encoder comprises at least one convolutional layer and one pooling layer.
The degree matrix D of the matrix W is calculated as follows, D is a diagonal matrix, the ith diagonal element D of Dii＝∑jWijWherein W isijIs the ith row and jth column element of W.
For an acquired medical image, some image preprocessing work is generally required, and the medical image to be recognized is subjected to correction, scaling, filtering and resolution adjustment.
The above examples are provided only for the purpose of describing the present invention, and are not intended to limit the scope of the present invention. The scope of the invention is defined by the appended claims. Various equivalent substitutions and modifications can be made without departing from the spirit and principles of the invention, and are intended to be within the scope of the invention.

Claims (10)

1. A medical image classification method is characterized by comprising the following steps:
step 1, collecting a large number of medical images
Figure FDA0002314387030000011
Using pairs of automatic encoders
Figure FDA0002314387030000012
Training to obtain
Figure FDA0002314387030000013
A set of feature vectors, i.e. a set of samples
Figure FDA0002314387030000014
To pair
Figure FDA0002314387030000015
Marking to obtain corresponding category label
Figure FDA0002314387030000016
Wherein the content of the first and second substances,
Figure FDA0002314387030000017
is a d-dimensional column vector and is,
Figure FDA0002314387030000018
the number of the labeled samples is a real number, l is the number of the labeled samples, n is the number of all the samples, and u-n-l is the number of the unlabeled samples;
initialization: the following parameters were manually set: model complexity coefficient gammaA>0, smooth conformity coefficient γI>0, connectivity fusion coefficient η ∈ (0,1), loss coefficient C>0, number of hidden layer nodes N>0；
Step 2, randomly generating an input weight vector of a hidden layer mapping function
Figure FDA0002314387030000019
And is offset from the input
Figure FDA00023143870300000110
The following were used:
randomly generating N w to obtain w1,...,wN(ii) a Randomly generating N b to obtain b1,...,bN；
Step 3, generating a hidden layer output function:
h(x)＝[G(w1,b1,x),…,G(WN,bN,x)]T
wherein G (w, b, x) is an activation function, and x represents a sample;
step 4, generating a hidden layer output matrix:
H＝[h(x1),…,h(xl+u)]T
step 5, constructing a graph Laplace matrix:
step 501, constructing a Laplace matrix L of the feature similarity graphG：
LG＝DG-WG
Wherein, WGIs a feature similarity matrix whose i row and j column elements [ W ]G]ijComprises the following steps:
Figure FDA00023143870300000111
wherein x isiAnd xjIs the sample, i, j ∈ {1, …, l + u }, σ>0 is the Gaussian kernel width; dGIs WGA degree matrix of (c);
step 502, construct the Laplace matrix L of the must-link graphm：
Lm＝Dm-Wm
Wherein, WmFor a must-join graph matrix, when xiAnd xjWhen it is a homogeneous sample, WmIth row and jth column element [ W ]m]ijWhen x is 1iAnd xjIf the same type of sample is unknown, [ W ]m]ij＝0；DmIs WmA degree matrix of (c);
step 503, construct the Laplace matrix L of the must-break graphc：
Lc＝Dc-Wc
Wherein, WcTo break the graph matrix, when xiAnd xjWhen it is a heterogeneous sample, WcIth row and jth column element [ W ]c]ijWhen x is 1iAnd xjIf the heterogeneous sample is unknown, [ W ]c]ij＝0；DcIs WcA degree matrix of (c);
and 6, predicting the category of the medical image by using the following models:
Figure FDA0002314387030000021
wherein, I is a unit array,
Figure FDA0002314387030000022
is a diagonal matrix, the first one diagonal elements of which are 1, and the other diagonal elements are 0; l is the number of marked samples, u is the number of unmarked samples,
Figure FDA0002314387030000023
2. a method as claimed in claim 1, wherein the activation function G (w, b, x) in step 3 is:
Figure FDA0002314387030000024
。
3. a method as claimed in claim 1, wherein the activation function G (w, b, x) in step 3 is:
Figure FDA0002314387030000025
。
4. a method for classifying medical images according to any of claims 1, 2 and 3, wherein N > d.
5. A method for classifying medical images as claimed in any one of claims 1, 2 and 3, wherein l > N.
6. A method for classifying medical images as claimed in any one of claims 1, 2 and 3, wherein η e (0.5, 1).
7. A method for classifying medical images as claimed in claim 4, wherein η e (0.5, 1).
8. A method for classifying medical images as claimed in claim 5, wherein η e (0.5, 1).
9. A method for classifying medical images according to any one of claims 1, 2 and 3, wherein said automatic encoder comprises at least one convolutional layer and one pooling layer.
10. The method of classifying medical images of claim 6, wherein said automated encoder includes at least one convolutional layer and one pooling layer.
CN201911271721.5A 2019-12-12 2019-12-12 Medical image classification method Withdrawn CN111046951A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201911271721.5A CN111046951A (en) 2019-12-12 2019-12-12 Medical image classification method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201911271721.5A CN111046951A (en) 2019-12-12 2019-12-12 Medical image classification method

Publications (1)

Publication Number Publication Date
CN111046951A true CN111046951A (en) 2020-04-21

Family

ID=70235853

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201911271721.5A Withdrawn CN111046951A (en) 2019-12-12 2019-12-12 Medical image classification method

Country Status (1)

Country Link
CN (1) CN111046951A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11257575B2 (en) * 2018-11-21 2022-02-22 Enlitic, Inc. Model-assisted annotating system

Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20140204092A1 (en) * 2012-04-09 2014-07-24 The Regents Of The University Of California Classification of high dimensional data
CN104408466A (en) * 2014-11-17 2015-03-11 中国地质大学（武汉） Semi-supervision and classification method for hyper-spectral remote sensing images based on local stream type learning composition
CN104463202A (en) * 2014-11-28 2015-03-25 苏州大学 Multi-class image semi-supervised classifying method and system
CN104881681A (en) * 2015-05-22 2015-09-02 浙江大学 Image sequence category labeling method based on mixed graph model
CN105551036A (en) * 2015-12-10 2016-05-04 中国科学院深圳先进技术研究院 Training method and device for deep learning network
CN105894517A (en) * 2016-04-22 2016-08-24 北京理工大学 CT image liver segmentation method and system based on characteristic learning
CN107392230A (en) * 2017-06-22 2017-11-24 江南大学 A kind of semi-supervision image classification method for possessing maximization knowledge utilization ability
CN108830301A (en) * 2018-05-28 2018-11-16 温州大学苍南研究院 The semi-supervised data classification method of double Laplace regularizations based on anchor graph structure
CN110363240A (en) * 2019-07-05 2019-10-22 安徽威奥曼机器人有限公司 A kind of medical image classification method and system

Patent Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20140204092A1 (en) * 2012-04-09 2014-07-24 The Regents Of The University Of California Classification of high dimensional data
CN104408466A (en) * 2014-11-17 2015-03-11 中国地质大学（武汉） Semi-supervision and classification method for hyper-spectral remote sensing images based on local stream type learning composition
CN104463202A (en) * 2014-11-28 2015-03-25 苏州大学 Multi-class image semi-supervised classifying method and system
CN104881681A (en) * 2015-05-22 2015-09-02 浙江大学 Image sequence category labeling method based on mixed graph model
CN105551036A (en) * 2015-12-10 2016-05-04 中国科学院深圳先进技术研究院 Training method and device for deep learning network
CN105894517A (en) * 2016-04-22 2016-08-24 北京理工大学 CT image liver segmentation method and system based on characteristic learning
CN107392230A (en) * 2017-06-22 2017-11-24 江南大学 A kind of semi-supervision image classification method for possessing maximization knowledge utilization ability
CN108830301A (en) * 2018-05-28 2018-11-16 温州大学苍南研究院 The semi-supervised data classification method of double Laplace regularizations based on anchor graph structure
CN110363240A (en) * 2019-07-05 2019-10-22 安徽威奥曼机器人有限公司 A kind of medical image classification method and system

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11257575B2 (en) * 2018-11-21 2022-02-22 Enlitic, Inc. Model-assisted annotating system
US20220215915A1 (en) * 2018-11-21 2022-07-07 Enlitic, Inc. Model-assisted annotating system and methods for use therewith
US11790297B2 (en) * 2018-11-21 2023-10-17 Enlitic, Inc. Model-assisted annotating system and methods for use therewith

Similar Documents

Publication Publication Date Title
Pang et al. Text matching as image recognition
Boom et al. Supporting ground-truth annotation of image datasets using clustering
CN111476315A (en) Image multi-label identification method based on statistical correlation and graph convolution technology
CN110909820A (en) Image classification method and system based on self-supervision learning
WO2009100417A2 (en) Method for training a learning machine having a deep multi-layered network with labeled and unlabeled training data
Naseer et al. Meta features-based scale invariant OCR decision making using LSTM-RNN
CN111881671B (en) Attribute word extraction method
CN110705490B (en) Visual emotion recognition method
CN110188766B (en) Image main target detection method and device based on convolutional neural network
CN113051914A (en) Enterprise hidden label extraction method and device based on multi-feature dynamic portrait
CN111062928A (en) Method for identifying lesion in medical CT image
CN112862024A (en) Text recognition method and system
CN110119448B (en) Semi-supervised cross-domain text classification method based on dual automatic encoders
CN110705384A (en) Vehicle re-identification method based on cross-domain migration enhanced representation
Tallapragada et al. Greek handwritten character recognition using inception V3
Juyal et al. Multilabel image classification using the CNN and DC-CNN model on Pascal VOC 2012 dataset
CN110363240B (en) Medical image classification method and system
CN111046951A (en) Medical image classification method
Wang et al. Crop pest detection by three-scale convolutional neural network with attention
CN111858939A (en) Text emotion classification method based on context information and convolutional neural network
Kataria et al. CNN-bidirectional LSTM based optical character recognition of Sanskrit manuscripts: A comprehensive systematic literature review
CN116486419A (en) Handwriting word recognition method based on twin convolutional neural network
Sheng et al. End-to-end chinese image text recognition with attention model
CN113158878A (en) Heterogeneous migration fault diagnosis method, system and model based on subspace
CN113449106A (en) Medical text classification method and device based on ATT-CN

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
WW01 Invention patent application withdrawn after publication
WW01 Invention patent application withdrawn after publication

Application publication date: 20200421