CN108427966A - A kind of magic magiscan and method based on PCA-LDA - Google Patents
A kind of magic magiscan and method based on PCA-LDA Download PDFInfo
- Publication number
- CN108427966A CN108427966A CN201810201133.3A CN201810201133A CN108427966A CN 108427966 A CN108427966 A CN 108427966A CN 201810201133 A CN201810201133 A CN 201810201133A CN 108427966 A CN108427966 A CN 108427966A
- Authority
- CN
- China
- Prior art keywords
- feature
- extraction
- image
- pca
- lda
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
- G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
-
- 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)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (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)
Abstract
The invention discloses a kind of magic magiscan and method based on PCA LDA, dimension-reduction treatment is carried out to pretreated feature using PCA LAD algorithms, PCA algorithms are used to carry out dimensionality reduction to pretreated feature vector first, so that the feature redundancy after dimensionality reduction reduces, and linear independence, it uses LDA algorithm to carry out dimensionality reduction again, obtains the low-dimensional feature vector of most discriminating power.It is preferred to the feature progress dimension-reduction treatment of extraction, feature using PCA LDA algorithms, realizing has the feature selecting of supervision, and the low-dimensional feature vector after dimensionality reduction has more identifiability, preferably shows the effect of classification, obtain better disaggregated model so that classification is more accurate and reliable.
Description
Technical field
The present invention relates to image group fields, and in particular to a kind of feature selecting having supervision and dimension-reduction treatment based on
The magic magiscan and method of PCA-LDA.
Background technology
Image group (Radiomics) is that image group is an emerging field[1], target is that (CT is swept from medical image
Retouch, the medical image that the modes such as positron emission tomography (PET) or magnetic resonance imaging obtain) in extraction and analysis largely there is height
The quantitative image feature of flux, and build description tumour and predict the model of clinical phenotypes, establish characteristics of image and clinical phenotypes
Or the association of gene molecule mark, and then carry out diagnosis and the clinical phenotypes prediction of tumour[2].In the clinical decision of thyroid cancer
In, the benign and malignant judgement for the state of an illness then can preferably carry out clinical decision using the method for image group.
2012, Dutch scholar Lambin was put forward for the first time image group concept, and thought source is in Tumor Heterogeneity[3]。
Lambin thinks that image group is " big measure feature to be extracted from irradiation image with high throughput, using automatically or semi-automatically analysis side
Method by iconography data be converted into it is high-resolution can mining data space ".Kumar etc.[4]It further expands, by image
Group, which is learned, to be defined as " extracted from CT, PET and MRI with high throughput and analyze a large amount of advanced Quantitative image features ".
Doroshow etc.[5]Publish an article in NatureReviewsClinicalOncology, it is indicated that image group be translational medicine not
Carry out one of developing direction.2014, the peak Radiological Society of North America (RadiologicalSocietyof NorthAmerica, RSNA)
It can theme as " Radiomics:FromClinicalImagestoOmics”.Gillies refers in conference keynote speech,
It can quantify microenvironment by the in-depth analysis to image, predict the degree of tumorgenesis heterogeneity.He thinks, compared to traditional
Clinical medicine only understands medical image from vision level, and image group, which can deeply excavate the biology essence of image and provide, faces
Bed decision support.
The general processing procedure of image group, image capturing and reconstruction, lesion segmentation, feature extraction, feature selecting, model
Structure etc..Currently, image group there are two main classes method, one kind is the method based on deep learning;Another kind of is to be based on machine
The method of study.Due to the problem that can not be explanatory of deep learning, the method for Major Epidemic machine learning.Pass through machine learning
It carries out in image group processing procedure, for different medical image means, used image group feature is also different.Have
Classified and predicted with big measure feature under limit sample, not only the calculating time is long, and effect also may not be optimal.The high pass of substantial amounts
After measuring Features extraction, feature selection approach need to be used to obtain the feature set of optimum performance performance, be input to accurate and reliable
Machine learning algorithm or statistics approach establish classification or prediction model.Therefore, feature selecting is the important step of image group
Suddenly
Using the processing of the method for machine learning expansion image group, there are two problems at present:First, unsupervised feature
Selection influences classifier performance;Second is that the sample characteristics dimension of extraction is higher (redundancy), correlation is higher, and medical image sample
This is less, and over-fitting is be easy to cause when building model.
Disadvantage one, unsupervised feature selecting influence classifier performance
Belong to two classification problems by the way that image group is general to the analysis of neoplastic problems, and tests sample used and all belong to
In the sample of tape label.Feature Dimension Reduction is also a kind of feature selecting, when carrying out Feature Dimension Reduction, has the dimensionality reduction of supervision better than no prison
The dimensionality reduction superintended and directed.Currently, method of the image group for Feature Dimension Reduction has more commonly used searching method and principal component analysis.This two
Class method belongs to unsupervised dimension reduction method, is affected for classifier performance.
Disadvantage two, characteristic dimension are high, and data volume is few, easy tos produce over-fitting
Correlation and redundancy property between feature can reduce the accuracy rate of classification, the characteristic dimension of image group extraction compared with
High (redundancy), correlation is higher, and medical image generally falls into small-sample learning, in data fitting, relative to compared with small sample
Amount, when fitting, can cause the dimension disaster of data, lead to over-fitting when building model so that the result of model is not applied for it
His data sample, the i.e. generalization ability of grader are poor.
Invention content
In order to solve technical problem present in existing image group processing procedure, the present invention provides one kind being based on PCA-
The magic magiscan and method of LDA, using PCA-LDA algorithms to the high dimensional feature vector of extraction carry out feature selecting,
Dimension-reduction treatment, realizing has the feature selecting of supervision, and the feature vector dimension obtained after dimensionality reduction is low, linear nothing between each dimension
It closes, and more discriminating power, can realize better disaggregated model, preferably show classifying quality.
On the one hand, the present invention is achieved through the following technical solutions:
A kind of magic magiscan based on PCA-LDA, including image collection module, characteristic extracting module and classification
Model building module;
The ROI that described image acquisition module is used to obtain original medical image and handle original medical image
Image;
The characteristic extracting module is used to carry out feature extraction to the original medical image and ROI image of acquisition;
The disaggregated model establishes module and carries out dimension-reduction treatment to the feature of extraction using PCA-LAD algorithms, obtains best
Feature vector verifies the sorting algorithm used based on best features vector, optimal classification algorithm is determined, to described optimal
Sorting algorithm carries out parameter regulation optimization, obtains final classification model.
Further, the feature extraction includes two parts:The extraction of Clinical symptoms and characteristics of image, the extraction of Clinical symptoms
It is the data by original medical image being identified on extraction image;The extraction of characteristics of image first carries out ROI image
Wavelet transformation, obtain wavelet transformation after image, then the ROI image to original ROI image and after wavelet transformation into
The extraction of row textural characteristics, gray level co-occurrence matrixes feature.
In order to enable subsequent characteristics selection, dimension-reduction treatment, determining that the processing procedures such as disaggregated model are more accurate reliable, into one
Step, it further includes being pre-processed to the feature vector of extraction before carrying out dimension-reduction treatment that the disaggregated model, which establishes module, institute
It includes filling default value, section scaling and standardization to state pretreatment.
Preferably, the dimension-reduction treatment carries out PCA dimensionality reductions first so that the feature redundancy after dimensionality reduction reduces, and linear
It is unrelated, then LDA dimensionality reductions are carried out, obtain the low-dimensional feature vector of most discriminating power.Using PCA-LDA algorithms to the feature of extraction
Progress dimension-reduction treatment, feature are preferred, realize there is the feature selecting of supervision, and the low-dimensional feature vector after dimensionality reduction is more recognizable
Property, it preferably shows the effect of classification, obtains better disaggregated model so that classification is more accurate and reliable.
On the other hand, the invention also provides a kind of medical image processing methods based on PCA-LDA, including following step
Suddenly:
Step 1: the ROI image for obtaining original medical image and being handled original medical image;
Step 2: the original medical image and ROI image to acquisition carry out feature extraction;
Step 3: carrying out dimension-reduction treatment to the feature of extraction using PCA-LAD algorithms, best features vector is obtained, is based on
Best features vector verifies the sorting algorithm used, determines optimal classification algorithm;
Step 4: carrying out parameter regulation optimization to the optimal classification algorithm, final classification model is obtained.
Further, feature extraction specifically includes two parts in the step 2:The extraction of Clinical symptoms and characteristics of image, faces
The extraction of bed feature is the data by original medical image being identified on extraction image;The extraction of characteristics of image is right first
ROI image carries out wavelet transformation, the image after wavelet transformation is obtained, then to original ROI image and after wavelet transformation
ROI image carry out textural characteristics, gray level co-occurrence matrixes feature extraction.
Further, carrying out dimension-reduction treatment to the feature of extraction using PCA-LAD algorithms in the step 3 is specially:First
PCA dimensionality reductions are carried out to the high dimensional feature vector of extraction so that the feature redundancy after dimensionality reduction reduces, and linear independence, then carries out
LDA dimensionality reductions obtain the low-dimensional feature vector of most discriminating power.The feature of extraction is carried out at dimensionality reduction using PCA-LDA algorithms
Reason, feature are preferred, and realizing has the feature selecting of supervision, and the low-dimensional feature vector after dimensionality reduction has more identifiability, preferably opens up
The effect for showing classification obtains better disaggregated model so that classification is more accurate and reliable.
Further, the feature vector to extraction is also needed in the step 3 before carrying out dimension-reduction treatment to the feature of extraction
It is pre-processed, specific preprocessing process includes:Fill default value, section scaling and standardization.In order to enable follow-up special
Sign selection, dimension-reduction treatment determine that the processing procedures such as disaggregated model are more accurate reliable.
The present invention has the following advantages and advantages compared to the prior art:
1, the present invention carries out feature selecting, dimension-reduction treatment using PCA-LDA algorithms to the high dimensional feature vector of extraction, realizes
A kind of having the feature selection process of supervision, and the feature vector dimension obtained after dimensionality reduction is low, linear independence between each dimension, and
More discriminating power can realize better disaggregated model, preferably show classifying quality.
2, the present invention is by there is the dimension-reduction treatment of supervision, and obtained feature vector dimension is low and linear independence, identifiability
Height can obtain better disaggregated model when building disaggregated model, more accurately, reliable show the effect of classification, and mould
The result of type can be perfectly suitable for a variety of data samples, and the generalization ability of classification is strong.
Description of the drawings
Attached drawing described herein is used for providing further understanding the embodiment of the present invention, constitutes one of the application
Point, do not constitute the restriction to the embodiment of the present invention.In the accompanying drawings:
Fig. 1 is the system construction drawing of the present invention.
Fig. 2 is flow chart of the method for the present invention.
Fig. 3 is the Medical Image Processing design sketch based on PCA algorithms.
Fig. 4 is the Medical Image Processing design sketch based on LDA algorithm.
Fig. 5 is the Medical Image Processing design sketch based on PCA-LDA algorithms.
Specific implementation mode
To make the objectives, technical solutions, and advantages of the present invention clearer, with reference to embodiment and attached drawing, to this
Invention is described in further detail, and exemplary embodiment of the invention and its explanation are only used for explaining the present invention, do not make
For limitation of the invention.
Embodiment 1
The present embodiment proposes a kind of magic magiscan based on PCA-LDA, which uses PCA-LDA algorithms
Feature selecting, dimension-reduction treatment are carried out to the high dimensional feature vector of extraction, realize a kind of feature selection process having supervision, and dimensionality reduction
The feature vector dimension that obtains afterwards is low, linear independence between each dimension, and more discriminating power, can realize better classification
Model preferably shows classifying quality.
As shown in Figure 1, the system, which specifically includes image collection module, characteristic extracting module and disaggregated model, establishes module;
The ROI that described image acquisition module is used to obtain original medical image and handle original medical image
Image;
The characteristic extracting module is used to carry out feature extraction to the original medical image and ROI image of acquisition;
Feature extraction in the present embodiment includes two parts:The extraction of Clinical symptoms and characteristics of image, Clinical symptoms carry
Take is by the way that the data extracted on image are identified to original medical image;The extraction of characteristics of image first to ROI image into
Row wavelet transformation obtains the image after wavelet transformation, then the ROI image to original ROI image and after wavelet transformation
Carry out the extraction of textural characteristics, gray level co-occurrence matrixes feature.
The disaggregated model establishes module and includes preprocessing module, Feature Dimension Reduction module, sorting algorithm authentication module and divide
Class model optimizes determining module.
The feature vector that the preprocessing module extracts characteristic extracting module pre-processes, and preprocessing process specifically wraps
It includes:Fill default value, section scaling and standardization;So that subsequent characteristics selection, dimension-reduction treatment, determining disaggregated model etc.
Processing procedure is more accurate reliable.
The Feature Dimension Reduction module carries out dimension-reduction treatment using PCA-LAD algorithms to pretreated feature, uses first
PCA algorithms carry out dimensionality reduction to pretreated feature vector so that and the feature redundancy after dimensionality reduction reduces, and linear independence, then
Dimensionality reduction is carried out using LDA algorithm, obtains the low-dimensional feature vector of most discriminating power.Using PCA-LDA algorithms to the spy of extraction
Sign progress dimension-reduction treatment, feature are preferred, realize there is the feature selecting of supervision, and the low-dimensional feature vector after dimensionality reduction is more recognizable
Property, it preferably shows the effect of classification, obtains better disaggregated model so that classification is more accurate and reliable.Specific PCA algorithms
It is as follows with LDA algorithm principle:
PCA algorithm principles
Principal component analysis (Principal Components Analysis, PCA) is to apply more dimension reduction method.It can
So that the feature space that high dimensional data is greatly reduced from original image spatial transformation for dimension, meanwhile, and retain original image number
According to most information.PCA algorithms are a kind of methods that data are analyzed in statistics, and the method is found in data space
One group of vector expresses the variance of data with this vector as far as possible, data is dropped to low latitude from higher-dimension, PCA algorithms are become using K-L
It changes and obtains the most low-dimensional identification space for approaching artwork image space.
The feature extracted in the process to image group using PCA algorithms is standardized:
In formula, X is the feature set of training sample, and size is N × P, and N is medical image sample size, and P is that training sample is special
The dimension of collection,For the mean value of training sample feature set, D is variance.
The principal component of training sample feature set can be calculate by the following formula to obtain:
UT(XXT) U=∧
In formula, ∧ be eigenvalue cluster corresponding to X at diagonal matrix, U is characterized what the corresponding feature vector of value formed
Orthogonal matrix.
Characteristic value is λi=(i=1,2 ..., N), λ1≥λ2≥...≥λNFeature vector is Ui(i=1,2 ... N), then:
Selection before a larger eigenvalue clusters of m (m is far smaller than N) at feature vector UmTo calculate the principal component of sample.Cause
This eigenmatrix for obtaining m training sample is:
W=UmX
Then training sample feature set X is projected as on training sample feature set subspace:
P=WXT
Thereby realize Feature Dimension Reduction.
LDA algorithm principle
Linear discriminant analysis (Linear Discriminant Analysis, LDA), it take full advantage of training sample oneself
The classification information known finds the projecting direction subspace most helpful in identification and classification, belongs to supervised learning method.Its purpose is
The low latitude feature of most discriminating power is extracted from high-dimensional feature space, these features can be helped the other institute of same class
There is sample to flock together, different classes of sample is separated as possible, i.e., selection is so that between-class scatter SBIn sample class
Dispersion SwThe maximum feature of ratio.Within class scatter matrix is to seek average value in class to the training sample of every one kind, then use
Each sample subtracts the mean value of respectively affiliated class.Matrix between samples SBWith dispersion matrix between sample classes SwDefinition
See below two formulas:
Wherein, C is classification number, PiIt is prior probability, μiIt is CiThe mean value of class sample, μ are the mean value of population sample, xk
It is k-th of feature of the i-th class sample, CiIt is the sample for belonging to the i-th class.
The sample that the projection line obtained by PCA methods allows for after projection can not divide again, and obtained by LDA methods
Projection line makes the sample after projection still have good separability.
The sample different classes of in low latitude space is got as far as possible after projection the more opens the more good, at the same time it is wished that inside each classification
Sample is intensive as possible, that is to say, that between-class scatter is the bigger the better, and within-class scatter is the smaller the better.Therefore, such as
Fruit SwIt is nonsingular matrix, optimal projecting direction W is exactly so that matrix between samples and within-class scatter square
Those of the determinant ratio maximum of battle array orthogonal eigenvectors, therefore, optimum mapping function is defined as:
By linear algebra theory it is found that W is exactly the solution for meeting following equation:
SBWi=λiSWWi(i=1,2 ..., m)
Namely correspond to matrixLarger eigenvalue λiFeature vector.
In the present embodiment, the sorting algorithm authentication module is based on after Feature Dimension Reduction module carries out PCA-LDA dimensionality reductions
Obtained low-dimensional feature vector verifies used sorting algorithm using the auxiliary tool in scikit-learn, choosing
The best sorting algorithm of classifying quality is selected, selection logistics regression algorithms are verified in scikit-learn in the present embodiment
As optimal classification algorithm.
In the present embodiment, the disaggregated model optimization determining module returns the optimal classification algorithm logistics of selection
Sorting algorithm carries out parameter optimization adjusting, the final classification prediction model for determining that classifying quality is optimal.
Embodiment 2
The magic magiscan of corresponding above-described embodiment, as shown in Fig. 2, the present invention, which proposes one kind, being based on PCA-
The medical image processing method of LDA, includes the following steps:
S01, the ROI image for obtaining original medical image and original medical image being handled;
S02, feature extraction is carried out to the original medical image and ROI image of acquisition;
Including two parts:The extraction of the extraction of Clinical symptoms and characteristics of image, Clinical symptoms is by primitive medicine figure
As the data on extraction image are identified;The extraction of characteristics of image carries out wavelet transformation to ROI image first, obtains small echo and becomes
Image after changing, then the ROI image progress textural characteristics to original ROI image and after wavelet transformation, gray scale symbiosis
The extraction of matrix character.
S03, in order to enable subsequent characteristics selection, dimension-reduction treatment, determining that the processing procedures such as disaggregated model are more accurate reliable;
The feature of extraction is carried out also needing to pre-process the feature vector of extraction before dimension-reduction treatment, specific preprocessing process packet
It includes:Fill default value, section scaling and standardization.Pretreated high dimensional feature vector is carried out using PCA-LAD algorithms
PCA dimensionality reductions so that the feature redundancy after dimensionality reduction reduces, and linear independence, then carries out LDA dimensionality reductions, obtains most discriminating power
Low-dimensional feature vector;Based on the low-dimensional feature vector of the more identity obtained after above-mentioned dimension-reduction treatment, scikit- is utilized
Auxiliary tool in learn assesses used sorting algorithm, selects logistic to return in scikit-learn
Algorithm is classified.
It is preferred to the feature progress dimension-reduction treatment of extraction, feature using PCA-LDA algorithms, realize that the feature for having supervision is selected
It selects, and the low-dimensional feature vector after dimensionality reduction has more identifiability, preferably shows the effect of classification, obtains mould of preferably classifying
Type so that classification is more accurate and reliable.
S04, to needed for logistic regression algorithms using to hyper parameter carry out tune ginseng, finally determine disaggregated model.
Embodiment 3
To one group of medical oncology image carried out the Medical Image Processing based on PCA, the Medical Image Processing based on LDA and
Medical Image Processing using the present invention based on PCA-LDA has carried out the processing of image group, obtains such as table 1-1 to 1-3 pairs of table
The classifying quality figure of the positive and negative sample test result and corresponding Fig. 3-5 answered.
Table 1-1
Table 1-2
Table 1-3
Table 1-1 is to predict that obtained test accuracy rate is 0.650000, such as Fig. 3 institutes only with the classification that PCA algorithms carry out
Show;Table 1-2 is to be predicted only with the classification that LDA algorithm carries out, and obtained test accuracy rate is 0.600000, as shown in Figure 4;Table
1-3 is using the present invention, that is, the classification prediction for using PCA-LDA algorithms to carry out, obtained test accuracy rate is 0.850000, such as
Shown in Fig. 5;It follows that the present invention uses the Feature Dimension Reduction for having supervision carried out based on PCA-LDA algorithms to handle, selection is most
The low-dimensional feature vector of discriminating power, obtained disaggregated model is more excellent, substantially increases the accuracy and reliability of classification.
Above-described specific implementation mode has carried out further the purpose of the present invention, technical solution and advantageous effect
It is described in detail, it should be understood that the foregoing is merely the specific implementation mode of the present invention, is not intended to limit the present invention
Protection domain, all within the spirits and principles of the present invention, any modification, equivalent substitution, improvement and etc. done should all include
Within protection scope of the present invention.
Claims (8)
1. a kind of magic magiscan based on PCA-LDA, which is characterized in that including image collection module, feature extraction
Module and disaggregated model establish module;
The ROI that described image acquisition module is used to obtain original medical image and is handled original medical image schemes
Picture;
The characteristic extracting module is used to carry out feature extraction to the original medical image and ROI image of acquisition;
The disaggregated model establishes module and carries out dimension-reduction treatment to the feature of extraction using PCA-LAD algorithms, obtains best features
Vector verifies the sorting algorithm used based on best features vector, optimal classification algorithm is determined, to the optimal classification
Algorithm carries out parameter regulation optimization, obtains final classification model.
2. system according to claim 1, which is characterized in that the feature extraction includes two parts:Clinical symptoms and figure
As the extraction of feature, the extraction of Clinical symptoms is the data by original medical image being identified on extraction image;Image
The extraction of feature first to ROI image carry out wavelet transformation, obtain wavelet transformation after image, then to original ROI image with
And the ROI image after wavelet transformation carries out the extraction of textural characteristics, gray level co-occurrence matrixes feature.
3. system according to claim 1, which is characterized in that it further includes carrying out dimensionality reduction that the disaggregated model, which establishes module,
The feature vector of extraction is pre-processed before processing, the pretreatment includes filling default value, section scaling and standard
Change.
4. system according to claim 1, which is characterized in that the dimension-reduction treatment carries out PCA dimensionality reductions first so that dimensionality reduction
Feature redundancy afterwards reduces, and linear independence, then carries out LDA dimensionality reductions, obtains the low-dimensional feature vector of most discriminating power.
5. a kind of medical image processing method based on PCA-LDA, which is characterized in that include the following steps:
Step 1: the ROI image for obtaining original medical image and being handled original medical image;
Step 2: the original medical image and ROI image to acquisition carry out feature extraction;
Step 3: carrying out dimension-reduction treatment to the feature of extraction using PCA-LAD algorithms, best features vector is obtained, based on best
Feature vector verifies the sorting algorithm used, determines optimal classification algorithm;
Step 4: carrying out parameter regulation optimization to the optimal classification algorithm, final classification model is obtained.
6. according to the method described in claim 5, it is characterized in that, in the step 2 feature extraction specifically include two parts:
The extraction of the extraction of Clinical symptoms and characteristics of image, Clinical symptoms is by the way that original medical image is identified on extraction image
Data;The extraction of characteristics of image carries out wavelet transformation to ROI image first, the image after wavelet transformation is obtained, then to original
Beginning ROI image and the ROI image after wavelet transformation carry out the extraction of textural characteristics, gray level co-occurrence matrixes feature.
7. according to the method described in claim 5, it is characterized in that, using PCA-LAD algorithms to extraction in the step 3
Feature carries out dimension-reduction treatment:PCA dimensionality reductions are carried out to the high dimensional feature vector of extraction first so that the feature after dimensionality reduction is superfluous
Remaining reduces, and linear independence, then carries out LDA dimensionality reductions, obtains the low-dimensional feature vector of most discriminating power.
8. according to the method described in claim 5, it is characterized in that, being carried out at dimensionality reduction in the feature to extraction in the step 3
It also needs to pre-process the feature vector of extraction before reason, specific preprocessing process includes:Fill default value, section scaling
And standardization.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810201133.3A CN108427966A (en) | 2018-03-12 | 2018-03-12 | A kind of magic magiscan and method based on PCA-LDA |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810201133.3A CN108427966A (en) | 2018-03-12 | 2018-03-12 | A kind of magic magiscan and method based on PCA-LDA |
Publications (1)
Publication Number | Publication Date |
---|---|
CN108427966A true CN108427966A (en) | 2018-08-21 |
Family
ID=63157827
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810201133.3A Pending CN108427966A (en) | 2018-03-12 | 2018-03-12 | A kind of magic magiscan and method based on PCA-LDA |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108427966A (en) |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109255313A (en) * | 2018-08-30 | 2019-01-22 | 中国科学院国家授时中心 | A kind of method of promotion signal recognition correct rate |
CN109901064A (en) * | 2019-03-15 | 2019-06-18 | 西安工程大学 | Fault Diagnosis for HV Circuit Breakers method based on ICA-LVQ |
CN110210559A (en) * | 2019-05-31 | 2019-09-06 | 北京小米移动软件有限公司 | Object screening technique and device, storage medium |
CN111292291A (en) * | 2020-01-15 | 2020-06-16 | 李东航 | Cancer CT image judgment device based on PCA or M wave band wavelet domain |
WO2020186432A1 (en) * | 2019-03-15 | 2020-09-24 | Hong Kong Applied Science and Technology Research Institute Company Limited | Apparatus and method of high dimensional data analysis in real-time |
CN111709492A (en) * | 2020-07-03 | 2020-09-25 | 大连东软教育科技集团有限公司 | Dimension reduction visualization method and device for high-dimensional electronic medical record list and storage medium |
CN111860529A (en) * | 2020-07-29 | 2020-10-30 | 山东超越数控电子股份有限公司 | Image preprocessing method, system, device and medium |
CN114127712A (en) * | 2019-05-15 | 2022-03-01 | 雷克斯股份有限公司 | System and method for generating a low dimensional space representing similarity between patents |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101093539A (en) * | 2007-07-27 | 2007-12-26 | 哈尔滨工程大学 | Matching identification method by extracting characters of vein from finger |
CN101819629A (en) * | 2010-04-08 | 2010-09-01 | 西京学院 | Supervising tensor manifold learning-based palmprint identification system and method |
CN102662167A (en) * | 2012-04-11 | 2012-09-12 | 西北工业大学 | Feature extraction method of radiated noise signal of underwater target |
CN102982322A (en) * | 2012-12-07 | 2013-03-20 | 大连大学 | Face recognition method based on PCA (principal component analysis) image reconstruction and LDA (linear discriminant analysis) |
CN103593844A (en) * | 2013-10-29 | 2014-02-19 | 华中科技大学 | Extraction method of multiple multi-dimensional features of medical images |
CN104102839A (en) * | 2013-11-16 | 2014-10-15 | 西安电子科技大学 | Alzheimer's disease cortex auto-classification method based on multi-scale mesh surface form features |
CN104700089A (en) * | 2015-03-24 | 2015-06-10 | 江南大学 | Face identification method based on Gabor wavelet and SB2DLPP |
CN106548178A (en) * | 2016-09-26 | 2017-03-29 | 深圳大学 | A kind of semantic feature auto-scoring method and system based on Lung neoplasm CT images |
-
2018
- 2018-03-12 CN CN201810201133.3A patent/CN108427966A/en active Pending
Patent Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101093539A (en) * | 2007-07-27 | 2007-12-26 | 哈尔滨工程大学 | Matching identification method by extracting characters of vein from finger |
CN101819629A (en) * | 2010-04-08 | 2010-09-01 | 西京学院 | Supervising tensor manifold learning-based palmprint identification system and method |
CN102662167A (en) * | 2012-04-11 | 2012-09-12 | 西北工业大学 | Feature extraction method of radiated noise signal of underwater target |
CN102982322A (en) * | 2012-12-07 | 2013-03-20 | 大连大学 | Face recognition method based on PCA (principal component analysis) image reconstruction and LDA (linear discriminant analysis) |
CN103593844A (en) * | 2013-10-29 | 2014-02-19 | 华中科技大学 | Extraction method of multiple multi-dimensional features of medical images |
CN104102839A (en) * | 2013-11-16 | 2014-10-15 | 西安电子科技大学 | Alzheimer's disease cortex auto-classification method based on multi-scale mesh surface form features |
CN104700089A (en) * | 2015-03-24 | 2015-06-10 | 江南大学 | Face identification method based on Gabor wavelet and SB2DLPP |
CN106548178A (en) * | 2016-09-26 | 2017-03-29 | 深圳大学 | A kind of semantic feature auto-scoring method and system based on Lung neoplasm CT images |
Cited By (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109255313A (en) * | 2018-08-30 | 2019-01-22 | 中国科学院国家授时中心 | A kind of method of promotion signal recognition correct rate |
CN109901064A (en) * | 2019-03-15 | 2019-06-18 | 西安工程大学 | Fault Diagnosis for HV Circuit Breakers method based on ICA-LVQ |
WO2020186432A1 (en) * | 2019-03-15 | 2020-09-24 | Hong Kong Applied Science and Technology Research Institute Company Limited | Apparatus and method of high dimensional data analysis in real-time |
US11494690B2 (en) | 2019-03-15 | 2022-11-08 | Hong Kong Applied Science and Technology Research Institute Company Limited | Apparatus and method of high dimensional data analysis in real-time |
CN114127712A (en) * | 2019-05-15 | 2022-03-01 | 雷克斯股份有限公司 | System and method for generating a low dimensional space representing similarity between patents |
CN114127712B (en) * | 2019-05-15 | 2024-01-05 | 雷克斯股份有限公司 | System and method for generating a low-dimensional space representing similarity between patents |
CN110210559A (en) * | 2019-05-31 | 2019-09-06 | 北京小米移动软件有限公司 | Object screening technique and device, storage medium |
CN110210559B (en) * | 2019-05-31 | 2021-10-08 | 北京小米移动软件有限公司 | Object screening method and device and storage medium |
CN111292291A (en) * | 2020-01-15 | 2020-06-16 | 李东航 | Cancer CT image judgment device based on PCA or M wave band wavelet domain |
CN111292291B (en) * | 2020-01-15 | 2024-02-02 | 李东航 | Cancer CT image judging device based on PCA or M wave band wavelet domain |
CN111709492A (en) * | 2020-07-03 | 2020-09-25 | 大连东软教育科技集团有限公司 | Dimension reduction visualization method and device for high-dimensional electronic medical record list and storage medium |
CN111860529A (en) * | 2020-07-29 | 2020-10-30 | 山东超越数控电子股份有限公司 | Image preprocessing method, system, device and medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108427966A (en) | A kind of magic magiscan and method based on PCA-LDA | |
US11449985B2 (en) | Computer vision for cancerous tissue recognition | |
CN112116605B (en) | Pancreas CT image segmentation method based on integrated depth convolution neural network | |
Ortiz et al. | Improving MR brain image segmentation using self-organising maps and entropy-gradient clustering | |
Gurcan et al. | Histopathological image analysis: A review | |
Lopes et al. | Prostate cancer characterization on MR images using fractal features | |
Basavanhally et al. | Computerized image-based detection and grading of lymphocytic infiltration in HER2+ breast cancer histopathology | |
US8295575B2 (en) | Computer assisted diagnosis (CAD) of cancer using multi-functional, multi-modal in-vivo magnetic resonance spectroscopy (MRS) and imaging (MRI) | |
Xue et al. | Application of deep learning in automated analysis of molecular images in cancer: a survey | |
US9025841B2 (en) | Method and system for segmentation of the prostate in 3D magnetic resonance images | |
Zheng et al. | STEP: Spatiotemporal enhancement pattern for MR‐based breast tumor diagnosis | |
Viswanath et al. | Enhanced multi-protocol analysis via intelligent supervised embedding (EMPrAvISE): detecting prostate cancer on multi-parametric MRI | |
Akselrod-Ballin et al. | Multimodal correlative preclinical whole body imaging and segmentation | |
Khotanlou et al. | Segmentation of multiple sclerosis lesions in brain MR images using spatially constrained possibilistic fuzzy C-means classification | |
Davarpanah et al. | Spatial possibilistic fuzzy C-mean segmentation algorithm integrated with brain mid-sagittal surface information | |
Escobar et al. | Voxel‐wise supervised analysis of tumors with multimodal engineered features to highlight interpretable biological patterns | |
Dayarathna et al. | Deep learning based synthesis of MRI, CT and PET: Review and analysis | |
Zheng et al. | Integrating semi-supervised label propagation and random forests for multi-atlas based hippocampus segmentation | |
CN111583192B (en) | MRI image and deep learning breast cancer image processing method and early screening system | |
Xiang et al. | Segmentation method of multiple sclerosis lesions based on 3D‐CNN networks | |
Gooya et al. | An EM algorithm for brain tumor image registration: A tumor growth modeling based approach | |
Platero et al. | A label fusion method using conditional random fields with higher-order potentials: Application to hippocampal segmentation | |
Menze et al. | Proceedings of the miccai challenge on multimodal brain tumor image segmentation (brats) 2012 | |
Landman et al. | Multiatlas segmentation | |
Illan et al. | Machine learning for accurate differentiation of benign and malignant breast tumors presenting as non-mass enhancement |
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 | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180821 |