CN105931224A - Pathology identification method for routine scan CT image of liver based on random forests - Google Patents
Pathology identification method for routine scan CT image of liver based on random forests Download PDFInfo
- Publication number
- CN105931224A CN105931224A CN201610231280.6A CN201610231280A CN105931224A CN 105931224 A CN105931224 A CN 105931224A CN 201610231280 A CN201610231280 A CN 201610231280A CN 105931224 A CN105931224 A CN 105931224A
- Authority
- CN
- China
- Prior art keywords
- image
- feature
- pathological changes
- liver
- characteristic
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10072—Tomographic images
- G06T2207/10081—Computed x-ray tomography [CT]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30056—Liver; Hepatic
Abstract
The invention discloses a pathology identification method for a routine scan CT image of the liver based on the random forests. The method comprises that image gray-level texture characteristic is extracted from a pathologic area of the routine scan CT image of the liver and serves as image characteristic vector expression, the random forests is used to select characteristics from the image characteristic vector of the pathologic area of the routine scan CT image of the liver to form a most effective characteristic combination, a most effective characteristic data set is trained and learned, the identification capability of a decision tree of random forests is balanced and optimized, and a final pathology identification model is obtained.
Description
Technical field
The present invention relates to a kind of liver plain CT image pathological changes recognition methods based on random forests algorithm, particularly to
The introducing of most effective feature selection approach and the improvement of random forests algorithm.
Background technology
Along with development and the maturation of medical imaging technology, medical image serves important work in diagnosis for liver disease
With.At present, hepatocarcinoma has become as one of the highest disease of fatality rate in the world, because the shortage for the treatment of means and early hepatocarcinoma
Pathological index is less obvious, is likely to result in mistaken diagnosis, thus misses optimal treatment time.Making a definite diagnosis of hepatocarcinoma relies primarily on liver
Dirty biopsy technique, but patient's liver can be caused certain damage by this technology, implements difficulty than high, post-operative recovery in addition
Slowly, therefore, diagnosis for hepatic disease at present is the most also to rely on medical image, such as Hepatic CT.
Summary of the invention
The technical problem to be solved in the present invention is to provide a kind of liver plain CT image pathological changes based on random forests algorithm
Recognition methods.
In order to solve above-mentioned technical problem, the present invention provides following technical scheme:
Present invention liver to be accomplished that plain CT image local pathological changes recognition methods, is specially and extracts liver plain CT figure
As the gradation of image textural characteristics of lesion region represents as image feature vector, then use random forests algorithm to Hepatic CT
Image lesion region image feature vector carries out feature selection, selects the combination of maximally effective feature, then to most effective feature
Data set is trained and learns, and the decision tree of random forest is identified ability balance optimizing, obtains final pathological changes
Identify model.
The block diagram that pathological changes identification model is set up as shown in Figure 1, is specifically divided into following steps:
1) CT image for liver lesion region characteristic data set is set up
For gradation of image textural characteristics, there are grey level histogram, gray level co-occurrence matrixes and gray scale ladder in image processing field
Degree three kinds of character representation methods of co-occurrence matrix.The diseased region of doctor's mark is extracted from the liver plain CT image of doctor's mark
Territory, extracts region as characteristics of lesion can cover the rectangle frame of lesion region, characteristics of lesion is extracted extracted region based on ash
Degree rectangular histogram, gray level co-occurrence matrixes and the characteristics of image of Gray level-gradient co-occurrence matrix.
A. grey level histogram feature extraction
Grey level histogram is for representing intensity profile and the statistical property of image, characteristics of image bag based on grey level histogram
Containing average, variance, skewness, kurtosis, energy, entropy etc..
B. gray level co-occurrence matrixes feature extraction
Gray level co-occurrence matrixes is for describing the gray-scale relation of neighbor in gray level image, figure based on gray level co-occurrence matrixes
As feature comprises angle second moment, contrast, unfavourable balance square, entropy, is correlated with.
C. Gray level-gradient co-occurrence matrix feature extraction
Gray level-gradient co-occurrence matrix features image slices vegetarian refreshments gray value and the mutual relation of Grad, depicts in image
Portion's pixel gray scale and the distribution situation of gradient, and embody a pixel and local, the space letter of pixel in its neighborhood
Breath, can well express the textural characteristics of image, and characteristics of image based on Gray level-gradient co-occurrence matrix comprises little gradient advantage, big
Gradient advantage, the inhomogeneities of intensity profile, the inhomogeneities of Gradient distribution, energy, gray scale is average, gradient is average, gray scale is equal
Variance, gradient mean square deviation, degree of association, gray level entropy, gradient entropy, the entropy of mixing, inertia, unfavourable balance square etc..
Combine the characteristics of image of above-mentioned three types, and combine the lesion type label of image as characteristic vector data collection
D。
2) most effective feature is selected
In step 1) in extract this three category feature in, comprise 26 eigenvalues altogether, these 26 eigenvalues are not all
Feature can embody the specificity of liver plain CT image local characteristics of lesion, so when selecting feature, probably due to select
Bad feature causes the identification Model Identification effect obtained poor, so the selecting for the identification established of validity feature
Model is most important, and can reduce the amount of calculation of algorithm.
The random forests algorithm that the present invention uses can provide the importance degree of each feature in characteristic vector, by following
Ring iterative rejects least key character, and the residue character after rejecting feature is set up new Random Forest model, finds out extensive
The model characteristic of correspondence combination that error is minimum, is the combination of maximally effective feature, and detailed iterative process is shown in specific embodiments.
3) foundation of random forest pathological changes identification model and improvement
For 2) step obtain most effective characteristic attribute combination, from initial characteristic data concentrate filter out most effective feature
Data set.Set up Random Forest model with most effective characteristic data set, and the decision tree in Random Forest model is optimized
And equilibrium, obtain final random forest pathological changes identification model.Concrete optimization method is shown in specific embodiments.
In the Random Forest model generated, comprising many decision trees, in these decision trees, some decision-tree model is known
Other effect is preferable, and some decision-tree model recognition effect is poor, therefore can reject the decision tree that those recognition effects are poor, but
The present invention in view of when the decision tree that screenability is higher, due to variety classes liver local patholoic change classification and recognition not
With, when screening decision tree, not from overall OOB estimation, but from the OOB estimation of single class specimen discerning effect, right
Each type focus characteristic, selects the decision tree the highest to single class classification performance of equivalent amount, new with these decision trees composition
Random forest.As a example by the random forest generating 40 decision trees, in 40 decision trees, having 10 is to normal type liver
Dirty feature identification OOB estimates optimum decision tree, has 10 hepatic haemangioma type characteristics of lesion identification OOB is estimated optimum
Decision tree, having 10 is that hepatic cyst type characteristics of lesion identification OOB is estimated optimum decision tree, and having 10 is to hepatocarcinoma type
Characteristics of lesion identification OOB estimates optimum decision tree, selects to optimize and after equilibrium through decision tree, can avoid overall situation screening
Decision tree causes the defect that certain type characteristics of lesion accuracy of identification is on the low side.
The present invention is accomplished that the automatic identification to liver plain CT image local pathological changes, mainly studies hepatocarcinoma, liver blood vessel
Several lesion type such as tumor, hepatic cyst, lesion region is embodied in grey scale change and texture variations with the image difference of normal region
On, what current gradation of image textural characteristics was conventional is based on grey level histogram, gray level co-occurrence matrixes and Gray level-gradient co-occurrence matrix
Characteristics of image, by extract suspected lesion area-of-interest as the character representation of image, on this basis, carry according to feature
Take algorithm and obtain the characteristic that quantizes, then use the random forest sorting algorithm of improvement characteristics of image to be trained, learns
Practising and prediction, recognition result can give doctor some diagnostic recommendations, although auxiliary diagnostic result cannot function as diagnostic criteria, but can
Obtaining more scientific diagnosis to be combined with doctor personal experience, thus reduce the error rate of diagnosis, this examined for the early stage of hepatocarcinoma
Disconnected have huge medical value.
The random forests algorithm that the present invention uses is a kind of integrated learning approach, with Bayes, neutral net, decision tree,
The single classifier machine learning algorithms such as support vector machine are compared, it is not easy to over-fitting problem, the study of single classifier model occur
Power limitations is in overall data sample, although can learn the data characteristics to whole data sample well, but it cannot be guaranteed that relatively
Strong generalization ability, i.e. lacks good predictive ability to unknown data sample.By contrast, random forests algorithm can solve
This problem, by integrated multiple Weak Classifiers, overcomes the defect of single grader learning capacity, uses bagging technology
Making each single decision tree classifier have the learning capacity that Partial Feature is stronger, the most each single classifier has local feature
Strong learning capacity rather than the learning capacity of global feature, each single classifier be responsible for learn Partial Feature, be combined with multiple
The random forests algorithm of single classifier just has higher learning capacity, and therefore single classifier algorithm can be compared to a comprehensive energy
The Learning machine that power is stronger, and random forest is equivalent to the combination learning machine of multiple expert composition, limited at data sample
In the case of, random forests algorithm has clear superiority, therefore the present invention uses random forests algorithm as the taxonomy of pathological changes identification
Practise device.
The innovative point of the present invention is that the introducing of most effective feature selection approach, decision tree select to optimize and decision tree identification
The improved methods such as ability equilibrium, the method for improvement has for the identification of liver plain CT image local pathological changes and preferably identifies standard
Really rate.
Accompanying drawing explanation
Below in conjunction with the accompanying drawings the detailed description of the invention of the present invention is described in further detail.
Fig. 1 is random forest pathological changes identification model；
Fig. 2 is the sequence of feature importance degree；
Fig. 3 is the impact that OOB is estimated by feature selection；
Fig. 4 is that the random forest improved compares with primal algorithm classification performance；
Fig. 5 is CT image for liver pathological changes identification operating process.
Detailed description of the invention
For realizing the liver plain CT image local pathological changes identification of the present invention, following two stages are used to carry out.
First stage: pathological changes identification model based on random forests algorithm is set up
1. liver plain CT image local lesion region image feature data collection is set up:
The present invention uses 3000 liver plain CT images from Hangzhou hospital mark, and size is 512*512, bag
Containing several types such as normal, hepatocarcinoma, hepatic haemangioma and hepatics cyst, extract the rectangle frame conduct of the lesion region that can cover mark
Area-of-interest, so obtains 3000 pathological changes tile images.
To each lesion region block, calculate the grey level histogram of pathological changes tile images, gray level co-occurrence matrixes and gray scale ladder
Degree co-occurrence matrix.
For pathological changes tile images, grey level histogram matrix H computational methods are as follows:
Wherein N is pathological changes block image pixel number amount, and L is gray level, and value of the present invention is 256, niRepresent pathological changes block
In image, gray level is the number of pixels of i, and grey level histogram H (i) represents that the number of pixels with certain gray level accounts for image pixel
The ratio of sum, the global characteristics illustrating image describes, and the size of H-matrix is 256*1.
For pathological changes tile images, the computational methods of gray level co-occurrence matrixes P are as follows:
P(Ii, I2)=P1(I1, I2)+P2(I1, I2)+P3(I1, I2)+P4(I1, I2)
Wherein I1, I2For grey scale pixel value, grey level L takes 256, P1(I1, I2) represent in pathological changes tile images level away from
It is respectively I from for 1 and 2 grey scale pixel values1, I2Pixel number account for pixel value to sum ratio, P2(I1, I2) represent
In pathological changes tile images, diagonal distance is 1 and 2 grey scale pixel values are respectively I1, I2Pixel number account for pixel value to always
The ratio of number, P3(I1, I2) represent that in pathological changes tile images, vertical dimension is 1 and 2 grey scale pixel values are respectively I1, I2Picture
Vegetarian refreshments number accounts for the pixel value ratio to sum, P4(I1, I2) represent in pathological changes tile images that back-diagonal distance is 1 and 2 point
Grey scale pixel value is respectively I1, I2Pixel number account for the pixel value ratio to sum, computing formula is respectively as follows:
Level:
Diagonal:
Vertical:
Back-diagonal:
Gray level co-occurrence matrixes P (I1, I2) then represent distance in pathological changes tile images be 1 and gray value be respectively I1, I2's
The number of two pixels pair accounts for pixel that distance is 1 ratio to sum.It is 256 that the present invention takes gray level, then distance is 1
Pixel be 65536 to sum, the size of P matrix is 256*256.
For pathological changes tile images, the computational methods of Gray level-gradient co-occurrence matrix T are as follows:
Gray value grey level LfTake 256, for a pixel (i, j), gradient calculation method is as follows:
gx(i, j)=f (i+1, j-1)+2f (i+1, j)+f (i+1, j+1)-f (i-1, j-1)-2f (i-1, j)-f (i-1, j
+1)
gy(i, j)=f (i-1, j+1)+2f (i, j+1)+f (i+1, j+1)-f (i-1, j+1)-2f (i, j-1)-f (i+1,
i-1)
Normalized gradient matrix G is:
G (i, j)=INT (g (i, j) × Lg/gM)+1
Wherein gradient number of stages is Lg, value of the present invention is 32, and image greatest gradient value is gM, INT is rounding operation, as
The Grad of vegetarian refreshments is i.e. defined by gradient matrix G, and the size of G matrix is 256*32.
Grey level histogram matrix H, gray level co-occurrence matrixes P and shade of gray according to pathological changes tile images derived above is altogether
Raw matrix T, calculates gradation of image textural characteristics based on these three matrix, and computational methods are respectively shown in following three form:
Table 1 grey level histogram characteristic measure
Table 2 gray level co-occurrence matrixes characteristic measure
Table 3 Gray level-gradient co-occurrence matrix characteristic measure
The most just the characteristic vector of pathological changes tile images has been obtained:
[f1,f2,f3,f4,f5,f6,f7,f8,f9,f10,f11,f12,f13,f14,f15,f16,f17,f18,f19,
f20,f21,f22,f23,f24,f25,f26,label]
Label is pathological changes label, and 1 represents normal, and 2 represent hepatocarcinoma, and 3 represent hepatic haemangioma, and 4 represent hepatic cyst, and 5 represent it
His lesion type.To all 3000 pathological changes block image zooming-out characteristic vectors, trained and test data set.
2. select most effective feature
For selecting most effective feature, it is necessary first to set up basic Random Forest model, take 3000 articles of spies in the 1st step
Levying vector data, as model training data, the building process of random forest is:
1) definition random forest CART decision tree quantity to be set up is 40, repeats 2), 3) step 40 time, generate 40
CART Decision-Tree Classifier Model；
2) by there being the sampling approach put back to concentrate one group of sample of extraction, sample size and original training set from training data
Equally；
3) from 26 sample attributes, randomly choose 10 attributes, be 2) in extraction sub-tree training dataset with choosing
10 attributes selected set up categorised decision tree.
CART decision tree uses Gini index as split criterion, it is assumed that node data collection T has K class, and sample point belongs to
The probability of kth class is pk, for the data set that node T is corresponding, Gini Index for Calculation is as follows:
For CART decision tree, if training set T is unsatisfactory for, " T broadly falls into the most surplus next sample in same category or T
This ", then this node is nonleaf node, so attempting each attribute according to sample and possible property value, enters sample
Row binary divide, it is assumed that after classification, T is divided into A and B, during wherein A accounts for T the ratio of sample be p, B be q (obvious p+q=1).The most miscellaneous
Matter knots modification: Gini (T)-p*Gini (A)-q*Gini (B), the purpose that each property value is attempted divide finds impurity to be somebody's turn to do exactly
The division that variable is maximum, this property value divides subtree and is optimum branching.The data set that node is corresponding is found every time
Disruptive features attribute and characteristic attribute value that impurity knots modification is maximum carry out node split, until the sample in node broadly falls into same
Till one class, obtain decision-tree model by such recurrence split vertexes.CART algorithm is existing algorithm, the most detailed
Explanation.
When setting up the CART decision tree of random forest, can concentrate from initial characteristic data can select at random in the way of putting back to
Selecting and the sample set of initial characteristic data collection equal number, the sample not being extracted is referred to as the outer data of bag, is called for short OOB, and
OOB data can be used to weigh the extensive error of Random Forest model, this extensive error to the test error of Random Forest model
It is referred to as OOB to estimate, is the meansigma methods of every extensive error of decision tree OOB.By the OOB of random forests algorithm, sample can be obtained
The importance degree of eigen attribute, for certain attribute X in sample, importance computational methods are as follows:
1) for each decision tree in random forest, use OOB data that decision-tree model is tested, calculate and survey
Examination error e rrOOB1；
2) randomly the attribute X of samples all in OOB is added interference noise, change sample the most at random at attribute X
Value, then recalculates extensive error e rrOOB2 of OOB；
3) set random forest decision tree quantity as k, take k=40, then the importance computing formula for attribute X is as follows:
By above-mentioned expression formula as the importance measures value of attribute X, if because adding noise at random to attribute X and cause
The extensive error of OOB reduces a lot, illustrates that this attribute X affects the biggest for the nicety of grading of sample.
During building random forest, importance journey is calculated the most respectively for 26 characteristic attributes
Degree, obtains importance degree vectorial:
[importance1, importance2 ..., importance26]
For based on grey level histogram, gray level co-occurrence matrixes, Gray level-gradient co-occurrence matrix characteristics of image in some tolerance
Index is invalid index for CT image for liver local patholoic change identification, therefore can be optimized by feature selection improve the most gloomy
The nicety of grading of woods algorithm, random forests algorithm model can provide the importance degree of each attribute of sample, for hepatopathy
Becoming provincial characteristics data, represent 26 characteristics of image by sequence number 1～26, the feature importance ranking result obtained is shown in accompanying drawing 2 institute
Showing, 26 features according to feature importance degree ranking results are:
f15,f22,f19,f9,f14,f2,f6,f5,f8,f4,f25,f3,f26,f1,f17,f21,f25,f7,f16,
f11,f12,f10,f18,f20,f23,f13
The step of most effective feature selection is as follows:
1) training sample set is set up random forest, and calculate OOB estimation and the importance degree of each characteristic attribute, press
Importance degree carries out characteristic attribute sequence.
2) reject the most unessential feature, this feature attribute of training sample set is rejected, obtains new training sample
This collection, builds random forest with new sample and calculates OOB estimation.
3) 2 are repeated), find out OOB and estimate minimum Random Forest model characteristic of correspondence property set, as most effective
Characteristic combinations of attributes.
According to most effective feature selection approach, oob estimates that the change that average error value changes along with feature selection quantity becomes
Shown in gesture as accompanying drawing 3.From accompanying drawing 3, when feature quantity is 19, oob estimates that average error value is minimum, therefore selects by weight
Front 19 features of 26 features of the property wanted degree sequence combine as most effective feature, and characteristic attribute is as follows:
f15,f22,f19,f9,f14,f2,f6,f5,f8,f4,f25,f3,f26,f1,f17,f21,f25,f7,f16
These 19 features are respectively average based on grey level histogram, variance, skewness, kurtosis, energy, entropy feature, base
In the angle second moment of gray level co-occurrence matrixes, contrast, unfavourable balance moment characteristics, and intensity profile based on Gray level-gradient co-occurrence matrix is not
Uniformity, Gradient distribution inhomogeneities, energy, gray scale are average, gray scale mean square deviation, relevant, gray level entropy, the entropy of mixing, inertia, unfavourable balance
Moment characteristics.
3. the foundation of random forest pathological changes identification model and improvement.
By the most effective characteristic attribute combination obtained in step 2, every the characteristic sieve concentrated from initial characteristic data
Select most effective characteristic, form new characteristic data set, new characteristic data set is set up Random Forest model, and to
Machine forest model carries out decision tree and selects to optimize and identification ability equilibrium, and step is as follows:
1) characteristic is concentrated maximally effective 19 obtained in selecting step 2 in 26 characteristic attributes of every sample
Characteristic attribute, sets up the Random Forest model comprising 400 decision trees, is designated as h1(x), h2(x) ..., h400(x)；
2) to these 400 decision trees, test by test data set, for the normal liver feature in test data set
Data, the forecast error of 400 decision trees is ordered as by little arrivalFor test data set
In liver cancer characteristic data, the forecast error of 400 decision trees is ordered as by little arrivalPin
To the hepatic haemangioma characteristic in test data set, the forecast error of 400 decision trees is ordered as by little arrival For the hepatic cyst characteristic in test data set, the forecast error of 400 decision trees
It is ordered as by little arrival
3) select These decision trees are as final Random Forest model.In these decision trees, repetition may be comprised
Decision tree, the decision tree of this kind of repetition has good recognition accuracy for multiple hepatic lesions type, and such decision tree is
The decision tree should being selected, such system of selection is equivalent to add the weight of such high-class performance decision tree,
Final random forest pathological changes identification model is just provided with more excellent recognition performance.
With the Random Forest model improved, test set is tested, it was predicted that error has obtained to a certain degree reducing, it was predicted that
The variation tendency that error increases with decision tree quantity contrasts as shown in Figure 4, and the asterisk line of side on the upper side is former random forests algorithm
Forecast error with the change curve of decision tree quantity, the cross wires of partial below is algorithm mould after new decision tree selects to optimize
The forecast error of type is with the change curve of decision tree quantity.
Second stage: liver plain CT image pathological changes identification
For a given liver plain CT image, it is necessary first to draw and take the suspected lesion region needing to identify, pass through
The rectangle frame region taking suspected lesion on CT image drawn by picture instrument, as the image in suspected lesion region.
1) the most effective characteristic attribute obtained for the first stage, extracts 19 of CT image for liver suspected lesion region
Validity feature attribute is as characteristic vector；
2) the random forest pathological changes identification model that obtains of first stage is used, to each decision tree to 1) spy that extracts of step
Levy vector and carry out lesion type prediction, the prediction voting results of all decision trees are added up, select type of prediction most
Lesion type, as final pathological changes recognition result.
Concrete pathological changes identification operating process is as shown in Figure 5.
Finally, in addition it is also necessary to be only several specific embodiments of the present invention it is noted that listed above.Obviously, this
Bright it is not limited to above example, it is also possible to have many deformation.Those of ordinary skill in the art can be from present disclosure
The all deformation directly derived or associate, are all considered as protection scope of the present invention.
Claims (2)
1. liver plain CT image pathological changes recognition methods based on random forests algorithm, is characterized in that including herein below:
The gradation of image textural characteristics extracting liver plain CT image lesion region represents as image feature vector, then uses
Random forests algorithm carries out feature selection to CT image for liver lesion region image feature vector, selects maximally effective feature group
Close, then most effective characteristic data set be trained and learn, and the decision tree of random forest is identified ability equilibrium
Optimize, obtain final pathological changes identification model.
Liver plain CT image pathological changes recognition methods based on random forests algorithm the most according to claim 1, its feature
It is: pathological changes identification model is set up, and comprises the steps:
1), CT image for liver lesion region characteristic data set is set up:
Including:
A. grey level histogram feature extraction；
B. gray level co-occurrence matrixes feature extraction；
C. Gray level-gradient co-occurrence matrix feature extraction；
Combine the characteristics of image of above-mentioned three types, and combine the lesion type label of image as characteristic vector data collection D；
2), most effective feature is selected:
Use random forests algorithm to provide the importance degree of each feature in characteristic vector, rejected by loop iteration the heaviest
Want feature, and the residue character after rejecting feature is set up new Random Forest model, find out the model pair that extensive error is minimum
The feature combination answered, is the combination of maximally effective feature；
3), the foundation of random forest pathological changes identification model and improvement
For 2) step obtain most effective characteristic attribute combination, from initial characteristic data concentrate filter out most effective characteristic
Collection；Set up Random Forest model with most effective characteristic data set, and the decision tree in Random Forest model is optimized and all
Weighing apparatus, obtains final random forest pathological changes identification model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610231280.6A CN105931224A (en) | 2016-04-14 | 2016-04-14 | Pathology identification method for routine scan CT image of liver based on random forests |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610231280.6A CN105931224A (en) | 2016-04-14 | 2016-04-14 | Pathology identification method for routine scan CT image of liver based on random forests |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN105931224A true CN105931224A (en) | 2016-09-07 |
Family
ID=56838951
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610231280.6A Pending CN105931224A (en) | 2016-04-14 | 2016-04-14 | Pathology identification method for routine scan CT image of liver based on random forests |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105931224A (en) |
Cited By (30)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106897821A (en) * | 2017-01-24 | 2017-06-27 | 中国电力科学研究院 | A kind of transient state assesses feature selection approach and device |
| CN107480702A (en) * | 2017-07-20 | 2017-12-15 | 东北大学 | Towards the feature selecting and Feature fusion of the identification of HCC pathological images |
| CN107632995A (en) * | 2017-03-13 | 2018-01-26 | 平安科技（深圳）有限公司 | The method and model training control system of Random Forest model training |
| CN107845098A (en) * | 2017-11-14 | 2018-03-27 | 南京理工大学 | Liver cancer image full-automatic partition method based on random forest and fuzzy clustering |
| CN108090507A (en) * | 2017-10-19 | 2018-05-29 | 电子科技大学 | A kind of medical imaging textural characteristics processing method based on integrated approach |
| WO2018098697A1 (en) * | 2016-11-30 | 2018-06-07 | 中国科学院深圳先进技术研究院 | Image feature repeatability measurement method and device |
| CN108320026A (en) * | 2017-05-16 | 2018-07-24 | 腾讯科技（深圳）有限公司 | Machine learning model training method and device |
| CN108335282A (en) * | 2017-01-20 | 2018-07-27 | 浙江京新术派医疗科技有限公司 | The cut surface generation method and generating means of operation on liver |
| CN108764355A (en) * | 2018-05-31 | 2018-11-06 | 清华大学 | Image processing apparatus and method based on textural characteristics classification |
| CN108805858A (en) * | 2018-04-10 | 2018-11-13 | 燕山大学 | Hepatopathy CT image computers assistant diagnosis system based on data mining and method |
| CN108961207A (en) * | 2018-05-02 | 2018-12-07 | 上海大学 | Lymph node Malignant and benign lesions aided diagnosis method based on multi-modal ultrasound image |
| CN109003659A (en) * | 2017-06-07 | 2018-12-14 | 万香波 | Stomach Helicobacter pylori infects pathological diagnosis and supports system and method |
| CN109117890A (en) * | 2018-08-24 | 2019-01-01 | 腾讯科技（深圳）有限公司 | A kind of image classification method, device and storage medium |
| CN109344907A (en) * | 2018-10-30 | 2019-02-15 | 顾海艳 | Based on the method for discrimination for improving judgment criteria sorting algorithm |
| CN109493886A (en) * | 2018-12-13 | 2019-03-19 | 西安电子科技大学 | Speech-emotion recognition method based on feature selecting and optimization |
| CN109840535A (en) * | 2017-11-29 | 2019-06-04 | 北京京东尚科信息技术有限公司 | The method and apparatus for realizing classification of landform |
| CN109858562A (en) * | 2019-02-21 | 2019-06-07 | 腾讯科技（深圳）有限公司 | A kind of classification method of medical image, device and storage medium |
| CN109934179A (en) * | 2019-03-18 | 2019-06-25 | 中南大学 | Human motion recognition method based on automated characterization selection and Ensemble Learning Algorithms |
| CN109965829A (en) * | 2019-03-06 | 2019-07-05 | 重庆金山医疗器械有限公司 | Imaging optimization method, image processing apparatus, imaging device and endoscopic system |
| CN110378875A (en) * | 2019-06-18 | 2019-10-25 | 中国科学院苏州生物医学工程技术研究所 | Internal lithangiuria ingredient discrimination method based on machine learning algorithm |
| CN110619633A (en) * | 2019-09-10 | 2019-12-27 | 武汉科技大学 | Liver image segmentation method based on multi-path filtering strategy |
| CN111291896A (en) * | 2020-02-03 | 2020-06-16 | 深圳前海微众银行股份有限公司 | Interactive random forest subtree screening method, device, equipment and readable medium |
| CN111445946A (en) * | 2020-03-26 | 2020-07-24 | 北京易康医疗科技有限公司 | Calculation method for calculating lung cancer genotyping by using PET/CT (positron emission tomography/computed tomography) images |
| WO2020199692A1 (en) * | 2019-04-04 | 2020-10-08 | 中国科学院深圳先进技术研究院 | Method and apparatus for screening predictive image features for cancer metastasis, and storage medium |
| WO2020233259A1 (en) * | 2019-07-12 | 2020-11-26 | 之江实验室 | Multi-center mode random forest algorithm-based feature importance sorting system |
| CN112633317A (en) * | 2020-11-02 | 2021-04-09 | 国能信控互联技术有限公司 | CNN-LSTM fan fault prediction method and system based on attention mechanism |
| CN112883962A (en) * | 2021-01-29 | 2021-06-01 | 北京百度网讯科技有限公司 | Fundus image recognition method, device, apparatus, storage medium, and program product |
| CN113593707A (en) * | 2021-09-29 | 2021-11-02 | 武汉楚精灵医疗科技有限公司 | Stomach early cancer model training method and device, computer equipment and storage medium |
| WO2021259003A1 (en) * | 2020-06-23 | 2021-12-30 | 平安科技（深圳）有限公司 | Feature recognition method and apparatus, and computer device and storage medium |
| CN116309593A (en) * | 2023-05-23 | 2023-06-23 | 天津市中西医结合医院（天津市南开医院） | Liver puncture biopsy B ultrasonic image processing method and system based on mathematical model |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20100158332A1 (en) * | 2008-12-22 | 2010-06-24 | Dan Rico | Method and system of automated detection of lesions in medical images |
| CN104751160A (en) * | 2015-03-12 | 2015-07-01 | 西安电子科技大学 | Mammary gland image processing method based on sparse automatic coding depth network |
| CN104866862A (en) * | 2015-04-27 | 2015-08-26 | 中南大学 | Strip steel surface area type defect identification and classification method |
| CN105427325A (en) * | 2015-12-07 | 2016-03-23 | 苏州大学 | Automatic lung tumour segmentation method based on random forest and monotonically decreasing function |
-
2016
- 2016-04-14 CN CN201610231280.6A patent/CN105931224A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20100158332A1 (en) * | 2008-12-22 | 2010-06-24 | Dan Rico | Method and system of automated detection of lesions in medical images |
| CN104751160A (en) * | 2015-03-12 | 2015-07-01 | 西安电子科技大学 | Mammary gland image processing method based on sparse automatic coding depth network |
| CN104866862A (en) * | 2015-04-27 | 2015-08-26 | 中南大学 | Strip steel surface area type defect identification and classification method |
| CN105427325A (en) * | 2015-12-07 | 2016-03-23 | 苏州大学 | Automatic lung tumour segmentation method based on random forest and monotonically decreasing function |
Non-Patent Citations (3)
| Title |
|---|
| 于殿泓: "《图像检测与处理技术》", 31 December 2006, 西安电子科技大学出版社 * |
| 姜慧 等: "基于双树复数小波的肝脏疾病分类", 《电脑与信息技术》 * |
| 胡峻峰: "基于机器视觉的实木地板分选技术研究", 《中国博士学位论文全文数据库》 * |
Cited By (40)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2018098697A1 (en) * | 2016-11-30 | 2018-06-07 | 中国科学院深圳先进技术研究院 | Image feature repeatability measurement method and device |
| CN108335282A (en) * | 2017-01-20 | 2018-07-27 | 浙江京新术派医疗科技有限公司 | The cut surface generation method and generating means of operation on liver |
| CN106897821B (en) * | 2017-01-24 | 2023-07-21 | 中国电力科学研究院 | Transient evaluation feature selection method and device |
| CN106897821A (en) * | 2017-01-24 | 2017-06-27 | 中国电力科学研究院 | A kind of transient state assesses feature selection approach and device |
| CN107632995B (en) * | 2017-03-13 | 2018-09-11 | 平安科技（深圳）有限公司 | The method and model training control system of Random Forest model training |
| CN107632995A (en) * | 2017-03-13 | 2018-01-26 | 平安科技（深圳）有限公司 | The method and model training control system of Random Forest model training |
| CN108320026B (en) * | 2017-05-16 | 2022-02-11 | 腾讯科技（深圳）有限公司 | Machine learning model training method and device |
| CN108320026A (en) * | 2017-05-16 | 2018-07-24 | 腾讯科技（深圳）有限公司 | Machine learning model training method and device |
| CN109003659A (en) * | 2017-06-07 | 2018-12-14 | 万香波 | Stomach Helicobacter pylori infects pathological diagnosis and supports system and method |
| CN107480702B (en) * | 2017-07-20 | 2021-03-30 | 东北大学 | Feature selection and feature fusion method for HCC pathological image recognition |
| CN107480702A (en) * | 2017-07-20 | 2017-12-15 | 东北大学 | Towards the feature selecting and Feature fusion of the identification of HCC pathological images |
| CN108090507A (en) * | 2017-10-19 | 2018-05-29 | 电子科技大学 | A kind of medical imaging textural characteristics processing method based on integrated approach |
| CN107845098A (en) * | 2017-11-14 | 2018-03-27 | 南京理工大学 | Liver cancer image full-automatic partition method based on random forest and fuzzy clustering |
| CN109840535A (en) * | 2017-11-29 | 2019-06-04 | 北京京东尚科信息技术有限公司 | The method and apparatus for realizing classification of landform |
| CN108805858A (en) * | 2018-04-10 | 2018-11-13 | 燕山大学 | Hepatopathy CT image computers assistant diagnosis system based on data mining and method |
| CN108961207A (en) * | 2018-05-02 | 2018-12-07 | 上海大学 | Lymph node Malignant and benign lesions aided diagnosis method based on multi-modal ultrasound image |
| CN108961207B (en) * | 2018-05-02 | 2022-11-04 | 上海大学 | Auxiliary diagnosis method for benign and malignant lymph node lesion based on multi-modal ultrasound images |
| CN108764355A (en) * | 2018-05-31 | 2018-11-06 | 清华大学 | Image processing apparatus and method based on textural characteristics classification |
| CN109117890A (en) * | 2018-08-24 | 2019-01-01 | 腾讯科技（深圳）有限公司 | A kind of image classification method, device and storage medium |
| CN109344907A (en) * | 2018-10-30 | 2019-02-15 | 顾海艳 | Based on the method for discrimination for improving judgment criteria sorting algorithm |
| CN109493886A (en) * | 2018-12-13 | 2019-03-19 | 西安电子科技大学 | Speech-emotion recognition method based on feature selecting and optimization |
| CN109858562A (en) * | 2019-02-21 | 2019-06-07 | 腾讯科技（深圳）有限公司 | A kind of classification method of medical image, device and storage medium |
| CN109965829A (en) * | 2019-03-06 | 2019-07-05 | 重庆金山医疗器械有限公司 | Imaging optimization method, image processing apparatus, imaging device and endoscopic system |
| CN109965829B (en) * | 2019-03-06 | 2022-05-06 | 重庆金山医疗技术研究院有限公司 | Imaging optimization method, image processing apparatus, imaging apparatus, and endoscope system |
| CN109934179A (en) * | 2019-03-18 | 2019-06-25 | 中南大学 | Human motion recognition method based on automated characterization selection and Ensemble Learning Algorithms |
| WO2020199692A1 (en) * | 2019-04-04 | 2020-10-08 | 中国科学院深圳先进技术研究院 | Method and apparatus for screening predictive image features for cancer metastasis, and storage medium |
| CN110378875A (en) * | 2019-06-18 | 2019-10-25 | 中国科学院苏州生物医学工程技术研究所 | Internal lithangiuria ingredient discrimination method based on machine learning algorithm |
| WO2020233259A1 (en) * | 2019-07-12 | 2020-11-26 | 之江实验室 | Multi-center mode random forest algorithm-based feature importance sorting system |
| CN110619633A (en) * | 2019-09-10 | 2019-12-27 | 武汉科技大学 | Liver image segmentation method based on multi-path filtering strategy |
| CN111291896A (en) * | 2020-02-03 | 2020-06-16 | 深圳前海微众银行股份有限公司 | Interactive random forest subtree screening method, device, equipment and readable medium |
| CN111445946B (en) * | 2020-03-26 | 2021-07-30 | 山东省肿瘤防治研究院（山东省肿瘤医院） | Calculation method for calculating lung cancer genotyping by using PET/CT (positron emission tomography/computed tomography) images |
| CN111445946A (en) * | 2020-03-26 | 2020-07-24 | 北京易康医疗科技有限公司 | Calculation method for calculating lung cancer genotyping by using PET/CT (positron emission tomography/computed tomography) images |
| WO2021259003A1 (en) * | 2020-06-23 | 2021-12-30 | 平安科技（深圳）有限公司 | Feature recognition method and apparatus, and computer device and storage medium |
| CN112633317A (en) * | 2020-11-02 | 2021-04-09 | 国能信控互联技术有限公司 | CNN-LSTM fan fault prediction method and system based on attention mechanism |
| CN112883962B (en) * | 2021-01-29 | 2023-07-18 | 北京百度网讯科技有限公司 | Fundus image recognition method, fundus image recognition apparatus, fundus image recognition device, fundus image recognition program, and fundus image recognition program |
| CN112883962A (en) * | 2021-01-29 | 2021-06-01 | 北京百度网讯科技有限公司 | Fundus image recognition method, device, apparatus, storage medium, and program product |
| CN113593707A (en) * | 2021-09-29 | 2021-11-02 | 武汉楚精灵医疗科技有限公司 | Stomach early cancer model training method and device, computer equipment and storage medium |
| CN113593707B (en) * | 2021-09-29 | 2021-12-14 | 武汉楚精灵医疗科技有限公司 | Stomach early cancer model training method and device, computer equipment and storage medium |
| CN116309593A (en) * | 2023-05-23 | 2023-06-23 | 天津市中西医结合医院（天津市南开医院） | Liver puncture biopsy B ultrasonic image processing method and system based on mathematical model |
| CN116309593B (en) * | 2023-05-23 | 2023-09-12 | 天津市中西医结合医院（天津市南开医院） | Liver puncture biopsy B ultrasonic image processing method and system based on mathematical model |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105931224A (en) | Pathology identification method for routine scan CT image of liver based on random forests | |
| CN109493308B (en) | Medical image synthesis and classification method for generating confrontation network based on condition multi-discrimination | |
| CN109145921A (en) | A kind of image partition method based on improved intuitionistic fuzzy C mean cluster | |
| CN110853011B (en) | Method for constructing convolutional neural network model for pulmonary nodule detection | |
| CN106529448A (en) | Method for performing multi-visual-angle face detection by means of integral channel features | |
| JP2008077677A (en) | Measurement of mitotic activity | |
| CN105913086A (en) | Computer-aided mammary gland diagnosing method by means of characteristic weight adaptive selection | |
| CN104282008B (en) | The method and apparatus that Texture Segmentation is carried out to image | |
| CN101081168A (en) | Method for shielding sex part on foetus image for preventing recognizing foetus sex | |
| CN109978880A (en) | Lung tumors CT image is carried out sentencing method for distinguishing using high dimensional feature selection | |
| CN106326834A (en) | Human body gender automatic identification method and apparatus | |
| CN106127735A (en) | A kind of facilities vegetable edge clear class blade face scab dividing method and device | |
| Beevi et al. | Detection of mitotic nuclei in breast histopathology images using localized ACM and Random Kitchen Sink based classifier | |
| CN109671060A (en) | Area of computer aided breast lump detection method based on selective search and CNN | |
| CN113191359B (en) | Small sample target detection method and system based on support and query samples | |
| Giusti et al. | A comparison of algorithms and humans for mitosis detection | |
| CN103279960B (en) | A kind of image partition method of human body cache based on X-ray backscatter images | |
| Hassanien et al. | Detection of spiculated masses in Mammograms based on fuzzy image processing | |
| CN110348320A (en) | A kind of face method for anti-counterfeit based on the fusion of more Damage degrees | |
| WO2022141201A1 (en) | Breast cancer grading method based on dce-mri | |
| Nasir et al. | Detection of acute leukaemia cells using variety of features and neural networks | |
| Cao et al. | 3D convolutional neural networks fusion model for lung nodule detection onclinical CT scans | |
| Fan et al. | Automated blood vessel segmentation in fundus image based on integral channel features and random forests | |
| CN107545565A (en) | A kind of solar energy half tone detection method | |
| CN103902997B (en) | Feature subspace integration method for biological cell microscope image classification |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into 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: 20160907 |