CN108805858A - Hepatopathy CT image computers assistant diagnosis system based on data mining and method - Google Patents
Hepatopathy CT image computers assistant diagnosis system based on data mining and method Download PDFInfo
- Publication number
- CN108805858A CN108805858A CN201810313764.4A CN201810313764A CN108805858A CN 108805858 A CN108805858 A CN 108805858A CN 201810313764 A CN201810313764 A CN 201810313764A CN 108805858 A CN108805858 A CN 108805858A
- Authority
- CN
- China
- Prior art keywords
- image
- liver
- textural characteristics
- diagnosis
- module
- 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
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/12—Edge-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/40—Analysis of texture
- G06T7/41—Analysis of texture based on statistical description of texture
- G06T7/45—Analysis of texture based on statistical description of texture using co-occurrence matrix computation
-
- 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
- G16H30/20—ICT specially adapted for the handling or processing of medical images for handling medical images, e.g. DICOM, HL7 or PACS
-
- 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
- G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
- G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems
-
- 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
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30061—Lung
-
- 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
- G06T2207/30096—Tumor; Lesion
Abstract
The invention discloses a kind of hepatopathy CT image computers assistant diagnosis system and method based on data mining.The system comprises input module, texture feature extraction module, classification diagnosis identification module and output modules.The method uses the system, and content includes：Contours segmentation and extraction are carried out to abdominal CT images in advance, and marked normal liver CT, hepatic cyst, liver cancer, pretreated image is imported into the system；Image texture characteristic extraction module carries out analyzing image texture to importing image, obtains 13 dimension gray scale symbiosis textural characteristics in CT images；The database that liver cancer, hepatic cyst image texture characteristic and the normal hepatocytes image liver textural characteristics that storage has confirmed in classification diagnosis module are constituted, according to Database diagnostic model, the CT image for liver textural characteristics of input are substituted into Classification of Diagnosis Models device and are handled, diagnostic result and accuracy are obtained.
Description
Technical field
The present invention relates to medical image analysis technical field more particularly to a kind of hepatopathy CT image meters based on data mining
Calculation machine assistant diagnosis system and method.
Background technology
Cancer (refers to malignant tumour), seriously threatens people's health, and liver cancer belongs to a kind of common malignant tumour.In
State, liver cancer all occupy second in terms of incidence is with the death rate, and in city, lung cancer occupy first, is only second in rural area
Gastric cancer.Due to liver cancer early stage non-evident sympton, when discovery, has just been in middle and advanced stage, and liver diagnosis is applied at present
Liver tissue bioptic puncture technique can cause huge injury and pain to the psychology and physiology of patient.Therefore appliance computer is examined
Disconnected technology interventional treatment is so as to Accurate Diagnosis and makes rational therapeutic scheme tool and has very great significance.
In recent years, with the continuous development of science and technology, digitlization has come into medical domain, and current era can be described as greatly
The epoch of data, however mass data generates and the workloads such as doctor is excessively increased, and easy tos produce various mistakes, such as mistaken diagnosis
With fail to pinpoint a disease in diagnosis etc, the consequence that this mistake generates in terms of medicine is hardly imaginable.Therefore, we are behaved using computer technology
Class service mitigates people's work load, and computer technology is a great development of modern medicine, is produced for medical diagnosis deep
It influences.
We carry out information decision using database, and data mining is currently one of research direction of this part, so-called
Data mining, exactly finds out that oneself is interested and rationally applied to it from mass data, obtains oneself desired knot
Fruit.In terms of medical domain, we utilize the core technology of data mining, clustering to utilize cluster under the support of big data
Structural classification device is analyzed by lesion classification, reduces misdiagnosis rate.
Invention content
The present invention mainly solves to diagnose liver disease by physical examination, ultrasound, radioisotope scan at present and lack there are many
Sunken problem provides a kind of hepatopathy CT image computers assistant diagnosis system and method based on data mining, this is a kind of liver
The high computer-aided diagnosis system of portion's pathological changes diagnosis accuracy and method.
The above-mentioned technical problem of the present invention is mainly to be addressed by following technical proposals：
A kind of hepatopathy CT image computer assistant diagnosis systems based on data mining, include sequentially connected input mould
Block, texture feature extraction module, classification diagnosis identification module and output module；
The input module is for importing the liver's CT images for passing through contours segmentation and extraction；
Described image texture feature extraction module is to carry out a series of images texture point for liver's CT images to extraction
Analysis, obtains the characteristics of image of target image, described image is characterized as 9 dimension gray scale symbiosis textural characteristics；
The classification diagnosis module, include in the module grader and have liver cancer, the hepatic cyst of many cases definitive pathological diagnosis with
And the database of normal liver CT image texture characteristics, the classification diagnosis module is according to the trained diagnosis mould in data with existing library
Type, and classification for statistical analysis to the image texture characteristic of input judge；
The output module is to show and generate report for exporting statistic of classification result, and statistical result includes that classification is sentenced
Disconnected result and accuracy statistical indicator.
A kind of hepatopathy CT image computer aided diagnosis methods based on data mining, this method are based on data using described
The hepatopathy CT image computer assistant diagnosis systems of excavation, content mainly include the following steps that：
(1) contours segmentation and extraction are carried out to abdominal CT images in advance, and have marked normal liver CT, hepatic cyst, liver cancer,
Pretreated image is imported into the system；
(2) image texture characteristic extraction module carries out analyzing image texture to importing image, obtains 13 dimensions in CT images
Gray scale symbiosis textural characteristics；
(3) liver cancer, hepatic cyst image texture characteristic and the normal hepatocytes image that storage has confirmed in classification diagnosis module
The database that liver textural characteristics are constituted, according to Database diagnostic model, by the CT image for liver textural characteristics generation of input
Enter and handled in Classification of Diagnosis Models device, obtains diagnostic result and accuracy.
The present invention carries out texture statistics analysis by computer to CT image for liver, and the automatically derived CT image for liver is examined
Disconnected result and several statistics indexs have non-invasive, quick timely feature, and do not need chemical reagent, at low cost；And
By the way that training sample characteristic optimization, the complexity of lesion thyroid gland textural characteristics sample set can be effectively reduced, into one
Step, which improves, differentiates accuracy；In addition the method for the present invention result avoids pathologic finding box due to not having artificial subjective factor influence
Other artificial subjective factors checked influence.
As a preferred embodiment, described that contours segmentation and extraction, mistake are carried out to abdominal CT images in step (1)
Cheng Wei：
Image is cut first to obtain ROI, then interested region is filled and is removed the place of noise
Reason, noise is removed using median filtering method, and image enhancement is carried out using histogram equalization method；And 256 are carried out to CT images
Grade gradation conversion, is then stored as double types, finally carries out 16 grades of gray compressions, and the ROI image of all extractions is constituted
CT image for liver collection.
As a preferred embodiment, in step (2), the 13 dimension gray scale symbiosis textural characteristics obtained in CT images,
Obtaining the process that wherein 9 tie up gray scale symbiosis textural characteristics is：
Appoint in the picture and take a pixel A (x, y), then obtains the one other pixel point B (x+ for being d with pixel A distances
A, y+b), pixel A and pixel B form a point pair, record the gray value (i, j) of this point pair, and the value of fixed a and b makes
Pixel A (x, y) is moved on the image, obtain a variety of pixels to combination, Ng be gradation of image rank value, i ∈ [0, Ng-1],
J ∈ [0, Ng-1], change d and θ, and θ is pixel to line and horizontal angle, the gray scale of the pixel pair in statistical picture
Value constitutes gray level co-occurrence matrixes P (i, j, θ, d), and wherein # { x } is the number of all elements in set x；Then according to gray scale
9 dimension textural characteristics of co-occurrence matrix extraction, respectively contrast, inverse difference moment, correlation, entropy, angular second moment, symbiosis and mean value are total to
Entropy, symbiosis and the difference of raw sum and the entropy of symbiosis difference.
As a preferred embodiment, described according to Database diagnostic model in step (3), establish diagnostic model
Process be：
N textural characteristics data sample is chosen out of database as training sample set, with training sample set pair grader
Be trained, wherein n be positive integer, and for entire database sample set 1/2, and by database in addition to training sample set
N textural characteristics data sample classifies to verification sample set as verification sample set, using grader after training, is tested
The discriminating accuracy rate of sample set is demonstrate,proved, n sample is then still further chosen and repeats the above steps, multi-pass operation is finally chosen and obtained
Maximum diagnosis accuracy rate dimension target establishes diagnostic model.
It is as a preferred embodiment, described n textural characteristics data sample to be chosen out of database as training sample set,
It is exactly the selection liver textural characteristics data sample out of database, is divided into training sample set and verification sample set, uses Relief
The feature of extraction is carried out dimension-reduction treatment by algorithm, is built grader using improved K-means clustering methods, is used instruction
Practice sample set to be trained grader.
Due to the adoption of the above technical scheme, the present invention has such advantageous effect compared with prior art：
1. the CT images used are obtained in biopsy or operation consent, therefore with non-invasive, have the characteristics that quick timely；
2. obtaining liver cancer by CT images to differentiate, living tissue identification etc. need not be taken, there is at low cost, reduction patient's pain
Bitter feature；
3. optimizing to the Feature Dimension Reduction of training sample by Relief algorithms, it is special that pathologic liver texture can be effectively reduced
The complexity for levying sample set, further increases discriminating accuracy；
4. differentiating the good pernicious of lump by the grader after the training of the textural characteristics data set of CT image for liver, tie
Fruit does not have artificial subjective factor influence, avoids the artificial subjective factor of pathologic finding and other inspections and influences；
5. read using three layer of two sub-argument, the problem of by four points of difficulty, the problem of being converted into three two points of difficulty, therefore
When being classified here, the k=2 of k-means algorithms is taken, repeatedly splits data into two classes；
6. using the Relief algorithms of optimization, will be weighted as the textural characteristics of classification foundation, or removal does not work
Characteristic, reduce operand and time, increase the precision of system model.
Description of the drawings
In order to illustrate the technical solution of the embodiments of the present invention more clearly, required use in being described below to embodiment
Attached drawing be briefly described, it should be apparent that, drawings in the following description are only some embodiments of the invention, for this
For the those of ordinary skill in field, without creative efforts, other are can also be obtained according to these attached drawings
Attached drawing.
Fig. 1 is the hepatopathy CT image computer assistant diagnosis system structural schematic diagrams based on data mining of the present invention；
Fig. 2 is liver CT image procossing block diagrams in the method for the present invention；
Fig. 3 is clustering block diagram in the method for the present invention；
Fig. 4 is analytic process figure in clustering method of the present invention；
Fig. 5 is the hepatopathy CT image computer aided diagnosis method overview flow charts based on data mining of the present invention.
Specific implementation mode
Below with reference to the embodiments and with reference to the accompanying drawing the technical solutions of the present invention will be further described.
Embodiment：
This example is a kind of hepatopathy CT image computer assistant diagnosis systems based on data mining, as shown in Figure 1, including
There are sequentially connected input module, texture feature extraction module, classification diagnosis identification module and output module.
The input module is for importing the liver's CT images for passing through contours segmentation and extraction；
Described image texture feature extraction module is to carry out a series of images texture point for liver's CT images to extraction
Analysis, obtains the characteristics of image of target image, described image is characterized as 9 dimension gray scale symbiosis textural characteristics；
The classification diagnosis module, include in the module grader and have liver cancer, the hepatic cyst of many cases definitive pathological diagnosis with
And the database of normal liver CT image texture characteristics, the classification diagnosis module is according to the trained diagnosis mould in data with existing library
Type, and classification for statistical analysis to the image texture characteristic of input judge；
The output module is to show and generate report for exporting statistic of classification result, and statistical result includes that classification is sentenced
Disconnected result and accuracy statistical indicator.
And a kind of hepatopathy CT image computer aided diagnosis methods based on data mining, as shown in figure 4, this method is adopted
The hepatopathy computer-aided diagnosis system based on data mining described in claim 1；The method content includes mainly following
Step：
Step 1：Contours segmentation and extraction are carried out to abdominal CT images in advance, and marked normal liver CT, hepatic cyst,
Pretreated image is imported into the system by liver cancer；As shown in Figure 2.
Abdominal CT images are acquired to patient by using CT scan instrument, CT scanner uses Siemens Sensation16
Slices spiral CT acquires the unenhanced cross-sectional image of abdominal CT, picture format DICOM.CT equipment sweep parameter is bulb voltage
120kV, tube current 220mAs, thickness 2-5mm, interlamellar spacing 2-5mm, screw pitch 1-1.5, image reconstruction type is B40, soft
Display window is organized, the resolution ratio of cross-sectional image is 512 × 512 pixels, 10-15 cross-sectional image of every patient.The present invention
Method will use normal liver tissue, liver cancer patient, and the data of hepatic cyst patient carries out analysis and utilization, then in CT scan image
Carry out profile separation and extraction.The process that abdominal CT images are carried out with contours segmentation and extraction is：Liver will be contained in the picture
Edge sketches out, specifically usable microMRI softwares.Then it is come out in the image zooming-out containing thyroid gland that will be sketched out, then
Interested region is filled and is removed the processing of noise, noise is removed using median filtering method, it is equal using histogram
Weighing apparatusization method carries out image enhancement.And 256 grades of gradation conversions are carried out to CT images, and double types are then stored as, it is most laggard
16 grades of gray compressions of row, specifically usable matlab programmings are realized.The liver ROI image of all extractions is constituted into Hepatic CT figure
Image set.While obtaining image set, by the corresponding normal liver of image, hepatic cyst, liver cancer.
Step 2：Image texture characteristic extraction module carries out analyzing image texture to the ROI image of extraction, obtains ROI figures
The 13 dimension gray scale symbiosis textural characteristics as in.
A pixel A (x, y) is taken as shown in figure 4, appointing in the picture, then acquisition and pixel A distances are another of d
Pixel B (x+a, y+b), pixel A and B form a point pair, record the gray value (i, j) of this point pair, fixed a's and b
Value, so that pixel A (x, y) is moved on the image, obtain a variety of pixels to combination, Ng be gradation of image rank value, i ∈ [0,
Ng-1], j ∈ [0, Ng-1] change d and θ, and θ is angle of the pixel to line and horizontal elder generation, the pixel pair in statistical picture
Gray value, constitute gray level co-occurrence matrixes P (i, j, θ, d), wherein # { x } is the number of all elements in set x；It sets in advance
Determine d values, calculates separately the gray level co-occurrence matrixes of 0 °, 45 °, 90 ° and 135 ° four direction, if tonal range is [0, Ng-1],
Then the size of the gray level co-occurrence matrixes in each direction is Ng × Ng.In the present embodiment, preset d=1, calculate separately 0 °,
45 °, 90 ° and 135 ° of gray level co-occurrence matrixes, the matrix for being then based on each direction calculates textural characteristics, by the line of four direction
Feature averaged is managed, the textural characteristics of invariable rotary are obtained.According to 9 dimension textural characteristics of gray level co-occurrence matrixes extraction.They
Respectively：
(i) contrast
Contrast can be understood as the clarity of texture image.In the picture, the rill of texture is deeper, and contrast is bigger,
The visual effect of the image is more clear.
(ii) inverse difference moment
Inverse difference moment reflects the regular degree and readability of texture, and texture is clear, regular, inverse difference moment
Value it is bigger, otherwise inverse difference moment is smaller.
(iii) correlation
Correlation refers to similarity degree of the textural characteristics element on line direction and column direction.Horizontal direction is occupied leading
Status, that is to say, that gray level co-occurrence matrixes correlation of the image at 0 ° is more than image at 45 °, 90 °, 135 ° of gray scale symbiosis square
Battle array correlation.
(iv) entropy
What entropy indicated is image complexity, is exactly the complexity for referring to texture image here, and texture is fewer in image,
Entropy is with regard to smaller, if being full of texture, entropy is bigger.
(v) angular second moment
Angular second moment f5 is used for measuring the uniformity coefficient of gradation of image.As can be seen that f5 refers to gray scale symbiosis from formula
The quadratic sum of matrix pixel values, therefore, also referred to as energy.When texture is relatively thick, the value of second moment is bigger, conversely, then
It is worth smaller.
(vi) symbiosis and mean value
(vii) entropy of symbiosis sum
(viii) symbiosis difference mean value
(ix) entropy of symbiosis difference
The textural characteristics of extraction CT images are programmed using Matlab, and the data extracted are automatically write into excel.Using
Data are normalized in min-max standardization.
Step 3：Storage first carries out Dimension Reduction Analysis by the ROI image textural characteristics extracted in classification diagnosis module, then structure
At database.
1) Relife algorithms are used to carry out Feature Dimension Reduction
The specific core concept of Relief algorithms helps us to being conducive to divide class another characteristic increase weight appropriate,
Remove feature that is useless or playing adverse effect.Figuratively, relief is when calculating, first in training sample set
Select a sample R, acquire later a similar nearest samples H (be referred to as NearHit) and it is non-similar one it is closest
Sample M (is referred to as Near Miss), then calculates the weight of feature by comparing the characteristic value of nearest sample.Algorithm advantage is meter
It calculates simply, and considers the correlation of attribute.In practical applications, that in most cases we need to distinguish is a variety of spies
Sign.Kononenko extended Relief algorithms in order to handle the ReliefF algorithms of multi-class problem in 1994.The algorithm
A sample x is randomly choosed from training seti, then from xiNearest samples PH is found in similar samplei, from xiNo
Adjacent sample PM is found in similar samplei, finally according to the weight of each feature of following Policy Updates, i.e.,
In formula：xiJ indicates sample xiValue about j-th of feature；D () indicates distance function, for calculating two samples
Distance about some feature；M is the number of randomly drawing sample.
When characterized by discrete features, Euclidean distance of two samples on feature j is：
When characterized by continuous feature, Euclidean distance of two samples on feature j is：
Wherein maxj, min (j) indicate the maximum value and minimum value in j-th of feature institute value respectively.In formula, Ri, RjIt is
Any two sample in sample set, value (I, R) are the values of the i-th feature of R.
By formula (1) if it is found that distance diff (I, R from of a sort two samples on feature Ii,Hj) smaller, or
Distance diff (I, R from inhomogeneous two samples on feature Ii,Mj(C)) bigger, then show the classification capacity of feature I
Stronger, the weights W (I) obtained is also bigger.The process needs iteration m times, finally obtain accumulation after feature weight to
Measure W (I).
2) textural characteristics select
For the entropy mean value of extraction, average energy value, contrast mean value, correlation mean value, local stationary mean value, symbiosis and
9 textural characteristics of entropy of mean value, the entropy of symbiosis sum, symbiosis and difference and symbiosis difference using ReliefF algorithms calculate special
Weight is levied, weight size is compared analysis, the feature of threshold value will be less than if there is weight to be removed.In the ring of Matlab
Under border, program is run 20 times.
Step 4：Classification diagnosis module is according to Database classification diagnosis model, by liver's CT image textures of input
Feature is substituted into diagnostic model and is handled, and obtains diagnostic result and accuracy.The grader of classification diagnosis module, which utilizes, to be changed
Into K-means algorithms built.
It introduces
Clustering algorithm is to worship the thought that things of a kind come together, people of a mind fall into the same group, which is different from sorting algorithm, and sorting algorithm is by a number
It is classified as the one type in the classification divided according to according to its characteristic, and it refers to utilizing algorithm by a heap data to cluster
It is divided into several classifications according to their own characteristic.Early in 1975, Hartugan just carried out detailed opinion to clustering algorithm
It states, k-means algorithms are obtained because of its efficient advantage easy to carry out in data mining, machine learning, statistics etc. field
It is widely applied, is most common clustering algorithm.Clustering algorithm is divided into four parts：The extraction and selection of feature, similitude
Grouping, obtains cluster result at classification calculating, as shown in Figure 3.
Since clustering algorithm is divided according to the similarity of feature, then describing this of difference between internal feature
Degree is exactly a critically important measurement index, general to describe the difference between data using the method for defining the distance between data
It is different.
Most common is exactly Euclidean distance, is defined as follows：
Dxp, xq=| xp1-xq1 | 2+ | xp2-xq2 | 2+ ...+| xpd-xqd | 2
Euclidean distance mathematically needs to meet some requirements：
d(xp,xq)≥0:The distance between two objects are necessary for nonnegative number；,
D (xp, xp)=0:The distance between object and itself are 0；
Dxp, xq=dxq, xp:The symmetry of distance；
dxp,xq≤dxp,xi+dxi,xq；Meet triangle relation formula.
Other than several inner range formulas in Europe also have it is several be commonly used for measurement scalar distinctiveness ratio range formulas, Manhattan away from
From and Minkowski Distance, be defined as follows：
Manhatton distance：
Dxp, xq=| xp1-xq1 |+| xp2-xq2 |+...+| xpd-xqd |
Minkowski Distance：
Dxp, xq=n | xp1-xq1 | n+ | xp2-xq2 | n+ ...+| xpd-xqd | n
K-means algorithms are to be based on Euclidean distance, and the core formula of the algorithm is as follows：
It can be seen that the essence of k-means formula is exactly to calculate in each sample point to the sample class of selection from this formula
Then the distance of heart point judges that sample point is closer apart from the distance of which class central point, just it assigned in this classification,
Then, recalculate class central point according to the classification divided, again iterate to calculate all the points to new class central point distance again
Classification, continuous iteration continues, and final class center is no more than some specified value apart from displacement distance, reaches a convergence effect
Fruit, that is, Clustering Effect.Image a bit, as shown in Figure 3.
It improves
The advantages of k-means algorithms is efficient and is easily achieved, but there are also shortcoming：
1.K-means algorithms are divided into K classes using k as parameter, n data so that similarity is high between class and inhomogeneity it
Between similarity it is relatively low.It is therefore desirable to which K values have been manually set in advance, the size of k values determines the number of class.Certainly, if thing
Several classes will be divided by first knowing, this problem is just readily solved, for example be to determine here about the classification of hepatopathy, Ke Yizhun
Really it is divided into liver cancer, hepatic cyst, hemangioma.
2.k-means algorithms are premised on Euclidean distance, selection and the nearest foreign peoples's sample of sample point and similar sample, from
It is exactly the size of space length from the point of view of in theorem in Euclid space, Euclidean distance is the difference of two sample points, square of each variable value difference
With finally find out square root again, purpose is exactly the overall distance i.e. dissimilarity for finding out therebetween.Unquestionable Euclidean away from
From being very useful, but it have the shortcomings that it is apparent.It all equally treats the different attribute relative difference of sample, this point
It is exactly undesirable when judging liver disease classification, each feature of liver disease is for judging classification
There is different importance, so what cannot be equal goes to treat.But weighting processing is done to the algorithm, make it more accurately will be sick
Become classification.
For not foot point 1 of K-means algorithms, the method for the present invention is read using three layer of two sub-argument, by four points of difficulty
Problem, the problem of being converted into three two points of difficulty, therefore when being classified here, take k=2, repeatedly split data into two classes.
It is using the Relief algorithms of optimization, the texture as classification foundation is special for not foot point 2 of k-means algorithms
Sign weighting, or the inoperative characteristic of removal, reduce operand and time, increase the precision of system model.After weighting
K-means core formula be：
Weight is realized in cluster process in the k-means functions that equivalent cluster is played a crucial role, and weighted
Wi be a nonnegative number.
Continue to choose other combined samples and repeat the above steps, multi-pass operation finishes until all combinations all calculate, such as schemes
Shown in 5.Finally selection obtains maximum diagnosis accuracy rate dimension target and establishes diagnostic model.The image texture characteristic substitution of input is examined
It is handled in disconnected model, obtains diagnostic result and accuracy.
Specific embodiments are merely illustrative of the spirit of the present invention described in the present embodiment.Skill belonging to the present invention
The technical staff in art field can make various modifications or additions to the described embodiments or using similar side
Formula substitutes, and however, it does not deviate from the spirit of the invention or beyond the scope of the appended claims.
Although the present invention has more used input module, image texture characteristic extraction module, classification diagnosis module and defeated
Go out module term, but it does not preclude the possibility of using other terms.The use of these items is only for more easily describe
With the essence for explaining the present invention；Any one of the additional limitations is construed as all to disagree with spirit of that invention.
Claims (6)
1. a kind of hepatopathy CT image computer assistant diagnosis systems based on data mining, it is characterised in that：The system comprises
There are sequentially connected input module, texture feature extraction module, classification diagnosis identification module and output module；
The input module is for importing the liver's CT images for passing through contours segmentation and extraction；
Described image texture feature extraction module is to carry out a series of images texture analysis for liver's CT images to extraction, is obtained
The characteristics of image of target image is obtained, described image is characterized as 9 dimension gray scale symbiosis textural characteristics；
The classification diagnosis module including grader and has the liver cancer of many cases definitive pathological diagnosis, a hepatic cyst and just in the module
The database of normal CT image for liver textural characteristics, the classification diagnosis module according to the trained diagnostic model in data with existing library,
And classification for statistical analysis to the image texture characteristic of input judges；
The output module is to show and generate report for exporting statistic of classification result, and statistical result includes that classification judges knot
Fruit and accuracy statistical indicator.
2. a kind of hepatopathy CT image computer aided diagnosis methods based on data mining, it is characterised in that：This method uses institute
The hepatopathy CT image computer assistant diagnosis systems based on data mining are stated, content mainly includes the following steps that：
(1) contours segmentation and extraction are carried out to abdominal CT images in advance, and has marked normal liver CT, hepatic cyst, liver cancer, it will be pre-
Treated, and image is imported into the system；
(2) image texture characteristic extraction module carries out analyzing image texture to importing image, obtains 13 dimension gray scales in CT images
Symbiosis textural characteristics；
(3) liver cancer, hepatic cyst image texture characteristic and the normal hepatocytes image liver that storage has confirmed in classification diagnosis module
The database that textural characteristics are constituted examines the CT image for liver textural characteristics substitution of input according to Database diagnostic model
It is handled in disconnected model classifiers, obtains diagnostic result and accuracy.
3. a kind of hepatopathy CT image computer aided diagnosis methods based on data mining according to claim 2, special
Sign is：It is described to be to abdominal CT images progress contours segmentation and extraction, process in step (1)：
Image is cut first to obtain ROI, then interested region is filled and is removed the processing of noise, adopt
Noise is removed with median filtering method, image enhancement is carried out using histogram equalization method；And 256 grades of gray scales are carried out to CT images
Conversion, is then stored as double types, finally carries out 16 grades of gray compressions, and the ROI image of all extractions is constituted Hepatic CT
Image set.
4. a kind of hepatopathy CT image computer aided diagnosis methods based on data mining according to claim 2, special
Sign is：In step (2), the 13 dimension gray scale symbiosis textural characteristics obtained in CT images obtain wherein 9 dimension gray scale symbiosis
The process of textural characteristics is：
Appoint in the picture and take a pixel A (x, y), then obtains the one other pixel point B (x+a, the y+ that are d with pixel A distances
B), pixel A and pixel B form a point pair, record the gray value (i, j) of this point pair, and the value of fixed a and b makes pixel
Point A (x, y) is moved on the image, obtains a variety of pixels to combination, Ng is gradation of image rank value, i ∈ [0, Ng-1], j ∈
[0, Ng-1], changes d and θ, and θ is pixel to line and horizontal angle, the gray value of the pixel pair in statistical picture,
Gray level co-occurrence matrixes P (i, j, θ, d) is constituted, wherein # { x } is the number of all elements in set x；Then according to gray scale symbiosis
Matrix extraction 9 dimension textural characteristics, respectively contrast, inverse difference moment, correlation, entropy, angular second moment, symbiosis and mean value, symbiosis and
Entropy, symbiosis and difference and symbiosis difference entropy.
5. a kind of hepatopathy CT image computer aided diagnosis methods based on data mining according to claim 2, special
Sign is：In step (3), described according to Database diagnostic model, the process for establishing diagnostic model is：
N textural characteristics data sample is chosen out of database as training sample set, is carried out with training sample set pair grader
Training, wherein n are positive integer, and are the 1/2 of entire database sample set, and the n in database in addition to training sample set is a
Textural characteristics data sample classifies to verification sample set as verification sample set, using grader after training, is verified
Then the discriminating accuracy rate of sample set still further chooses n sample and repeats the above steps, multi-pass operation is finally chosen and obtained most
Big accuracy rate of diagnosis dimension target establishes diagnostic model.
6. a kind of hepatopathy CT image computer aided diagnosis methods based on data mining according to claim 5, special
Sign is：It is described that n textural characteristics data sample is chosen out of database as training sample set, exactly chosen out of database
Liver textural characteristics data sample, be divided into training sample set and verification sample set, using Relief algorithms by the feature of extraction into
Row dimension-reduction treatment, using improved K-means clustering methods build grader, using training sample set pair grader into
Row training.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810313764.4A CN108805858A (en) | 2018-04-10 | 2018-04-10 | Hepatopathy CT image computers assistant diagnosis system based on data mining and method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810313764.4A CN108805858A (en) | 2018-04-10 | 2018-04-10 | Hepatopathy CT image computers assistant diagnosis system based on data mining and method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108805858A true CN108805858A (en) | 2018-11-13 |
Family
ID=64095546
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810313764.4A Pending CN108805858A (en) | 2018-04-10 | 2018-04-10 | Hepatopathy CT image computers assistant diagnosis system based on data mining and method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108805858A (en) |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109712705A (en) * | 2018-12-19 | 2019-05-03 | 中国石油大学(华东) | A kind of cholelithiasis intelligent diagnostics APP based on deep learning |
| CN110085288A (en) * | 2019-04-19 | 2019-08-02 | 四川大学华西医院 | A kind of liver and gall surgical department Internet-based treatment information sharing system and sharing method |
| CN110767293A (en) * | 2019-11-07 | 2020-02-07 | 辽宁医汇智健康科技有限公司 | Brain auxiliary diagnosis system |
| CN110910991A (en) * | 2019-11-21 | 2020-03-24 | 张军 | Medical automatic image processing system |
| CN111340824A (en) * | 2020-02-26 | 2020-06-26 | 青海民族大学 | Image feature segmentation method based on data mining |
| CN111754485A (en) * | 2020-06-24 | 2020-10-09 | 成都市温江区人民医院 | Artificial intelligence ultrasonic auxiliary system for liver |
| CN111950595A (en) * | 2020-07-14 | 2020-11-17 | 十堰市太和医院（湖北医药学院附属医院） | Liver focus image processing method, system, storage medium, program, and terminal |
| CN112330731A (en) * | 2020-11-30 | 2021-02-05 | 深圳开立生物医疗科技股份有限公司 | Image processing apparatus, image processing method, image processing device, ultrasound system, and readable storage medium |
| CN113314202A (en) * | 2020-02-26 | 2021-08-27 | 张瑞明 | System for processing medical images based on big data |
| CN113408603A (en) * | 2021-06-15 | 2021-09-17 | 西安理工大学 | Coronary artery stenosis degree identification method based on multi-classifier fusion |
| WO2023040164A1 (en) * | 2021-09-14 | 2023-03-23 | 之江实验室 | Method and apparatus for training pet/ct-based lung adenocarcinoma and squamous carcinoma diagnosis model |
| CN116188786A (en) * | 2023-05-04 | 2023-05-30 | 潍坊医学院附属医院 | Image segmentation system for hepatic duct and biliary tract calculus |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20130030278A1 (en) * | 2011-07-25 | 2013-01-31 | Seong Yeong-Kyeong | Apparatus and method for detecting lesion and lesion diagnosis apparatus |
| CN103593844A (en) * | 2013-10-29 | 2014-02-19 | 华中科技大学 | Extraction method of multiple multi-dimensional features of medical images |
| CN104000619A (en) * | 2014-06-16 | 2014-08-27 | 彭文献 | Thyroid CT image computer-aided diagnosis system and method |
| CN105931224A (en) * | 2016-04-14 | 2016-09-07 | 浙江大学 | Pathology identification method for routine scan CT image of liver based on random forests |
| CN105956620A (en) * | 2016-04-29 | 2016-09-21 | 华南理工大学 | Liver ultrasonic image identification method based on sparse expression |
| CN107748900A (en) * | 2017-11-08 | 2018-03-02 | 山东财经大学 | Tumor of breast sorting technique and device based on distinction convolutional neural networks |
-
2018
- 2018-04-10 CN CN201810313764.4A patent/CN108805858A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20130030278A1 (en) * | 2011-07-25 | 2013-01-31 | Seong Yeong-Kyeong | Apparatus and method for detecting lesion and lesion diagnosis apparatus |
| CN103593844A (en) * | 2013-10-29 | 2014-02-19 | 华中科技大学 | Extraction method of multiple multi-dimensional features of medical images |
| CN104000619A (en) * | 2014-06-16 | 2014-08-27 | 彭文献 | Thyroid CT image computer-aided diagnosis system and method |
| CN105931224A (en) * | 2016-04-14 | 2016-09-07 | 浙江大学 | Pathology identification method for routine scan CT image of liver based on random forests |
| CN105956620A (en) * | 2016-04-29 | 2016-09-21 | 华南理工大学 | Liver ultrasonic image identification method based on sparse expression |
| CN107748900A (en) * | 2017-11-08 | 2018-03-02 | 山东财经大学 | Tumor of breast sorting technique and device based on distinction convolutional neural networks |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109712705A (en) * | 2018-12-19 | 2019-05-03 | 中国石油大学(华东) | A kind of cholelithiasis intelligent diagnostics APP based on deep learning |
| CN110085288A (en) * | 2019-04-19 | 2019-08-02 | 四川大学华西医院 | A kind of liver and gall surgical department Internet-based treatment information sharing system and sharing method |
| CN110767293A (en) * | 2019-11-07 | 2020-02-07 | 辽宁医汇智健康科技有限公司 | Brain auxiliary diagnosis system |
| CN110767293B (en) * | 2019-11-07 | 2023-11-21 | 辽宁医汇智健康科技有限公司 | Auxiliary brain diagnosis system |
| CN110910991A (en) * | 2019-11-21 | 2020-03-24 | 张军 | Medical automatic image processing system |
| CN113314202A (en) * | 2020-02-26 | 2021-08-27 | 张瑞明 | System for processing medical images based on big data |
| CN111340824A (en) * | 2020-02-26 | 2020-06-26 | 青海民族大学 | Image feature segmentation method based on data mining |
| CN111754485A (en) * | 2020-06-24 | 2020-10-09 | 成都市温江区人民医院 | Artificial intelligence ultrasonic auxiliary system for liver |
| CN111950595A (en) * | 2020-07-14 | 2020-11-17 | 十堰市太和医院（湖北医药学院附属医院） | Liver focus image processing method, system, storage medium, program, and terminal |
| CN112330731A (en) * | 2020-11-30 | 2021-02-05 | 深圳开立生物医疗科技股份有限公司 | Image processing apparatus, image processing method, image processing device, ultrasound system, and readable storage medium |
| CN113408603A (en) * | 2021-06-15 | 2021-09-17 | 西安理工大学 | Coronary artery stenosis degree identification method based on multi-classifier fusion |
| CN113408603B (en) * | 2021-06-15 | 2023-10-31 | 西安华企众信科技发展有限公司 | Coronary artery stenosis degree identification method based on multi-classifier fusion |
| WO2023040164A1 (en) * | 2021-09-14 | 2023-03-23 | 之江实验室 | Method and apparatus for training pet/ct-based lung adenocarcinoma and squamous carcinoma diagnosis model |
| CN116188786A (en) * | 2023-05-04 | 2023-05-30 | 潍坊医学院附属医院 | Image segmentation system for hepatic duct and biliary tract calculus |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108805858A (en) | Hepatopathy CT image computers assistant diagnosis system based on data mining and method | |
| Qi et al. | Automated diagnosis of breast ultrasonography images using deep neural networks | |
| CN109036547A (en) | A kind of lung CT image computer aided system and method based on clustering | |
| Lee et al. | Random forest based lung nodule classification aided by clustering | |
| Andrade et al. | Classifier approaches for liver steatosis using ultrasound images | |
| CN101742961B (en) | Diagnosis support device and system | |
| CN101103924A (en) | Galactophore cancer computer auxiliary diagnosis method based on galactophore X-ray radiography and system thereof | |
| WO2022110525A1 (en) | Comprehensive detection apparatus and method for cancerous region | |
| Cabral et al. | Fractal analysis of breast masses in mammograms | |
| Wang et al. | MARnet: Multi-scale adaptive residual neural network for chest X-ray images recognition of lung diseases | |
| Anshad et al. | Recent methods for the detection of tumor using computer aided diagnosis—A review | |
| Al-Tam et al. | Breast cancer detection and diagnosis using machine learning: a survey | |
| Lu et al. | Breast tumor computer-aided detection system based on magnetic resonance imaging using convolutional neural network | |
| CN115564756A (en) | Medical image focus positioning display method and system | |
| Aggarwal et al. | Patient-Wise Versus Nodule-Wise Classification of Annotated Pulmonary Nodules using Pathologically Confirmed Cases. | |
| Zhang et al. | Cost-sensitive ensemble classification algorithm for medical image | |
| Liu et al. | Novel Approach for Automatic Region of Interest and Seed Point Detection in CT Images Based on Temporal and Spatial Data. | |
| CN111265234A (en) | Method and system for judging properties of lung mediastinal lymph nodes | |
| Xiong et al. | Deep Ensemble Learning Network for Kidney Lesion Detection | |
| Kim et al. | Robust local explanations for healthcare predictive analytics: An application to fragility fracture risk modeling | |
| Umamaheswari et al. | Literature review on breast cancer diagnosis using 3D images: methods and performance analysis | |
| Dhimmar et al. | Breast Cancer Detection Using Classification Algorithms | |
| Liu et al. | Texture feature extraction from thyroid MR imaging using high-order derived mean CLBP | |
| Kim et al. | Mass lesions classification in digital mammography using optimal subset of BI-RADS and gray level features | |
| Li et al. | Prediction of Short-Term Breast Cancer Risk with Fusion of CC-and MLO-Based Risk Models in Four-View Mammograms |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20181113 |
|
| WD01 | Invention patent application deemed withdrawn after publication |