CN104217213B - A kind of medical image multistage sorting technique based on symmetric theory - Google Patents
A kind of medical image multistage sorting technique based on symmetric theory Download PDFInfo
- Publication number
- CN104217213B CN104217213B CN201410409810.2A CN201410409810A CN104217213B CN 104217213 B CN104217213 B CN 104217213B CN 201410409810 A CN201410409810 A CN 201410409810A CN 104217213 B CN104217213 B CN 104217213B
- Authority
- CN
- China
- Prior art keywords
- image
- medical image
- classification
- symmetry
- texture
- 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.)
- Active
Links
Landscapes
- Apparatus For Radiation Diagnosis (AREA)
- Image Analysis (AREA)
- Measuring And Recording Apparatus For Diagnosis (AREA)
Abstract
The invention belongs to medical information technical field, and in particular to a kind of medical image multistage sorting technique based on symmetric theory.The present invention includes：Image to be classified proposes classification request, and image to be classified should be raw medical image data；Image preprocessing process：Image modeling；Multistage classifies；Show result.The concept of weak symmetry proposed by the present invention and strong symmetry is that one of medical image is redefined.It is proposed that weak symmetrical Deciding Algorithm and strong symmetrical Deciding Algorithm realize the multistage classification of medical image.The classification accuracy of this multistage classification is very high, each stage, directly linking was successively goed deep into, so as to improve the diagnostic accuracy of doctor and shorten Diagnostic Time.The classification of medical image is realized using symmetric theory, makes the medical image multistage sorting technique based on symmetric theory that there is the accuracy of higher.
Description
Technical field
The invention belongs to medical information technical field, and in particular to a kind of medical image multistage based on symmetric theory
Sorting technique.
Background technology
Due to containing abundant image and medical information in medical image, in recent years towards the data mining of medical image
Technology becomes the hot spot of medicine and computer cross discipline research.With the fast development of medical digital equipment, medical information
Database is widely used.The structured text information of patient, and substantial amounts of unstructured medical image, are medicine figure
The data mining of picture provides abundant data resource.Medical image can effectively aid in doctor during diagnosis to pathology
Region of variation is detected, positions and judges the good pernicious of it, therefore is widely used in clinical diagnostic process.However,
Even if the doctor with different knowledge background to same medical image there may be different judgements, so, maintenance data dig
Technique study classification method of medical image is dug, auxiliary doctor is diagnosed according to medical image, improves its efficiency and precision, is had
There are higher learning value and actual application prospect.
At present, both at home and abroad in terms of Medical Images Classification research, the sorting technique mainly used includes statistical method, nerve
Network method, Fuzzy Pattern Recognition Method, machine learning method etc..Existing sorting technique is to be just different by Medical Images Classification
Often or abnormal position is positioned, and the successively deep multistage sorting technique of medical image is temporarily not carried out.Medical image
Imaging results show that the perpendicular bisector both sides on image are near symmetricals, and healthy image is in image gray levels distribution and image
Approximate symmetrical structure is all presented on shaped position.And the image that lesion occurs will destroy the structure of this near symmetrical.For this
It is proposed a kind of knowledge that can make full use of the symmetry of medical image in itself realize multistage sorting technique be one urgently
Solve the problems, such as.
The content of the invention
The purpose of the present invention is to propose to a kind of doctor that medical image multistage classification accuracy is improved based on symmetric theory
Learn image classification method.
The object of the present invention is achieved like this：
The present invention includes the following steps：
(1) image to be classified proposes classification request：Image to be classified should be raw medical image data；
(2) image preprocessing process：To original medical image extraction area-of-interest (ROI, Region Of
Interest), the grey level histogram of image ROI region is calculated, obtains the trough list of the grey level histogram of image ROI region,
According to trough list to image grading texture feature extraction, according to the classification texture image standardization that actual needs will obtain to system
One size；ROI region is divided into the left and right sides by the perpendicular bisector on image ROI region, in classification texture image
Vertical line is divided into the left and right sides；
(3) image modeling：According to the strong symmetry of medical image and the concept of weak symmetry, the medical image multistage is established
The graph model of classification；
(4) multistage classifies：Weak symmetry decision method based on grey level histogram Intersection, to medical image thicker
The classification of first stage has been carried out in granularity；Strong symmetry decision method based on point symmetry, with reference to weak symmetry decision method,
More fine-grained second stage classification has been carried out for abnormal image to first stage classification results, located the position of lesion region
Put；Finally, using the feature extracted to lesion region, the classification of phase III has been carried out to lesion region；
(5) result is shown：Multistage sorting technique based on symmetric theory is real by the image in raw image database
Now classify.
Image modeling is：Medical image multistage disaggregated model is established according to the concept of weak symmetry and strong symmetry, it is weak
Symmetry is：For a medical image G, weak symmetry refer to D in G (L) and D (R) each group away from K in share pixel number
Mesh, weak symmetry is weighed with their intersecting distances
Wherein, group is the width of every group of histogram away from K.
Symmetry is by force：Strong symmetry refer to the number of the point in circle or on circle in TI number, with their radius model
Enclose to weigh strong symmetry rad (v (i))=mov (v (i)).
Multistage is categorized as：Weak symmetry decision method is applied to carry out the first rank compared with coarseness again medical image first
The classification of section；Then it is to first stage classification results using strong symmetry decision method and combining for weak symmetry decision method
Abnormal image carries out the classification of more fine-grained second stage, the position in lesion region；Finally, using to lesion region
The feature extracted, has carried out lesion region the classification of phase III.
The beneficial effects of the present invention are：
The concept of weak symmetry proposed by the present invention and strong symmetry is that one of medical image is redefined.It is proposed weak
Symmetry decision method and strong symmetry decision method realize the multistage classification of medical image.Point of this multistage classification
Class accuracy rate is very high, each stage, directly linking was successively goed deep into, so that when improving the diagnostic accuracy of doctor and shortening diagnosis
Between.The classification of medical image is realized using symmetric theory, makes the medical image multistage sorting technique based on symmetric theory
Accuracy with higher.
Brief description of the drawings
Fig. 1 is the process on perpendicular bisector symmetry division；
Fig. 2 is MI iconic models；
Fig. 3 is weak symmetry example；
Fig. 4 is the example for actively changing and passively changing
Fig. 5 is strong symmetry example；
Fig. 6 is the medical image multistage sorting technique flow chart based on symmetric theory；
Embodiment
The present invention is further illustrated with specific embodiment below in conjunction with the accompanying drawings：
Medical image is pre-processed first：
1. each original brain CT image zooming-out ROI region in pair original image storehouse；
2. interception ROI region simultaneously corrects；
3. calculating the trough distribution situation of image ROI region grey level histogram, the trough table of grey level histogram is obtained；
4. threshold value is set repeatedly to extract texture to image according to trough table, so as to obtain Mipmap image；
5. it is finally image that size is COLUMN × ROW by Mipmap image normalization；
6. image ROI region is divided into left and right two parts on perpendicular bisector；
7. the texture image after size of standardizing is divided into left and right two parts on perpendicular bisector；
Dividing processing is carried out to image and is stored in corresponding database, by above procedure, each original image
Correspond to four equal-sized images.Calculate grey image R OI regions d (L, R) values and texture image rad (v (i))=
Mov (v (i)) value；The value of gray level image d (L, R) is calculated, realizes the classification of the first stage of medical image；Then using being based on
The weak symmetry d (L, R) of gray level image and the joint realization doctor based on the strong symmetry rad (v (i)) of Texture Points on texture image
Learn the classification of the second stage of image, the position in lesion region；Feature finally is extracted to lesion region, utilizes these features
Realize the classification of the phase III of medical image；
A specific image classification process is as follows：
1. a couple medical image MI to be sorted proposes classification request；
2. a couple medical image MI to be sorted is pre-processed, it is (symmetrically grey on perpendicular bisector to obtain corresponding four images
Spend image L, R and texture image Lh、Rh)；
3. d (L, R) value of gray level image L and R is obtained according to the definition of weak symmetry, using weak symmetry decision method pair
Medical image realizes the classification of first stage, that is, is divided into normal medical image and abnormal medical image；
4. the relation for actively changing and passively changing in medical image is defined according to the medical knowledge of medical image, i.e., it is main
Dynamic change is less than passive change, and the relation for actively changing and passively changing is present in normal medical image and abnormal medical image
In, calculated by val=d (L, L ')-d (R, R '), wherein the value of val is the number of magnitude relationship between actively change and passive change
Value, d (L, L ') are the values of the weak symmetry on the left of normal medical image and on the left of abnormal medical image, and d (R, R ') it is normally to cure
Learn the value of the weak symmetry on the right side of image and on the right side of abnormal medical image；
5. abnormal medical image in pair first stage carries out the classification of second stage, calculate abnormal medical image with it is normal
The active change of medical image and relation value val=d (L, L ')-d (R, R ') passively changed, if val<0, then abnormal medical figure
It happens is that in the left side L ' of picture and actively change, right side R ' happens is that passive change.So abnormal area T is present in abnormal doctor
In the left side L ' for learning image.Otherwise val is worked as>0, abnormal area is present in the right side R ' of abnormal medical image.Point of second stage
Class realizes the positioning to abnormal area T；
6. the edge contour in extraction T regions simultaneously calculates following characteristics value： Wherein C is tight
Degree of gathering, davg are that Average normalized radius, Ro are roughness, δ2It is that variance, Sk are that degree of skewness, Pe are peak values, A is tumor's profiles
Area, p is the girth of tumor's profiles.N is the number put on edge, if the center of the minimum enclosed rectangle of ROI region is
(x0, y0), radius d (i) is each marginal point (xi, yi) arrive picture centre (x0, y0) distance, K is the quantity put on edge, LiIt is
Euclidean distance between putting and putting on edge, μ is average, using iteration stochastical sampling method, passes through spectral clustering in sampling every time
Estimation does not mark the class label of sample, carries out model learning using SVM, successive optimization model, to improve classification accuracy, is realized
The classification of phase III, i.e., be categorized as benign and malignant two class by abnormal area T；
Above example shows that the medical image multistage sorting technique proposed by the present invention based on symmetric theory has real
The application value on border.
The present invention also has so some technical characteristics：
The medical image model 1. the multistage classifies：
One medical image (Medical Image, MI) be a set G=(L, R, T) | L={ v1, v2..., vn, n
It is the number of MI left pixels point, vi∈[0,255].R={ u1, u2..., um, m is the number of MI right pixels point, ui∈[0,
255], T={ w1, w2..., wp, 0<p<N or 0<p<M, wi∈[0,255]}。
Wherein L and R is the set that MI is divided equally rear left side and right pixel by median perpendicular respectively.T is abnormal area in MI
Pixel set,Or.If MI is normal, T collection is combined into sky, and otherwise T set is not empty.
In G, since exception occur in some genes in organ, when the chromosome of normal cell has run into this gene just
It can lure that cell is abnormal breeding into, hinder the growth of normal cell, so that abnormal side cell mushrooms out, and opposite side
Then since the reasons such as extruding covering hinder the growth of normal cell, it is set passively to deform upon.Since the presence in T regions makes L
Change with R set.It is both that abnormal side pixel set varies widely on the image that this medical knowledge, which embodies, and
The change that non-abnormal side pixel set occurs is small compared with abnormal side.Fig. 2 (a) is normal medicine brain image on brain
Median perpendicular is bisected into left and right two parts, and pixel is evenly distributed during L and R gathers in G.Fig. 2 (b) is abnormal brain
Portion's image instance, we can be found that in G there are an abnormal area T to gather.Due to the presence of T set, L and R gathers in G
There is very big difference in the distribution of middle pixel.
In abnormal image (Abnormal Image, AMI), if, then the change that the left side of AMI occurs is referred to as actively
Change, the change that right side occurs are known as passive change；If, then the change that the right side of AMI occurs is known as actively changing, left
The change that side occurs is known as passive change.
In Fig. 3 (b), L and R pixel set difference is very big, causes the presence in T regions., i.e., it is abnormal to appear in a left side
Side, so the change that left side occurs in Fig. 3 (b) is exactly actively to change, right side is passively to change.How in G L and R pixel is determined
The difference of setIf D (X) be pixel set X containing K group away from grey level histogram, weak symmetry is given below
Definition.
For a medical image G, weak symmetry refer to D in G (L) and D (R) each group away from K in share pixel number
Mesh, weak symmetry is weighed with their intersecting distances
Wherein, group is the width of every group of histogram away from K.
Fig. 3 is the grey level histogram of left, right parts of images in Fig. 2 (a).Abscissa represents the distribution of gray value, takes
It is worth scope [0 255], ordinate represents the number K of the pixel on a certain gray value.L and R can be learnt according to above formula
The distance that intersects of pixel set.The part that weak bigger explanation L and the R of symmetry d (L, R) intersects is more, and difference is smaller.It is weak symmetrical
The part that property d (L, R) smaller explanation L and R intersects is fewer, and difference is bigger.
Include normal medical image (Normal Medical Image, NMI) and the medical image of exception in G
How (Abnormal Medical Image, AMI), distinguish them, be given below in NMI and AMI and actively change
The relation passively changed.The relation for actively changing and passively changing is present in the contrast of the weak symmetry of NMI and AMI, and main
Dynamic change is less than passive change.
The relation for actively changing and passively changing is as shown in Figure 4.Fig. 4 (a) is NMI, and Fig. 4 (b) is AMI.Fig. 4 (c) (d) is
The histogram on the left of Fig. 4 (a) and right side, Fig. 4 (e) (f) are the histograms on the left of Fig. 4 (b) and right side.If the left side of known NMI
L and R are combined into right pixel collection, the left side of AMI and right pixel collection are combined into L ' and R '.Position abnormal areaOr
Using the following formula.
Val=d (L, L')-d (R, R')
If val<0, then it happens is that in the L ' of AMI and actively change, R ' happens is that passive change.So T regions exist
In the L ' of AMI.Otherwise val is worked as>0, T region is present in R '.Ash can be positioned according to active change and the relation passively changed
Spend the position of abnormal area T in level medical image.According to formula, the val of (a) and (b) in Fig. 4>0, i.e. T regions appear in Fig. 4
(b) right side.
The texture image of medical image is made of more textures, is made of per texture several points.The corresponding textures of NMI
Image divide equally on brain median perpendicular after left and right sides near symmetrical, the definition of strong symmetry is given according to this principle.
In the texture image of AMI, the texture bar number that the side presentation of actively change occurs is more than the side texture bar number passively changed.
One texture image (Texture Image, TI) is a set F={ (Lh, Rh)|Lh={ Lh1,Lh2...,
Lhn, LhiIt is LhIn the i-th texture, Lhi={ v (1), v (2) ..., v (n) }, v (i) represent composition LhiThe point of texture.Rh=
{Rh1,Rh2..., Rhn, RhiIt is RhIn the i-th texture, Rhj={ v (1), v (2) ..., v (m) }, v (i) represent composition Rhi
The point of texture }, wherein LhAnd RhIt is the set for the left side and right side texture divided equally on median perpendicular.
In texture image TI, to form the point v (i) of texture as the center of circle, the mobility mov (v (i)) of point v (i) is half
Do circle in footpath.Point in circle or on circle is regarded as and the point symmetry.
Strong symmetry refer to the number of the point in circle or on circle in TI number, it is strong right to be weighed with their radius
Title property rad (v (i))=mov (v (i)).
Fig. 5 (a) is that texture image a TI, Fig. 5 (b) and Fig. 5 (c) are the partial enlarged views of Fig. 5 (a) and on center line pair
Claim.Side in selection TI more than Texture Points is that left side is used as template, the rad (v each put according to the concept of strong symmetry mark
(i)), as shown in Fig. 5 (d), represent that significance index a little is different with the point of different colours in figure.Fig. 5 (f) be Fig. 5 (d) on
The symmetric graph of center line.Fig. 5 (g) is the exemplary plot of strong symmetry rad (v (i)).
2. the specific steps of weak symmetry decision method：
For G, the distribution of pixel is important feature.Judge whether T regions deposit by weak symmetry d (L, R)
, it is possible to determine that whether medical image is normal.Compare for image application feature 1 existing for T regions in NMI and AMI and actively become
The relation changed and passively changed.Known L and R are the set in the left side and right pixel in NMI, L ' and R ' be in AMI left side and
The set of right pixel.The weak symmetry d (R, R ') of weak symmetry d (L, L ') and the R and R ' of L and L ' is calculated, if d (L, L ')<d
The change that pixel distribution occurs in (R, R '), then in L and L ' is more than the change that pixel distribution occurs in R and R '.Show that T regions go out
In the L ' of present AMI, i.e.,.Otherwise in the R ' for appearing in AMI, i.e.,
The specific steps of the last 3. symmetry decision method：
Since weak symmetric characteristics are insensitive to the change such as the direction of image or image-region, size, position, so weak right
Title property feature cannot catch the local attribute of objects in images well, and set forth herein the strong symmetry decision method solution of image
This problem.Different from grey level histogram feature, textural characteristics are not based on the feature of pixel, it needs including multiple pictures
Statistics calculating is carried out in the region of vegetarian refreshments.The position of gray-level pixels point in the local positioning image of textural characteristics energy.Each
What texture was all made of the point of gray scale acute variation, the integrated symmetric of texture image is judged by the definition of strong symmetry
Property.Each point v (i) divides the point v ' (i) for being mapped to opposite side equally on median perpendicular on texture, can according to the definition of strong symmetry
To draw the mobile range of point v ' (i).Point so in mobile range is regarded as and origin symmetry.
Texture image set F={ Lh, RhIn, L in the texture image of normal medical imagehAnd RhIt is near symmetrical.Such as
There is lesion region in fruit, then the side texture bar number of lesion region will be more, that is, the points for forming texture increase.Scan LhWith
RhThe number of the point of middle composition texture, it is assumed that the template of first time is used as using the side more than the number of point.Mapped on symmetry axis
Elimination method to opposite side application Texture Points eliminates all points in mobile range.Remaining point after eliminating for the first time
Markf=suml-sumr.Then L is calculatedhAnd RhIn remaining point number, secondary template is used as using the side of non-template
Eliminate the point of opposite side.Remaining point markl=suml-sumr after second of cancellation.Compare the number sum of both sides left point
=markf+markl, if sum >=0 illustrates that there are T regions, sum in left side<There are T regions on 0 explanation right side.It is right by force here it is image
Claim the basic thought of sex determination method.
4. the specific steps of medical image multistage sorting technique：
Learnt by the investigation to doctor, it carries out diagnosis process using medical image and is：First according to the medicine of doctor
Knowledge judges whether medical image MI is normal.Secondly, AMI is continued to study, it is different according to the clinical diagnosis micro-judgment of doctor
Often appear in L or R.Finally, it is further to the region T in AMI to be studied, judged according to the conventional diagnostic experiences of doctor
T regions it is good pernicious.The diagnosis process for simulating doctor herein proposes MI multistage graders (multi-stage
Classification, MSC) sorting technique, it realizes that process is as shown in Figure 6.
MSC-1：First stage classification has been carried out to medical image MI using WSDA, i.e., NMI and AMI have been categorized as to MI.
MSC-2：Further processing obtains corresponding texture image F to the image AMI for being to mark, while applies WSDA
Judge left sides or right side of the abnormal area T in medical image with SSDA.
MSC-3：Herein according to the edge contour of abnormal area T, the multiple features realization for extracting T is good pernicious to AMI progress
Classification.
The key of the present invention is to analyze medical image using medical image symmetric theory, so that using weak right
Sex determination method and strong symmetry decision method is claimed to carry out multistage classification to image.
The present invention takes into full account medical image structural and regular in itself.By divide the image into for left and right approximation it is right
Two parts of title can quickly realize multistage classification, also disclosed the feature of the medical image of lesion, so that this is more
Stage Classification method is in itself closer to the right mind process of the mankind.
Claims (1)
1. a kind of medical image multistage sorting technique based on symmetric theory, it is characterised in that include the following steps：
(1) image to be classified proposes classification request, and image to be classified should be raw medical image data；
(2) image preprocessing process：Region of interest ROI is extracted to original medical image, the gray scale for calculating image ROI region is straight
Fang Tu, obtains the trough list of the grey level histogram of image ROI region, special to image grading extraction texture according to trough list
Sign, according to the classification texture image standardization that actual needs will obtain to unified size；Perpendicular bisector on image ROI region
ROI region is divided into the left and right sides, it is the left and right sides that the perpendicular bisector on being classified texture image, which will be classified Study Of Segmentation Of Textured Images,；
(3) image modeling：According to the strong symmetry of medical image and the concept of weak symmetry, the classification of medical image multistage is established
Graph model；
(4) multistage classifies：Weak symmetry decision method based on grey level histogram Intersection, to medical image compared with coarseness
On carried out the classification of first stage；Strong symmetry decision method based on point symmetry, with reference to weak symmetry decision method, to
One stage classification results have carried out more fine-grained second stage classification for abnormal image, located the position of lesion region,
Specifically include：Texture image set F={ Lh, RhIn, L in the texture image of normal medical imagehAnd RhIt is near symmetrical,
If there is lesion region, then the side texture bar number of lesion region will be more, that is, the points for forming texture increase；Scan Lh
And RhThe number of the point of middle composition texture, it is assumed that the template of first time is used as using the side more than the number of point；Reflected on symmetry axis
The elimination method for being mapped to opposite side application Texture Points eliminates all points in mobile range；Remaining point after eliminating for the first time
Markf=suml-sumr；Then L is calculatedhAnd RhIn remaining point number, secondary template is used as using the side of non-template
Eliminate the point of opposite side；Remaining point markl=suml-sumr after second of cancellation；Compare the number sum=of both sides left point
Markf+markl, if sum >=0 illustrates that there are T regions, sum in left side<There are T regions on 0 explanation right side；Finally, using to lesion
The feature that region is extracted, has carried out lesion region the classification of phase III；
(5) result is shown：Multistage sorting technique based on symmetric theory, which realizes the image in raw image database, divides
Class；
The image modeling process is：Medicine is established according to the concept of the concept of the strong symmetry of medical image and weak symmetry
The graph model of image multistage classification；
The weak symmetry is：For a medical image G, weak symmetry refers to D in G (L) and D (R) at each group away from K
Shared number of pixels, distances for being intersected with them weigh weak symmetry, and D (X) is that pixel set X contains K group
Away from grey level histogram, L and R are the set that medical image MI is divided equally rear left side and right pixel by median perpendicular respectively
Wherein, group is the width of every group of histogram away from K；
Strong symmetry refer to the number of the point in circle or on circle in texture image TI number, weighed with their radius
Strong symmetry rad (v (i))=mov (v (i))；V (i) represents composition RhiThe point of texture, RhiIt is RhIn the i-th texture, RhIt is
The set for the right side texture divided equally on median perpendicular, mov (v (i)) are the mobility of point v (i)；
The multistage assorting process is：Weak symmetry decision method based on grey level histogram Intersection, to medical image
The classification of first stage has been carried out on compared with coarseness；Strong symmetry decision method based on point symmetry, sentences with reference to weak symmetry
Determine method, having carried out more fine-grained second stage to first stage classification results for abnormal image classifies, and located lesion
The position in region；Finally, using the feature extracted to lesion region, the classification of phase III has been carried out to lesion region；Tool
Body includes：
(4.1) classification request is proposed to medical image MI to be sorted；
(4.2) medical image MI to be sorted is pre-processed, corresponding four images is obtained, on the symmetrical gray scale of perpendicular bisector
Image L, R and texture image Lh、Rh,；
(4.3) d (L, R) value of gray level image L and R is obtained according to the definition of weak symmetry, using weak symmetry decision method pair
Medical image realizes the classification of first stage, that is, is divided into normal medical image and abnormal medical image；
(4.4) relation for actively changing and passively changing in medical image is defined according to the medical knowledge of medical image, i.e., it is main
Dynamic change is less than passive change, and the relation for actively changing and passively changing is present in normal medical image and abnormal medical image
In, calculated by val=d (L, L ')-d (R, R '), wherein the value of val is the number of magnitude relationship between actively change and passive change
Value, d (L, L ') are the values of the weak symmetry on the left of normal medical image and on the left of abnormal medical image, and d (R, R ') it is normally to cure
Learn the value of the weak symmetry on the right side of image and on the right side of abnormal medical image；
(4.5) classification of second stage is carried out to the abnormal medical image in the first stage, calculate abnormal medical image with it is normal
The active change of medical image and relation value val=d (L, L ')-d (R, R ') passively changed, if val<0, then abnormal medical figure
It happens is that in the left side L ' of picture and actively change, right side R ' happens is that passive change；So abnormal area T is present in abnormal doctor
In the left side L ' for learning image；Otherwise val is worked as>0, abnormal area is present in the right side R ' of abnormal medical image；Point of second stage
Class realizes the positioning to abnormal area T；
(4.6) extract the edge contour in T regions and calculate following characteristics value： Wherein C is tight
Degree of gathering, davg are that Average normalized radius, Ro are roughness, δ2It is that variance, Sk are that degree of skewness, Pe are peak values, A is tumor's profiles
Area, p is the girth of tumor's profiles；N is the number put on edge, if the center of the minimum enclosed rectangle of ROI region is
(x0, y0), radius d (i) is each marginal point (xi, yi) arrive picture centre (x0, y0) distance, K is the quantity put on edge, LiIt is
Euclidean distance between putting and putting on edge, μ is average, using iteration stochastical sampling method, passes through spectral clustering in sampling every time
Estimation does not mark the class label of sample, carries out model learning using SVM, successive optimization model, to improve classification accuracy, is realized
The classification of phase III, i.e., be categorized as benign and malignant two class by abnormal area T；
The pretreated database of the medical image is：By being pre-processed to every image in existing image library,
Then multistage classification image modeling is carried out to pretreated image set and obtains a multistage classification atlas D={ G1,
G2,…,Gn}。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410409810.2A CN104217213B (en) | 2014-08-20 | 2014-08-20 | A kind of medical image multistage sorting technique based on symmetric theory |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410409810.2A CN104217213B (en) | 2014-08-20 | 2014-08-20 | A kind of medical image multistage sorting technique based on symmetric theory |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104217213A CN104217213A (en) | 2014-12-17 |
| CN104217213B true CN104217213B (en) | 2018-04-27 |
Family
ID=52098676
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410409810.2A Active CN104217213B (en) | 2014-08-20 | 2014-08-20 | A kind of medical image multistage sorting technique based on symmetric theory |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104217213B (en) |
Families Citing this family (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104573742B (en) * | 2014-12-30 | 2018-03-23 | 中国科学院深圳先进技术研究院 | Classification method of medical image and system |
| CN104933712B (en) * | 2015-06-12 | 2017-12-01 | 哈尔滨工程大学 | Graph model methods of exhibiting based on brain CT images |
| CN106446923B (en) * | 2016-05-25 | 2019-08-06 | 哈尔滨工程大学 | Classification method of medical image based on corners Matching |
| CN109272508B (en) * | 2018-08-02 | 2021-10-26 | 哈尔滨工程大学 | Petri network image segmentation method based on rough set and rough entropy |
| CN113705723A (en) * | 2019-05-24 | 2021-11-26 | 上海联影智能医疗科技有限公司 | Image classification system, method and computer equipment |
| CN111488933B (en) * | 2020-04-13 | 2024-02-27 | 上海联影智能医疗科技有限公司 | Image classification method, network, computer device, and storage medium |
| CN111242144B (en) * | 2020-04-26 | 2020-08-21 | 北京邮电大学 | Method and device for detecting abnormality of power grid equipment |
| CN113313203B (en) * | 2021-06-22 | 2022-11-01 | 哈尔滨工程大学 | Medical image classification method based on extension theory and deep learning |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102467667A (en) * | 2010-11-11 | 2012-05-23 | 江苏大学 | Classification method of medical image |
| CN102509109A (en) * | 2011-10-20 | 2012-06-20 | 西北民族大学 | Method for distinguishing Thangka image from non-Thangka image |
| CN103226582A (en) * | 2013-04-08 | 2013-07-31 | 哈尔滨工程大学 | Medical image retrieving method on basis of uncertain fixed point image |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104798105B (en) * | 2012-11-20 | 2019-06-07 | 皇家飞利浦有限公司 | Using the integrated phenotype of image texture characteristic |
-
2014
- 2014-08-20 CN CN201410409810.2A patent/CN104217213B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102467667A (en) * | 2010-11-11 | 2012-05-23 | 江苏大学 | Classification method of medical image |
| CN102509109A (en) * | 2011-10-20 | 2012-06-20 | 西北民族大学 | Method for distinguishing Thangka image from non-Thangka image |
| CN103226582A (en) * | 2013-04-08 | 2013-07-31 | 哈尔滨工程大学 | Medical image retrieving method on basis of uncertain fixed point image |
Non-Patent Citations (2)
| Title |
|---|
| 人脑CT图像特征提取及其分类;林伟曦;《中国优秀硕士学位论文全文数据库 信息科技辑》;20111215(第S1期);I138-1663 * |
| 多模态影像脑部疾病检索研究;黎维娟;《中国优秀硕士学位论文全文数据库 信息科技辑》;20110415(第04期);I138-1144 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104217213A (en) | 2014-12-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104217213B (en) | A kind of medical image multistage sorting technique based on symmetric theory | |
| CN110443798B (en) | Autism detection method, device and system based on magnetic resonance image | |
| CN107886514B (en) | Mammary gland molybdenum target image lump semantic segmentation method based on depth residual error network | |
| Hariraj et al. | Fuzzy multi-layer SVM classification of breast cancer mammogram images | |
| CN106296653B (en) | Brain CT image hemorrhagic areas dividing method and system based on semi-supervised learning | |
| Li et al. | Dilated-inception net: multi-scale feature aggregation for cardiac right ventricle segmentation | |
| CN112101451B (en) | Breast cancer tissue pathological type classification method based on generation of antagonism network screening image block | |
| CN109447998B (en) | Automatic segmentation method based on PCANet deep learning model | |
| CN108257135A (en) | The assistant diagnosis system of medical image features is understood based on deep learning method | |
| CN110942446A (en) | Pulmonary nodule automatic detection method based on CT image | |
| CN108648182B (en) | Breast cancer nuclear magnetic resonance image tumor region segmentation method based on molecular subtype | |
| CN109145921A (en) | A kind of image partition method based on improved intuitionistic fuzzy C mean cluster | |
| Pan et al. | Mitosis detection techniques in H&E stained breast cancer pathological images: A comprehensive review | |
| CN109978880A (en) | Lung tumors CT image is carried out sentencing method for distinguishing using high dimensional feature selection | |
| CN102737379A (en) | Captive test (CT) image partitioning method based on adaptive learning | |
| CN104616289A (en) | Removal method and system for bone tissue in 3D CT (Three Dimensional Computed Tomography) image | |
| CN112669254A (en) | Deep learning prostate cancer bone metastasis identification system based on whole-body bone scanning image | |
| CN112819747A (en) | Method for automatically diagnosing benign and malignant nodules based on lung tomography image | |
| CN104281856A (en) | Image preprocessing method and system for brain medical image classification | |
| Oliver et al. | A comparison of breast tissue classification techniques | |
| Hao et al. | Vp-detector: A 3d multi-scale dense convolutional neural network for macromolecule localization and classification in cryo-electron tomograms | |
| Behara et al. | Artificial intelligence in medical diagnostics: A review from a South African context | |
| Roy et al. | Automated medical image segmentation: a survey | |
| CN108877927A (en) | A kind of medical image diagnosis method | |
| Cao et al. | 3D convolutional neural networks fusion model for lung nodule detection onclinical CT scans |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |