CN104463215A - Tiny aneurysm occurrence risk prediction system based on retina image processing - Google Patents

Tiny aneurysm occurrence risk prediction system based on retina image processing Download PDF

Info

Publication number
CN104463215A
CN104463215A CN201410757731.0A CN201410757731A CN104463215A CN 104463215 A CN104463215 A CN 104463215A CN 201410757731 A CN201410757731 A CN 201410757731A CN 104463215 A CN104463215 A CN 104463215A
Authority
CN
China
Prior art keywords
image
small
aneurysm
candidate
small aneurysm
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
Application number
CN201410757731.0A
Other languages
Chinese (zh)
Inventor
丁山
马文翼
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Northeastern University China
Original Assignee
Northeastern University China
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Northeastern University China filed Critical Northeastern University China
Priority to CN201410757731.0A priority Critical patent/CN104463215A/en
Publication of CN104463215A publication Critical patent/CN104463215A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
    • G06F18/24155—Bayesian classification
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • 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
    • G06T2207/30041—Eye; Retina; Ophthalmic

Abstract

The invention relates to a tiny aneurysm occurrence risk prediction system based on retina image processing. The system comprises a retina image acquisition unit, a retina image preprocessing unit, a tiny aneurysm image rough extraction unit, a tiny aneurysm image fine extraction unit and a tiny aneurysm occurrence risk prediction unit. Image enhancement processing is conducted on acquired retina images through image enhancement and wavelet transformation, multi-dynamic-template matching is conducted on the enhanced images, tiny aneurysm images are roughly extracted, vascular tissue on the roughly-extracted images is removed, feature values of the tiny aneurysm images are extracted, a naive Bayes classifier is established, and tiny aneurysm occurrence risk prediction is achieved. The weighted naive Bayes classifier is adopted in the system, the sensitivity of tiny aneurysm occurrence risk prediction is improved, and the false drop rate of tiny aneurysms is decreased.

Description

Based on the small aneurysm occurrence risk prognoses system of retinal images process
Technical field
The invention belongs to pattern identification research field, be specifically related to the small aneurysm occurrence risk prognoses system based on retinal images process.
Background technology
Digital fundus image is adopted to carry out the important applied field that the small aneurysmal automatic detection of retina is current pattern-recognition.The people such as Fleming use the method for local standard contrast to achieve small aneurysmal automatic detection (A.D.Fleming in digital fundus image, S.Philip, K.A.Goatman, J.A.Olson, and P.F.Sharp.Automatedmicroaneurysms detection using local contrast normalization and local vessel detection [J], IEEEtransactions on medical imaging, 2006, 25:1223-1232.), eliminate the capillary vessel member not easily distinguished with small aneurysm to a certain extent.The people such as Walter propose a kind of (T.Walter of the detection based on mathematical morphology, P.Massin, A.Erginay, R.Ordonez, C.Jeulin, and J.-C.Klein.Automatic detection of microaneurysms in colorfundus images [J], Medical image analysis, 2007,11 (6): 555-566.), realize bright shadow by image standardization to compensate, to the method that small aneurysmal detection then adopts diameter to approach, make up the deficiency that top cap transfer pair curved blood vessel detects.Niemeijer combines supervision pixel classifier (Supervised Pixel Classification) method and proposes a kind of hybrid detection algorithm (M.Niemeijer, B.van Ginneken, J.Staal, M.S.A.Suttorp-Schulten, and M.D.Abr à moff.Automatic detection of red lesions in digital color fundus photographs [J], IEEE transactions onmedical imaging, 2005, 24 (5): 584-592.), this algorithm can detect all retinal vessels, comprise the lesion information that small aneurysm etc. is similar, by removing the linear structure be sticked together, obtain small aneurysm candidate collection, the method can detect the symptom information such as retinal hemorrhage simultaneously.A kind of method that the people such as Bob then have employed multiple dimensioned related coefficient realizes small aneurysmal automatic detection, form coupling is carried out according to small aneurysmal shape facility, its matching degree adopts related coefficient to weigh, small aneurysm set (Z.Bob is obtained again through feature extraction, W.X.Qian, Y.Jane, et al.Detection ofmicroaneurysms using multi-scale correlation coefficients [J], Pattern recognition, 2010,43:2237-2248.).
Except the small aneurysm based on Mathematical Morphology Method detects automatically, the people such as Sinthanayothin adopt recursive region to grow the detection achieving small aneurysm and blood vessel, and end user's artificial neural networks extracts blood vessel, small aneurysm (the C.Sinthanayothin that will detect is obtained after removing the blood vessel extracted, J.F.Boyce, T.H.Williamson, H.L.Cook, E.Mensah, S.Lal, and D.Usher.Automated detection of diabetic retinopathy on digital fundus images [J], Diabetic medicine, 2002, 19:105-112.).The persons such as Quellec use supervised learning method, after wavelet transformation, template matches is carried out to wavelet sub-band after conversion, Threshold segmentation is carried out to matching result and obtains required small aneurysm set (G.Quellec, M.Lamard, P.M.Josselin, G.Cazuguel, B.Cochener, and C.Roux.Optimal waveletstransform for the detection of microaneurysms in retina photographs [J], IEEE transactions onmedical imaging, 2008, 27 (9): 1230-1241.) based on the difference of stencil-chosen method, common classical template matching algorithm has template base to mate (Template Library, TL) algorithm sum functions template (Function Template, FT) matching algorithm.
The up-to-date coupling extracting method occurred in small aneurysmal first extraction at present, its flow process is simple and easy to use, but its testing result is unsatisfactory, detection accuracy is on the low side and false drop rate is higher, and inadequate to the adaptability of the retinal images under different mode, the retinal images obtained under being not suitable with certain imaging pattern even completely.
Summary of the invention
For the deficiencies in the prior art, the present invention proposes the small aneurysm occurrence risk prognoses system based on retinal images process.
Technical solution of the present invention is as follows:
Based on the small aneurysm occurrence risk prognoses system of retinal images process, comprise retinal images acquiring unit, retinal images pretreatment unit, the thick extraction unit of small aneurysm image, small aneurysm image essence extraction unit and small aneurysm occurrence risk predicting unit;
Described retinal images acquiring unit, opens the amphiblestroid image of known small aneurysmal rgb format and N opens the amphiblestroid image of rgb format to be detected for obtaining N;
Described retinal images pretreatment unit, open the amphiblestroid image of rgb format to be detected carry out image enhancement processing for opening the amphiblestroid image of small aneurysmal rgb format and N to N, carry out three layers of haar wavelet transformation again, the high frequency imaging be enhanced after processing, the image namely after the conversion of 2N Zhang Xiaobo;
The described thick extraction unit of small aneurysm image, for using small aneurysmal retinal images known in the image after wavelet transformation as training set image, using retinal images to be detected in the image after wavelet transformation as test set image, set up k the dynamic multiparameter template of small aneurysm, k the dynamic multiparameter template of small aneurysm is utilized to carry out template matches to training set image and test set image respectively, obtain the image after the image after training set template matches and test set template matches, in image, each pixel forms small aneurysm Candidate Set, candidate's knurl in small aneurysm Candidate Set on image medium vessels in small aneurysm Candidate Set on image medium vessels after extracting training set template matches respectively after candidate's knurl and test set template matches is also rejected, obtain the rejecting blood-vessel image of training set and the rejecting blood-vessel image of test set, the method increased utilizing region is reduced, obtain the image after the image after the growth of training set region and the growth of test set region,
Described small aneurysm image essence extraction unit, feature extraction is carried out for the image after increasing the image after the growth of training set region and test set region, obtain the set of training set eigenwert and the set of test set eigenwert, and set up the Naive Bayes Classifier of weighting according to the set of training set eigenwert;
Described small aneurysm occurrence risk predicting unit, for the set of test set eigenwert being brought into the Naive Bayes Classifier of weighting, when belong to small aneurysmal probability be greater than do not belong to small aneurysmal probability time, this retinal images exist there is small aneurysmal risk.
Described retinal images pretreatment unit, comprising: obtain image background module, shadow correction module, top cap conversion module, gaussian filtering module and Wavelet transformation module;
Described acquisition image background module, carrying out negate for 2N being opened the image of the amphiblestroid image of rgb format under green channel, obtaining the image that 2N opens film result;
Described shadow correction module, carries out shadow correction for image 2N being opened to film result, obtains 2N and opens the image removing shade;
Described top cap conversion module, carries out top cap conversion for opening the image removing shade to 2N, obtains 2N and open the image after pushing up cap conversion;
Described gaussian filtering module, carries out gaussian filtering for the image after 2N being opened to top cap conversion, obtains 2N and open the image after strengthening process;
Described Wavelet transformation module, carries out three layers of haar wavelet transformation for opening the image after strengthening process to 2N, obtains 2N and opens the high frequency imaging strengthening the rear image of process, the image namely after the conversion of 2N Zhang Xiaobo.
The described thick extraction unit of small aneurysm image, comprises template matches module, blood vessel removes module and region increases module;
Described template matches module, for using small aneurysmal retinal images known in the image after wavelet transformation as training set image, using retinal images to be detected in the image after wavelet transformation as test set image, set up k the dynamic multiparameter template of small aneurysm, utilize k the dynamic multiparameter template of small aneurysm to carry out template matches to training set image and test set image respectively, obtain the image after the image after training set template matches and test set template matches;
Many dynamic parameters of described small aneurysm dynamic multiparameter template comprise: radius parameter, steepness parameter, grey level face parameter and gray scale height parameter;
Described radius parameter utilizes enumerative technique to set;
Described steepness parameter utilizes the average error summation ASE of candidate's knurl in all small aneurysm Candidate Sets in training set image to calculate;
Described grey level face parameter is the background intensity values of candidate's knurl in small aneurysm Candidate Set to be matched in the image after wavelet transformation;
Described gray scale height parameter is the versus grayscale height of the central point of candidate's knurl in small aneurysm Candidate Set to be matched in the image after wavelet transformation.
The image of described training set template matches is the image after training set related coefficient image after Threshold segmentation and error and image carry out logical and; The image of described test set template matches is the image after test set related coefficient image after Threshold segmentation and error and image carry out logical and.
The related coefficient image of described training set is the image that each pixel maximum with each pixel related coefficient in small aneurysm dynamic multiparameter template in training set image is formed;
The related coefficient image of described test set is the image that each pixel maximum with each pixel related coefficient in small aneurysm dynamic multiparameter template in test set image is formed;
The error of described training set and image are the image formed with each pixel error in small aneurysm dynamic multiparameter template and maximum each pixel in training set image;
The error of described test set and image are the image formed with each pixel error in small aneurysm dynamic multiparameter template and maximum each pixel in test set image.
Described blood vessel removes module, for adopting the method for matched filter to training set image and test set image, blood vessel in image is rejected, obtain training set blood-vessel image and test set blood-vessel image, candidate's knurl in small aneurysm Candidate Set on image medium vessels after utilizing training set blood-vessel image to extract training set template matches is also rejected, and obtains the rejecting blood-vessel image of training set; Candidate's knurl in small aneurysm Candidate Set on image medium vessels after utilizing test set blood-vessel image to extract test set template matches is also rejected, and obtains the rejecting blood-vessel image of test set;
Described region increases module, reduces for the method utilizing region to increase the rejecting blood-vessel image of training set, obtains the image after the growth of training set region; The method that the rejecting blood-vessel image of test set utilizes region to increase is reduced, obtains the image after the growth of test set region.
Described small aneurysm image essence extraction unit, comprising: characteristic extracting module, feature weight coefficient training module, structure classifier modules;
Described characteristic extracting module, for 48 eigenwerts of candidate's knurl in each small aneurysm Candidate Set in the image after the growth of calculation training collection region, obtains the set of training set eigenwert; Calculate 48 eigenwerts of candidate's knurl in each small aneurysm Candidate Set in the image after the growth of test set region, obtain the set of test set eigenwert;
In described small aneurysm Candidate Set, 48 eigenwerts of candidate's knurl comprise the pixel ratio in the pixel ratio in small aneurysmal shape facility information, small aneurysmal pixel characteristic information, small aneurysmal matching characteristic information, small aneurysm region and its minimal convex polygon, small aneurysm region and its minimum boundary rectangle.
Described feature weight coefficient training module, for the weight coefficient utilizing genetic algorithm to obtain 48 eigenwerts of the set of training set eigenwert;
Described structure classifier modules, the set for calculation training collection eigenwert belongs to average and the variance of each property value of candidate's knurl in small aneurysmal small aneurysm Candidate Set; Average and the variance of each property value of candidate's knurl in small aneurysmal small aneurysm Candidate Set is not belonged in the set of calculation training collection eigenwert; The Naive Bayes Classifier that the average that do not belong to each property value of candidate's knurl in small aneurysmal small aneurysm Candidate Set in the set of the average of each property value of candidate's knurl in small aneurysmal small aneurysm Candidate Set and variance, training set eigenwert and variance set up weighting is belonged in set according to the weight coefficient of 48 eigenwerts of the set of training set eigenwert, training set eigenwert.
Beneficial effect of the present invention:
The present invention is applicable to the template matches that in retinal images, small aneurysm is tentatively extracted, and design achieves dynamic multiparameter template DMPT, this template can adapt to the most of metamorphosis of small aneurysm in gamma characteristic, and the coupling of this template is carried out in wavelet field with outstanding feature in one direction.In addition, the present invention combines the contradictory relation between coupling correlativity and matching error in the design of matching degree, employs the strategy that a kind of error and SE and related coefficient CC restrict matching degree jointly.At final small aneurysm positioning stage, then take the Bayes classifier based on mathematical statistics.
The present invention adopts the dynamic masterplate of multiparameter carried out in wavelet field, can adapt to different gray scale, size, background, sharpness and show different small aneurysm on the low frequency, high frequency aliquot of image, enhancing the extraction accuracy in thick leaching process; The present invention adopts the Naive Bayes Classifier of weighting, improves the sensitivity of small aneurysm occurrence risk prediction, reduces small aneurysmal false drop rate.
Accompanying drawing explanation
Fig. 1 is the structured flowchart of the small aneurysm occurrence risk prognoses system based on retinal images process of the specific embodiment of the invention;
Fig. 2 is the amphiblestroid image of known small aneurysmal rgb format of the specific embodiment of the invention;
Fig. 3 is the image after the top cap conversion of the specific embodiment of the invention;
Fig. 4 is the image after the enhancing process of the specific embodiment of the invention;
Fig. 5 is the image after the ground floor haar wavelet transformation of the specific embodiment of the invention;
Wherein, (a) is the low frequency signal image after ground floor haar wavelet transformation; B () is the high-frequency signal image after ground floor haar wavelet transformation;
Fig. 6 is the image after the second layer haar wavelet transformation of the specific embodiment of the invention;
Wherein, (a) is the low frequency signal image after second layer haar wavelet transformation; B () is the high-frequency signal image after second layer haar wavelet transformation;
Fig. 7 is the image after the third layer haar wavelet transformation of the specific embodiment of the invention;
Wherein, (a) is the low frequency signal image after third layer haar wavelet transformation; B () is the high-frequency signal image (image after wavelet transformation) after third layer haar wavelet transformation;
Fig. 8 is the image after the template matches of the specific embodiment of the invention;
Fig. 9 is the blood-vessel image of the specific embodiment of the invention;
Figure 10 is the image after the region of the specific embodiment of the invention increases;
Figure 11 is the small aneurysm image finally extracted of the specific embodiment of the invention;
Figure 12 be the specific embodiment of the invention based on the small aneurysm detection method of wavelet transformation and Bayes's weighting classification device and the algorithm effect comparison diagram of people such as score tactful effect and Bob etc.
Embodiment
Below in conjunction with accompanying drawing to specific embodiment of the invention detailed description in addition.
Based on the small aneurysm occurrence risk prognoses system of retinal images process, as shown in Figure 1, retinal images acquiring unit, retinal images pretreatment unit, the thick extraction unit of small aneurysm image, small aneurysm image essence extraction unit and small aneurysm occurrence risk predicting unit is comprised.
Retinal images acquiring unit, for obtaining 50 amphiblestroid image I of known small aneurysmal rgb format 1i 50with 50 amphiblestroid image I of rgb format to be detected 51i 100.
The amphiblestroid image of known small aneurysmal rgb format as shown in Figure 2.
Retinal images pretreatment unit, for 50 amphiblestroid image I of known small aneurysmal rgb format 1i 50with 50 amphiblestroid image I of rgb format to be detected 51i 100carry out image enhancement processing, then carry out three layers of haar wavelet transformation, the high frequency imaging after the process that is enhanced, the image I namely after the conversion of 2N Zhang Xiaobo hr1i hr100.
Retinal images pretreatment unit, comprising: obtain image background module, shadow correction module, top cap conversion module, gaussian filtering module and Wavelet transformation module.
Obtain image background module, for by 100 the image Is of the amphiblestroid image of rgb format under green channel g1i g100carry out negate, obtain the image I of 100 film results bg1i bg100.
Shadow correction module, for the image I to 100 film results bg1i bg100carry out shadow correction, revise the image shade because uneven illumination produces, obtain the image I that 100 are removed shade sc1i sc10.
Top cap conversion module, for removing the image I of shade to 100 sc1i sc100carry out top cap conversion, remove the linear structure obviously existed in image, as the trunk in retina, obtain the image I after 100 top cap conversion th1i th100.
In present embodiment, choosing the anglec of rotation is that the linear structure element of β is to the image I removing shade sc1i sc100carry out top cap conversion, when structural element rotates, each angle rotated is set to 15 °, and the angle direction β value namely chosen is as shown in formula (1):
{0°，15°，30°，45°，60°，75°，90°，105°，120°，135°，150°，165°} (1)
The structural element of above-mentioned different rotary angle is used to remove the image I of shade sc1i scc100carry out morphology opening operation, if each image after opening operation has 12 respectively.Select the maximal value of each pixel corresponding point in 12 images to be pixel in opening operation image, obtain opening operation image, remove the image I of shade sc1i scc100deduct the opening operation image of its correspondence, namely obtain the image I after pushing up cap conversion th1i th100.
Image after the top cap conversion of the specific embodiment of the invention as shown in Figure 3.
Gaussian filtering module, for the image I pushed up after caps conversion 100 th1i th100carry out gaussian filtering, strengthen the contrast of candidate's knurl target area in small aneurysm Candidate Set, obtain 100 and strengthen the image I after processing gt1i gt100.
To the image I after the cap conversion of top th1i th100carry out the contrast that gaussian filtering is used for strengthening candidate's knurl target area in small aneurysm Candidate Set.
In present embodiment, select size to be 11 × 11, the 2-d gaussian filters device of intensive parameter σ=1 is to the image I after the cap conversion of top th1i th100carry out boostfiltering, the image I after the process that is enhanced gt1i gt100.
Image after the enhancing process of the specific embodiment of the invention as shown in Figure 4.
Wavelet transformation module, for the image I strengthened after process 100 gt1i gt100carry out three layers of haar wavelet transformation, obtain 100 and strengthen the image I after processing gt1i gt100high frequency imaging, namely 100 Zhang Xiaobos conversion after image I hr1i hr100.
As shown in Figure 5, wherein, (a) is the low frequency signal image after ground floor haar wavelet transformation to image after the ground floor haar wavelet transformation of the specific embodiment of the invention; B () is the high-frequency signal image after ground floor haar wavelet transformation.
As shown in Figure 6, wherein, (a) is the low frequency signal image after second layer haar wavelet transformation to image after the second layer haar wavelet transformation of the specific embodiment of the invention; B () is the high-frequency signal image after second layer haar wavelet transformation.
As shown in Figure 7, wherein, wherein, (a) is the low frequency signal image after third layer haar wavelet transformation to image after the third layer haar wavelet transformation of the specific embodiment of the invention; B () is the high-frequency signal image (image after wavelet transformation) after third layer haar wavelet transformation.
The thick extraction unit of small aneurysm image, for by the image I after wavelet transformation hr1i hr100in known small aneurysmal retinal images as training set image I hr1i hr50, by the image I after wavelet transformation hr1i hr100in retinal images to be detected as test set image I hr51i hr100, set up k the dynamic multiparameter template of small aneurysm, utilize k the dynamic multiparameter template of small aneurysm respectively to training set image I hr1i hr50with test set image I hr51i hr100carry out template matches, obtain the image I after training set template matches bv1i bv50with the image I after test set template matches bv51i bv100; In image, each pixel forms small aneurysm Candidate Set, extracts the image I after training set template matches respectively bv1i bv50image I in small aneurysm Candidate Set on medium vessels after candidate's knurl and test set template matches bv51i bv100candidate's knurl in small aneurysm Candidate Set on medium vessels is also rejected, and obtains the rejecting blood-vessel image I of training set tb1i tb50with the rejecting blood-vessel image I of test set tb51i tb100, the method increased utilizing region is reduced, and obtains the image I after the growth of training set region re1i re50image I after increasing with test set region re51i re100.
The thick extraction unit of small aneurysm image, comprises template matches module, blood vessel removes module and region increases module.
Template matches module, for by the image I after wavelet transformation hr1i hr100in known small aneurysmal retinal images as training set image I hr1i hr50, by the image I after wavelet transformation hr1i hr100in retinal images to be detected as test set image I hr51i hr100, set up k the dynamic multiparameter template of small aneurysm, utilize k the dynamic multiparameter template of small aneurysm to carry out template matches to training set image and test set image respectively, obtain the image I after training set template matches bv1i bv50with the image I after test set template matches bv51i bv100.
By the image I after wavelet transformation hr1i hr100in known small aneurysmal retinal images as training set image I hr1i hr50, by the image I after wavelet transformation hr1i hr100in retinal images to be detected as test set image I hr51i hr100.
Set up k=4 the dynamic multiparameter template of small aneurysm.
Radius parameter utilizes enumerative technique to set, radius parameter value set R v{ r 1r 4, radius parameter r 1r 4∈ [3,10] v ∈ (1 ... 4), in present embodiment, the value set of the radius parameter utilizing enumerative technique to set is as R v={ 3,5,7,10}.
Steepness parameter utilizes the average error summation ASE of candidate's knurl in all small aneurysm Candidate Sets in training set image to calculate: the initial value s of setting steepness parameter 0with incremental steps s p, utilize training set image I hr1i hr50in in all small aneurysm Candidate Sets the average error summation ASE of candidate's knurl as compromise function, the ratio that this average error summation ASE is reduced is less than the scale parameter q% of setting as judge condition, calculates steepness parameter s.
In present embodiment, the initial value s of the steepness parameter of setting 0be 0.5, step parameter is s p=0.5, the scale parameter of setting is 20%, and the steepness parameter s calculated is 3.
Grey level face parameter l is the image I after wavelet transformation hr1i hr100in the background intensity values b of candidate's knurl in small aneurysm Candidate Set to be matched.
Grey level face parameter l is as shown in formula (2):
l = b = Σ I hrM ( x , y , r v + ω ) - I hrM ( x , y , r v ) ( r v + ω ) - r v 2 - - - ( 2 )
Wherein, the value of b is expressed as the image I after surrounding wavelet transformation hr1i hr1000in candidate knurl I in small aneurysm Candidate Set to be matched hrM(x, y, r v) width be w annular region in the average gray value of pixel.
Gray scale height parameter h is the image I after wavelet transformation hr1i hr100in the versus grayscale height of the central point of candidate's knurl in small aneurysm Candidate Set to be matched.
Gray scale height parameter h is as shown in formula (3):
According to radius parameter r v, steepness parameter s, grey level face parameter l and gray scale height parameter h, the k of a foundation dynamic multiparameter template of small aneurysm be wherein, e is the truth of a matter of natural logarithm, d be pixel in dynamic multiparameter template (x ', y ') with dynamic multiparameter template center point (x ' 0, y ' 0) between distance.
Obtain the pixel in 4 dynamic multiparameter templates of small aneurysm and training set image I hr1i hr50in small aneurysm Candidate Set in the related coefficient of pixel of candidate's knurl and error and, obtain the pixel in 4 dynamic multiparameter templates of small aneurysm and test set image I hr51i hr100in small aneurysm Candidate Set in the related coefficient of pixel of candidate's knurl and error and.
In small aneurysm dynamic multiparameter template each pixel 4 related coefficients in maximum related coefficient as the matching degree of this pixel, using the matching degree of this pixel as pixel, obtain the related coefficient image I of training set cc1i cc50with the related coefficient image I of test set cc51i cc100;
In small aneurysm dynamic multiparameter template each pixel 4 errors and in minimum error and the matching degree as this pixel, using the matching degree of this pixel as pixel, obtain error and the image I of training set se1i se50with error and the image I of test set se51i se100;
By the related coefficient image I of training set cc1i cc50middle related coefficient minimum value as correlation coefficient threshold, to the related coefficient image I of training set cc1i cc50with the related coefficient image I of test set cc51i cc100carry out correlation coefficient threshold segmentation, obtain the training set related coefficient image I after splitting ccd1i ccd50with the test set related coefficient image I after segmentation ccd51i ccd100
The error of training set and matching degree image I se1i se50medial error and maximal value as error and threshold value, to error and the image I of training set se1i se50with error and the image I of test set se51i se100carry out error and Threshold segmentation, obtain the training set error after splitting and image I sed1i sed50with the test set error after segmentation and image I sed51i sed100
By the training set related coefficient image I after segmentation ccd1i ccd50with the training set error after segmentation and image I sed1i sed50carry out logical and calculating, obtain the image I after training set template matches tm1i tm50, by the test set related coefficient image I after segmentation ccd51i ccd100with the test set error after segmentation and image I sed51i sed100carry out logical and calculating, obtain the image I after test set template matches tm51i tm100.
Image after the template matches of the specific embodiment of the invention as shown in Figure 8.
Blood vessel removes module, for training set image I hr1i hr50with test set image I hr51i hr100adopt the method for matched filter, the blood vessel in image is rejected, obtains training set blood-vessel image I bv1i bv50with test set blood-vessel image I bv51i bv100, utilize training set blood-vessel image I bv1i bv50extract the image I after training set template matches tm1i tm50candidate's knurl in small aneurysm Candidate Set on medium vessels is also rejected, and obtains the rejecting blood-vessel image I of training set tb1i tb50; Utilize test set blood-vessel image I bv51i bv100extract the image I after test set template matches tm51i tm100candidate's knurl in small aneurysm Candidate Set on medium vessels is also rejected, and obtains the rejecting blood-vessel image I of test set tb51i tb100.
In present embodiment, the method for matched filter is adopted to reject the blood vessel in image, first calculation training collection image I hr1i hr50with test set image I hr51i hr100gauss matched filtering response and the response of Gauss matched filtering device first order derivative, automatically blood vessel segmentation threshold value is regulated according to the response of image to Gauss matched filtering device first order derivative, re-use the segmentation that this adaptive threshold realizes Gauss matched filtering response image, thus amphiblestroid blood vessel structure and class frontier district are separated.
The blood-vessel image of the specific embodiment of the invention as shown in Figure 9.
Region increases module, for the rejecting blood-vessel image I by training set tb1i tb50the method utilizing region to increase is reduced, and obtains the image I after the growth of training set region re1i re50; By the rejecting blood-vessel image I of test set tb51i tb100the method utilizing region to increase is reduced, and obtains the image I after the growth of test set region re51i re100.
Image after the region growth of the specific embodiment of the invention as shown in Figure 10.
Small aneurysm image essence extraction unit, for the image I after increasing training set region re1i re50image I after increasing with test set region re51i re100carry out feature extraction, obtain the set A of training set eigenwert zcm(A zc1, A zc2a zc48) and the set a of test set eigenwert yhn(a yh1, a yh2a yh48), and according to the set A of training set eigenwert zcm(A zc1, A zc2a zc48) set up the Naive Bayes Classifier of weighting, wherein z ∈ 1 ... 50, c ∈ 1 ... E z, m ∈ 1 ... 48, E zfor the image I after the growth of training set region rezin the number of candidate's knurl in small aneurysm Candidate Set, y ∈ 51 ... 100, h ∈ 1 ... H y, n ∈ 1 ... 48, H yfor the image I after the growth of test set region reyin the number of candidate's knurl in small aneurysm Candidate Set.
In present embodiment, E zbe 24189, H ybe 23704.
Small aneurysm image essence extraction unit, comprising: characteristic extracting module, feature weight coefficient training module, structure classifier modules;
Characteristic extracting module, the image I after increasing for calculation training collection region re1i re50in 48 eigenwerts of candidate's knurl in each small aneurysm Candidate Set, obtain the set A of training set eigenwert zcm(A zc1, A zc2a zc48); Calculate the image I after the growth of test set region re51i re100in 48 eigenwerts of candidate's knurl in each small aneurysm Candidate Set, obtain the set a of test set eigenwert yhn(a yh1, a yh2a yh48).
In each small aneurysm Candidate Set, 48 eigenwerts of candidate's knurl comprise the pixel ratio in pixel ratio in small aneurysmal shape facility information (as shown in table 1), small aneurysmal pixel characteristic information (as shown in table 2), small aneurysmal matching characteristic information (as shown in table 3), small aneurysm region and its minimal convex polygon and small aneurysm region and its minimum boundary rectangle
Comprise the pixel ratio (as shown in table 4) in the pixel ratio in the matching characteristic information (as shown in table 3) of candidate's knurl in the pixel characteristic information (as shown in table 2) of candidate's knurl in the shape facility information (as shown in table 1) of candidate's knurl in small aneurysm Candidate Set, small aneurysm Candidate Set, small aneurysm Candidate Set, small aneurysm region and its minimal convex polygon, small aneurysm region and its minimum boundary rectangle:
The small aneurysmal shape facility information of table 1
The small aneurysmal pixel characteristic information of table 2
The matching characteristic information of candidate's knurl in the small aneurysm Candidate Set of table 3
The further feature information of candidate's knurl in the small aneurysm Candidate Set of table 4
Feature weight coefficient training module, for the set A utilizing genetic algorithm to obtain training set eigenwert zcm(A zc1, A zc2a zc48) in the weight coefficient ω of 48 eigenwerts of candidate's knurl in small aneurysm Candidate Set m.
In present embodiment, obtain the set A of training set eigenwert zcm(A zc1, A zc2a zc48) in the weight coefficient ω of 48 eigenwerts of candidate's knurl in small aneurysm Candidate Set mas shown in table 5:
The weight coefficient ω of 48 eigenwerts of the set of table 5 training set eigenwert m
Build classifier modules, for the set A of calculation training collection eigenwert zcm(A zc1, A zc2a zc48) in belong to the average μ of each property value of candidate's knurl in small aneurysmal small aneurysm Candidate Set 1and variance the set A of calculation training collection eigenwert zcm(A zc1, A zc2a zc48) in do not belong to the average μ of each property value of candidate's knurl in small aneurysmal small aneurysm Candidate Set 2and variance wherein, C 1represent that in small aneurysm Candidate Set, candidate's knurl is small aneurysm, C 2represent that in this small aneurysm Candidate Set, candidate's knurl is not small aneurysm.
According to the set A of training set eigenwert zcm(A zc1, A zc2a zc48) in the weight coefficient ω of 48 eigenwerts of candidate's knurl in small aneurysm Candidate Set m, calculation training collection eigenwert set A zcm(A zc1, A zc2a zc48) in belong to the average μ of each property value of candidate's knurl in small aneurysmal small aneurysm Candidate Set 1and variance the set A of training set eigenwert zcm(S zc1, A zc2a zc48) in do not belong to the average μ of each property value of candidate's knurl in small aneurysmal small aneurysm Candidate Set 2and variance set up the Naive Bayes Classifier of weighting.
In present embodiment, the Naive Bayes Classifier of weighting is as shown in formula (4):
C nb ( C i ) X = arg max C i ∈ C P ( C i ) ΠP ( X | C i ) ω m - - - ( 4 )
Wherein, C i∈ (C 1, C 2);
P(C i)＝E/E Z；
The number that E waits for small aneurysm in the image after the growth of training set region;
E zfor the number of candidate's knurl in small aneurysm Candidate Set in the image after the growth of training set region;
P ( X | C i ) = g ( X , μ i , σ C i ) = 1 2 πσ C i e - ( x - μ i ) 2 2 σ C i 2 ;
i＝1，2。
Small aneurysm occurrence risk predicting unit, for the set a by test set eigenwert yhn(a yh1, a yh2a yh48) weighted input Naive Bayes Classifier in, when test set region increase after image I re51i re100in in a certain small aneurysm Candidate Set candidate's knurl in the Naive Bayes Classifier of weighting, belong to small aneurysmal probability be greater than when not belonging to small aneurysmal probability, this retinal images exists and has small aneurysmal risk.
In present embodiment, by the set a of test set eigenwert yhn(a yh1, a yh2a yh48) Naive Bayes Classifier of weighted input is as shown in formula (5):
C nb ( C i ) a yhn = arg max C i ∈ C P ( C i ) ΠP ( a yhn | C i ) ω m - - - ( 5 )
Image I after test set region increases re51i re100in the Naive Bayes Classification probability of the weighting of candidate's knurl in a certain small aneurysm Candidate Set then this retinal images exists and has small aneurysmal risk.
The small aneurysm image finally extracted of the specific embodiment of the invention as shown in figure 11.
As shown in figure 12, what can find out this invention thus is highly effective for small aneurysmal occurrence risk prediction for the small aneurysm occurrence risk prognoses system based on retinal images process of the specific embodiment of the invention and the algorithm effect comparison diagram of people such as score tactful effect and Bob etc.

Claims (8)

1. based on the small aneurysm occurrence risk prognoses system of retinal images process, it is characterized in that, comprise retinal images acquiring unit, retinal images pretreatment unit, the thick extraction unit of small aneurysm image, small aneurysm image essence extraction unit and small aneurysm occurrence risk predicting unit;
Described retinal images acquiring unit, opens the amphiblestroid image of known small aneurysmal rgb format and N opens the amphiblestroid image of rgb format to be detected for obtaining N;
Described retinal images pretreatment unit, open the amphiblestroid image of rgb format to be detected carry out image enhancement processing for opening the amphiblestroid image of known small aneurysmal rgb format and N to N, carry out three layers of haar wavelet transformation again, the high frequency imaging be enhanced after processing, the image namely after the conversion of 2N Zhang Xiaobo;
The described thick extraction unit of small aneurysm image, for using small aneurysmal retinal images known in the image after wavelet transformation as training set image, using retinal images to be detected in the image after wavelet transformation as test set image, set up k the dynamic multiparameter template of small aneurysm, k the dynamic multiparameter template of small aneurysm is utilized to carry out template matches to training set image and test set image respectively, obtain the image after the image after training set template matches and test set template matches, in image, each pixel forms small aneurysm Candidate Set, candidate's knurl in small aneurysm Candidate Set on image medium vessels in small aneurysm Candidate Set on image medium vessels after extracting training set template matches respectively after candidate's knurl and test set template matches is also rejected, obtain the rejecting blood-vessel image of training set and the rejecting blood-vessel image of test set, the method increased utilizing region is reduced, obtain the image after the image after the growth of training set region and the growth of test set region,
Described small aneurysm image essence extraction unit, feature extraction is carried out for the image after increasing the image after the growth of training set region and test set region, obtain the set of training set eigenwert and the set of test set eigenwert, and set up the Naive Bayes Classifier of weighting according to the set of training set eigenwert;
Described small aneurysm occurrence risk predicting unit, for the set of test set eigenwert being brought into the Naive Bayes Classifier of weighting, when belong to small aneurysmal probability be greater than do not belong to small aneurysmal probability time, this retinal images exist there is small aneurysmal risk.
2. the small aneurysm occurrence risk prognoses system based on retinal images process according to claim 1, it is characterized in that, described retinal images pretreatment unit, comprising: obtain image background module, shadow correction module, top cap conversion module, gaussian filtering module and Wavelet transformation module;
Described acquisition image background module, carrying out negate for 2N being opened the image of the amphiblestroid image of rgb format under green channel, obtaining the image that 2N opens film result;
Described shadow correction module, carries out shadow correction for image 2N being opened to film result, obtains 2N and opens the image removing shade;
Described top cap conversion module, carries out top cap conversion for opening the image removing shade to 2N, obtains 2N and open the image after pushing up cap conversion;
Described gaussian filtering module, carries out gaussian filtering for the image after 2N being opened to top cap conversion, obtains 2N and open the image after strengthening process;
Described Wavelet transformation module, carries out three layers of haar wavelet transformation for opening the image after strengthening process to 2N, obtains 2N and opens the high frequency imaging strengthening the rear image of process, the image namely after the conversion of 2N Zhang Xiaobo.
3. the small aneurysm occurrence risk prognoses system based on retinal images process according to claim 1, is characterized in that, the described thick extraction unit of small aneurysm image, comprises template matches module, blood vessel removes module and region increases module;
Described template matches module, for using small aneurysmal retinal images known in the image after wavelet transformation as training set image, using retinal images to be detected in the image after wavelet transformation as test set image, set up k the dynamic multiparameter template of small aneurysm, utilize k the dynamic multiparameter template of small aneurysm to carry out template matches to training set image and test set image respectively, obtain the image after the image after training set template matches and test set template matches;
Described blood vessel removes module, for adopting the method for matched filter to training set image and test set image, blood vessel in image is rejected, obtain training set blood-vessel image and test set blood-vessel image, candidate's knurl in small aneurysm Candidate Set on image medium vessels after utilizing training set blood-vessel image to extract training set template matches is also rejected, and obtains the rejecting blood-vessel image of training set; Candidate's knurl in small aneurysm Candidate Set on image medium vessels after utilizing test set blood-vessel image to extract test set template matches is also rejected, and obtains the rejecting blood-vessel image of test set;
Described region increases module, reduces for the method utilizing region to increase the rejecting blood-vessel image of training set, obtains the image after the growth of training set region; The method that the rejecting blood-vessel image of test set utilizes region to increase is reduced, obtains the image after the growth of test set region.
4. the small aneurysm occurrence risk prognoses system based on retinal images process according to claim 1, it is characterized in that, described small aneurysm image essence extraction unit, comprising: characteristic extracting module, feature weight coefficient training module, structure classifier modules;
In described small aneurysm Candidate Set, 48 eigenwerts of candidate's knurl comprise the pixel ratio in the pixel ratio in small aneurysmal shape facility information, small aneurysmal pixel characteristic information, small aneurysmal matching characteristic information, small aneurysm region and its minimal convex polygon, small aneurysm region and its minimum boundary rectangle;
Described feature weight coefficient training module, for the weight coefficient utilizing genetic algorithm to obtain 48 eigenwerts of the set of training set eigenwert;
Described structure classifier modules, the set for calculation training collection eigenwert belongs to average and the variance of each property value of small aneurysmal small aneurysm candidate point; Average and the variance of each property value of candidate's knurl in small aneurysmal small aneurysm Candidate Set is not belonged in the set of calculation training collection eigenwert; The Naive Bayes Classifier that the average that do not belong to each property value of candidate's knurl in small aneurysmal small aneurysm Candidate Set in the set of the average of each property value of candidate's knurl in small aneurysmal small aneurysm Candidate Set and variance, training set eigenwert and variance set up weighting is belonged in set according to the weight coefficient of 48 eigenwerts of the set of training set eigenwert, training set eigenwert.
5. the small aneurysm occurrence risk prognoses system based on retinal images process according to claim 3, it is characterized in that, many dynamic parameters of described small aneurysm dynamic multiparameter template comprise: radius parameter, steepness parameter, grey level face parameter and gray scale height parameter;
Described radius parameter utilizes enumerative technique to set;
Described steepness parameter utilizes the average error summation ASE of candidate's knurl in all small aneurysm Candidate Sets in training set image to calculate;
Described grey level face parameter is the background intensity values of candidate's knurl in small aneurysm Candidate Set to be matched in the image after wavelet transformation;
Described gray scale height parameter is the versus grayscale height of the central point of candidate's knurl in small aneurysm Candidate Set to be matched in the image after wavelet transformation.
6. the small aneurysm occurrence risk prognoses system based on retinal images process according to claim 3, it is characterized in that, the image of described training set template matches is the image after training set related coefficient image after Threshold segmentation and error and image carry out logical and; The image of described test set template matches is the image after test set related coefficient image after Threshold segmentation and error and image carry out logical and.
7. the small aneurysm occurrence risk prognoses system based on retinal images process according to claim 4, it is characterized in that, in described small aneurysm Candidate Set, 48 eigenwerts of candidate's knurl comprise the pixel ratio in pixel ratio in small aneurysmal shape facility information, small aneurysmal pixel characteristic information, small aneurysmal matching characteristic information and small aneurysm region and its minimal convex polygon and small aneurysm region and its minimum boundary rectangle.
8. the small aneurysm occurrence risk prognoses system based on retinal images process according to claim 7, it is characterized in that, the related coefficient image of described training set is the image that each pixel maximum with each pixel related coefficient in small aneurysm dynamic multiparameter template in training set image is formed;
The related coefficient image of described test set is the image that each pixel maximum with each pixel related coefficient in small aneurysm dynamic multiparameter template in test set image is formed;
The error of described training set and image are the image formed with each pixel error in small aneurysm dynamic multiparameter template and maximum each pixel in training set image;
The error of described test set and image are the image formed with each pixel error in small aneurysm dynamic multiparameter template and maximum each pixel in test set image.
CN201410757731.0A 2014-12-10 2014-12-10 Tiny aneurysm occurrence risk prediction system based on retina image processing Pending CN104463215A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201410757731.0A CN104463215A (en) 2014-12-10 2014-12-10 Tiny aneurysm occurrence risk prediction system based on retina image processing

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201410757731.0A CN104463215A (en) 2014-12-10 2014-12-10 Tiny aneurysm occurrence risk prediction system based on retina image processing

Publications (1)

Publication Number Publication Date
CN104463215A true CN104463215A (en) 2015-03-25

Family

ID=52909229

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201410757731.0A Pending CN104463215A (en) 2014-12-10 2014-12-10 Tiny aneurysm occurrence risk prediction system based on retina image processing

Country Status (1)

Country Link
CN (1) CN104463215A (en)

Cited By (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105844379A (en) * 2015-12-09 2016-08-10 东软集团股份有限公司 Vehicle risk assessment method and vehicle risk assessment device based on principal component analysis
CN108230322A (en) * 2018-01-28 2018-06-29 浙江大学 A kind of eyeground feature detection device based on weak sample labeling
CN109003279A (en) * 2018-07-06 2018-12-14 东北大学 Fundus retina blood vessel segmentation method and system based on K-Means clustering labeling and naive Bayes model
CN109523153A (en) * 2018-11-12 2019-03-26 平安科技（深圳）有限公司 Acquisition methods, device, computer equipment and the storage medium of illegal fund collection enterprise
CN110120027A (en) * 2019-04-16 2019-08-13 华南理工大学 The CT sectioning image Enhancement Method and device of Machine oriented learning system data
CN111415321A (en) * 2018-12-19 2020-07-14 上海联影医疗科技有限公司 Aneurysm rupture risk detection device and equipment
WO2021224726A1 (en) * 2020-05-05 2021-11-11 International Business Machines Corporation Real-time detection and correction of shadowing in hyperspectral retinal images
CN114494196A (en) * 2022-01-26 2022-05-13 南通大学 Retina diabetic depth network detection method based on genetic fuzzy tree
CN116823833A (en) * 2023-08-30 2023-09-29 山东科技大学 Method, system and equipment for detecting intracranial aneurysm by using omnibearing MIP image

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103458772A (en) * 2011-04-07 2013-12-18 香港中文大学 Method and device for retinal image analysis
WO2014186838A1 (en) * 2013-05-19 2014-11-27 Commonwealth Scientific And Industrial Research Organisation A system and method for remote medical diagnosis

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103458772A (en) * 2011-04-07 2013-12-18 香港中文大学 Method and device for retinal image analysis
WO2014186838A1 (en) * 2013-05-19 2014-11-27 Commonwealth Scientific And Industrial Research Organisation A system and method for remote medical diagnosis

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
SHAN DING等: "An Accurate Approach for Microaneurysm Detection in Digital Fundus Images", 《PATTERN RECOGNITION(ICPR),2014 22ND INTERNATIONAL CONFERENCE ON. IEEE》 *

Cited By (16)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105844379A (en) * 2015-12-09 2016-08-10 东软集团股份有限公司 Vehicle risk assessment method and vehicle risk assessment device based on principal component analysis
CN108230322A (en) * 2018-01-28 2018-06-29 浙江大学 A kind of eyeground feature detection device based on weak sample labeling
CN108230322B (en) * 2018-01-28 2021-11-09 浙江大学 Eye ground characteristic detection device based on weak sample mark
CN109003279B (en) * 2018-07-06 2022-05-13 东北大学 Fundus retina blood vessel segmentation method and system based on K-Means clustering labeling and naive Bayes model
CN109003279A (en) * 2018-07-06 2018-12-14 东北大学 Fundus retina blood vessel segmentation method and system based on K-Means clustering labeling and naive Bayes model
CN109523153A (en) * 2018-11-12 2019-03-26 平安科技（深圳）有限公司 Acquisition methods, device, computer equipment and the storage medium of illegal fund collection enterprise
CN111415321A (en) * 2018-12-19 2020-07-14 上海联影医疗科技有限公司 Aneurysm rupture risk detection device and equipment
CN111415321B (en) * 2018-12-19 2024-02-27 上海联影医疗科技股份有限公司 Aneurysm rupture risk detection device and equipment
CN110120027A (en) * 2019-04-16 2019-08-13 华南理工大学 The CT sectioning image Enhancement Method and device of Machine oriented learning system data
US11200670B2 (en) 2020-05-05 2021-12-14 International Business Machines Corporation Real-time detection and correction of shadowing in hyperspectral retinal images
GB2610743A (en) * 2020-05-05 2023-03-15 Ibm Real-time detection and correction of shadowing in hyperspectral retinal images
WO2021224726A1 (en) * 2020-05-05 2021-11-11 International Business Machines Corporation Real-time detection and correction of shadowing in hyperspectral retinal images
CN114494196A (en) * 2022-01-26 2022-05-13 南通大学 Retina diabetic depth network detection method based on genetic fuzzy tree
CN114494196B (en) * 2022-01-26 2023-11-17 南通大学 Retinal diabetes mellitus depth network detection method based on genetic fuzzy tree
CN116823833A (en) * 2023-08-30 2023-09-29 山东科技大学 Method, system and equipment for detecting intracranial aneurysm by using omnibearing MIP image
CN116823833B (en) * 2023-08-30 2023-11-10 山东科技大学 Method, system and equipment for detecting intracranial aneurysm by using omnibearing MIP image

Similar Documents

Publication Publication Date Title
CN104463215A (en) Tiny aneurysm occurrence risk prediction system based on retina image processing
Ren ANN vs. SVM: Which one performs better in classification of MCCs in mammogram imaging
Priya et al. SVM and neural network based diagnosis of diabetic retinopathy
CN109472781B (en) Diabetic retinopathy detection system based on serial structure segmentation
CN106651899B (en) Eye fundus image aneurysms detection system based on Adaboost
CN106295588A (en) The automatic identifying method of leukocyte in a kind of leucorrhea micro-image
CN111815562B (en) Retina blood vessel segmentation method combining U-Net and self-adaptive PCNN
CN109977955A (en) A kind of precancerous lesions of uterine cervix knowledge method for distinguishing based on deep learning
Palanivel et al. Retinal vessel segmentation using multifractal characterization
CN109636824A (en) A kind of multiple target method of counting based on image recognition technology
CN104881631B (en) The cervical cell image characteristics extraction that multiple features are integrated and recognition methods and its device
Aravind et al. Automatic detection of microaneurysms and classification of diabetic retinopathy images using SVM technique
Kaur et al. Neural network technique for diabetic retinopathy detection
Zhang et al. Microaneurysm (MA) detection via sparse representation classifier with MA and Non-MA dictionary learning
CN106780439A (en) A kind of method for screening eye fundus image
CN108109159A (en) It is a kind of to increase the retinal vessel segmenting system being combined based on hessian matrixes and region
Zhou et al. Automatic microaneurysms detection based on multifeature fusion dictionary learning
Tang et al. Retinal vessel segmentation from fundus images using DeepLabv3+
Ponni Bala et al. Early detection and classification of microaneurysms in retinal fundus images using sequential learning methods
Zhao et al. Convolutional neural networks for branch retinal vein occlusion recognition?
Shojaeipour et al. Using image processing methods for diagnosis diabetic retinopathy
Van Der Sommen et al. Computer-aided detection of early cancer in the esophagus using HD endoscopy images
Phyo et al. Automatic detection of optic disc and blood vessels from retinal images using image processing techniques
Zhou et al. Automated detection of red lesions using superpixel multichannel multifeature
CN108230341A (en) A kind of eye fundus image blood vessel segmentation method that nomography is scratched based on layering

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
WD01 Invention patent application deemed withdrawn after publication

Application publication date: 20150325

WD01 Invention patent application deemed withdrawn after publication