CN105551038A - Method for fully automatically classifying and segmenting retinal branch artery obstruction based on three-dimensional OCT image - Google Patents

Abstract

The invention discloses a method for fully automatically classifying and segmenting retinal branch artery obstruction based on a three-dimensional OCT image, which comprises the following steps: pretreatment: layering the retina through a graph search algorithm, and then flattening each layer of the retina according to the pigment epithelium layer; automatically classifying acute stage and atrophic stage of retinal branch artery occlusion by using an AdaBoost classifier; segmentation of retinal branch artery occlusion acute phase: firstly, initializing and segmenting a blocking area by adopting Bayesian posterior probability; then accurately segmenting the blocking area based on a graph search-graph segmentation algorithm; (4) segmentation of retinal branch artery occlusion atrophy phase: the occlusion regions during atrophy are automatically segmented by building an inner retinal thickness model. The method can accurately classify and segment the blocked area of the branch artery of the retina, and can replace manual classification and segmentation.

Description

A kind of method that automated classification based on three-dimensional OCT image and segmentation Branch Retinal Artery block

Technical field

The present invention relates to the classification of pathology and the dividing method of lesion region in the retinal images of SD-OCT (domain optical coherence fault imaging), be specifically related to a kind of method of automated classification based on three-dimensional OCT image and segmentation Branch Retinal Artery obstruction, belong to the method and technology field of classification and segmentation retinal images.

Background technology

It is one of acute disease of ophthalmology that Branch Retinal Artery blocks.Its prognosis is poor, and morbidity is quick, the normally monocular vision obstacle of Silent Neuritis.Retinal arterial obstruction makes corresponding retinal area nutrition supply interrupt, and causes retina regional area anoxic, ischemic, and form oedema, retina cell is sharply dead, thus causes visual disorder.

Up to the present, great majority block relevant work to Branch Retinal Artery and all concentrate on the analysis qualitatively of blocking Branch Retinal Artery, as: the people such as H.Chen propose the framework analyzing each layer light intensity of retina in OCT image; The manual measurement macula luteas such as CKSLeng and the relation investigating Branch Retinal Artery obstruction patient's retina 26S Proteasome Structure and Function depending on nipple peripheral retinal tear retinal nerve fiber layer thickness and visual sensitivity; B.Asefzadeh and K.Ninyo analyzes longitudinal Fundus oculi changes of the optic disk peripheral nerve fiber layer thickness that Branch Retinal Artery blocks.

These methods are all the qualitative analyses of blocking Branch Retinal Artery, can not completely automatically detect and split the region of obstruction.Therefore, quantitative information accurately about congested areas can not be provided to clinician, as shape, size and position etc.Generally speaking, there is following defect in the method that current Branch Retinal Artery blocks: (1) most of algorithm does not all block Branch Retinal Artery and classifies (acute stage and atrophy phase), and Branch Retinal Artery blocks the amphiblestroid institutional framework of different times and differs greatly.(2) most of method is not completely automatically, by hand dipping or mark.(3) most of algorithm does not all have the congested areas of blocking for Branch Retinal Artery to carry out concrete analysis.

Branch Retinal Artery blocks many caused by embolus or thrombosis, and visual impairment degree and fundus Oculi Manifestations are determined according to obstructive position and degree.If the region of obstruction automatically accurately can be partitioned into, doctor just can be helped well to carry out diagnosing and formulate corresponding therapeutic scheme, to patient's vision restoration highly significant.But due to the shape of the congested areas that Branch Retinal Artery blocks, size, the position of appearance all has arbitrariness, and the boundary of congested areas and surrounding tissue is very fuzzy, adds that retina OCT image itself is with noise.Therefore, full automation Ground Split Branch Retinal Artery congested areas is a challenging task.

Summary of the invention

For the deficiency that prior art exists, the object of the invention is to provide a kind of method of automated classification based on three-dimensional OCT image and segmentation Branch Retinal Artery obstruction, can classify to Branch Retinal Artery congested areas accurately and split, manual classification and segmentation can be substituted.

To achieve these goals, the present invention realizes by the following technical solutions:

The method that a kind of automated classification based on three-dimensional OCT image of the present invention and segmentation Branch Retinal Artery block, bag

Draw together following step:

(1) pre-service: by graph search algorithm (existing algorithm), layering is carried out to retina, then according to pigment epithelial layer, each layer of retina is evened up;

(2) acute stage using AdaBoost (a kind of algorithm being produced final strong classifier by iteration Weak Classifier) sorter to block Branch Retinal Artery and atrophy phase carry out automatic classification;

(3) Branch Retinal Artery blocks the segmentation of acute stage: first adopt Bayes posterior probability to carry out initialize partition to congested areas; Then cut algorithm based on graph search-Tu and Accurate Segmentation is carried out to congested areas;

(4) Branch Retinal Artery blocks the segmentation of atrophy phase: carry out auto Segmentation by setting up interior retinal thickness model to the congested areas of atrophy phase.

In step (1), the cost function of described graph search algorithm is defined as:

$E\left(S\right)=\underset{v\∈S}{\Σ}{c}_v+\underset{(p,q)\∈N}{\Σ}{h}_{p,q}\left(S\right(p)-S(q\left)\right)$

Wherein, v is a voxel, and S is the surface of requirement, c _vbe a cost based on edge, comprise the possibility inverse correlation of voxel v with S, (p, q) is a pair adjacent voxel, and N is the set of voxel in image, p and q all in image N, h _p,qbe the cost of shape in the upper change of p, q of surperficial S, S (p) is the position of voxel p on surperficial S.

In step (2), the method that the acute stage using AdaBoost sorter to block Branch Retinal Artery and atrophy phase carry out automatic classification is as follows:

A () extracts amphiblestroid textural characteristics, shape facility and position feature;

B () adopts AdaBoost sorter train the feature in step (a) and select, three-dimensional OCT image 23 example that classification uses altogether Branch Retinal Artery to block, wherein 12 routine Branch Retinal Artery block acute stage image, and 11 routine Branch Retinal Artery block the image of atrophy phase.And a method is abandoned in use, namely selects the 3-D view of a patient to test, is trained by remaining image at every turn, thus Branch Retinal Artery is divided into by each image to block acute stage or atrophy phase.

In step (3), the method adopting Bayes posterior probability to carry out initialize partition to congested areas is as follows:

(3 ?1) use Bayes posterior probability to estimate that each voxel belongs to the possibility of congested areas, and the computing formula of this possibility probability is:

$P\left(occ\right|I_p)=\frac{P\left(I_P\right|occ)\·P\left(occ\right)}{P\left(I_P\right|occ)\·P\left(occ\right)+P\left(I_P\right|non)\·P\left(non\right)}---\left(1\right)$

Wherein, I _prepresent the brightness value of voxel, occ represents obstruction, and non represents it is not block; P (occ) and P (non) represents congested areas in whole image and is not the probability of congested areas; P (I _p| occ) and P (I _p| non) represent that brightness is I _pvoxel p belong to congested areas and be not the probability of congested areas; In the training stage, interior amphiblestroid each voxel is marked as and blocks or be not block, and this mark is the hand labeled that instructs according to clinician; At test phase, probability between interior given one 0 to 1 of amphiblestroid each voxel estimates that it is the possibility of congested areas, thus obtain the probability graph of whole image, described probability graph is used in the cost function that graph search-Tu when splitting cuts as constraint;

After (3 ?2) obtain described probability graph, first, mean filter is used to carry out smoothed image; Then, separated by point high for probability by threshold value, these points are regarded as the region of obstruction; Finally, initialize partition result is obtained by morphologic open and close operation.

In step (3), cut algorithm (existing algorithm) based on graph search-Tu, with max-flow-minimal cut algorithm, carry out auto Segmentation according to cost function is minimum to image, the cost function that graph search-Tu cuts algorithm is defined as:

E(f)＝E(Surfaces)+E(Regions)+E(Interactions)

(2)

Wherein, E (Surfaces) represents all costs relevant to surface segmentation, and E (Regions) represents the cost relevant to cut zone, the cost that E (Interactions) retrains between presentation surface and region;

Based on initialize partition result, with morphological erosion operation (prior art) obtain described figure Sou Suo ?figure cut foreground seeds point needed for algorithm, then use expansive working (prior art) obtain described figure Sou Suo ?figure cut background Seed Points needed for algorithm.

In step (4), as follows by setting up the method that interior retinal thickness model carries out auto Segmentation to the congested areas of atrophy phase:

(4-1) image alignment: because retinal images divides left eye and right eye, its retinal structure becomes mirror image; In order to obtain retinal thickness model in unification, left-eye image being overturn, makes it consistent with right eye;

(4-2) set up interior retinal thickness model: in retina OCT image, centered by macula lutea, the interior amphiblestroid thickness at record diverse location place, forms the thickness model of a 2D;

(4-3) atrophy phase congested areas segmentation: the retina OCT image choosing several normal eyes, average as retinal thickness model in the normal eye of standard with retinal thickness model in them, compare with retinal thickness in atrophy phase patient, thus be partitioned into congested areas.

The acute stage that the present invention uses AdaBoost sorting algorithm automatic distinguishing Branch Retinal Artery to block and atrophy phase, and use different dividing methods to carry out auto Segmentation to it according to its institutional framework and textural characteristics; Cut algorithm in conjunction with Bayes posterior probability and graph search-Tu, be accurately partitioned into the congested areas that Branch Retinal Artery blocks acute stage; And retinal thickness model is efficiently partitioned into the congested areas of atrophy phase rapidly in using; Classification and segmentation all have higher accuracy, and therefore this method can substitute manual classification and segmentation, and the Clinics and Practices for clinical relevant ophthalmology disease can play important booster action.

Accompanying drawing explanation

Fig. 1 (a) is retina layering and the effect (layering of atrophy phase) of evening up;

Fig. 1 (b) is retina layering and the effect (layering of acute stage) of evening up;

Fig. 2 is the graph of a relation of posterior probability and voxel intensity;

Fig. 3 (a) is the initialized original image of Bayes;

Fig. 3 (b) is the initialized probability graph of Bayes;

Fig. 3 (c) is the initialized segmentation result figure of Bayes;

Fig. 4 (a) is retinal thickness model in normal retina;

Fig. 4 (b) is the interior retinal thickness model that first Branch Retinal Artery blocks atrophy phase patient;

Fig. 4 (c) is the interior retinal thickness model that first Branch Retinal Artery blocks atrophy phase patient;

Fig. 5 (a) is the segmentation result (first is classified as former figure, and second is classified as normative reference, and the 3rd is classified as segmentation result) of Branch Retinal Artery obstruction acute stage;

Fig. 5 (b) is the segmentation result (first is classified as former figure, and second is classified as normative reference, and the 3rd is classified as segmentation result) of Branch Retinal Artery obstruction atrophy phase.

Embodiment

The technological means realized for making the present invention, creation characteristic, reaching object and effect is easy to understand, below in conjunction with embodiment, setting forth the present invention further.

The present invention includes following four steps: Image semantic classification, the segmentation of blocking acute stage based on the classification of AdaBosst, Branch Retinal Artery and Branch Retinal Artery block the segmentation of atrophy phase.

(1) Image semantic classification

In order to obtain the information of each layer of retina, graph search method is used to realize amphiblestroid layering.Its cost function is defined as:

$E\left(S\right)=\underset{v\∈S}{\Σ}{c}_v+\underset{(p,q)\∈N}{\Σ}{h}_{p,q}\left(S\right(p)-S(q\left)\right)$

Wherein, S is the surface of requirement, c _vbe a cost based on edge, comprise the possibility inverse correlation of voxel v with S.(p, q) is a pair adjacent voxel.H _p,qthe cost of shape in the upper change of p, q of surperficial S.

Then according to pigment epithelial layer, each layer of retina is evened up.Retina layering and the effect of evening up are as shown in Fig. 1 (a), Fig. 1 (b).Only marked 4 layers that the present invention uses in figure, be respectively retinal nerve fiber coboundary from top to bottom, outer plexiform layer lower boundary, outer nuclear layer lower boundary and pigment epithelial layer lower boundary.Retina is divided into interior retina and outer retina by these 4 layers, and interior retina is between Article 1 line and Article 2 line (from top to bottom), and outer retina is between Article 3 line and Article 4 line (from top to bottom).

(2) based on the classification of AdaBoost

Have acute stage and atrophy phase because Branch Retinal Artery blocks, both retinal structures are not identical with texture, therefore need first to block Branch Retinal Artery to classify.The acute stage that the present invention uses AdaBoost sorter to block Branch Retinal Artery and atrophy phase classify, and comprise two parts, specifically describe as follows:

A () puies forward feature

The present invention extracts 50 features to each voxel, comprises interior retinal thickness, voxel intensity, gray level co-occurrence matrixes energy, entropy, inertia, relevant and gal cypress (Gabor) filtering transformation, as shown in table 1.These feature interpretation texture of each voxel, structure and positional information.

Table 1 is classified feature used

Feature sequence number	Feature interpretation
		1	Interior retinal thickness
2	Voxel intensity
		3-34	Gray level co-occurrence matrixes energy, entropy, inertia, relevant average in 8 directions and standard deviation
35-50	The output of Jia Bai (Gabor) wave filter, two centre frequencies, 8 directions

B () AdaBoost classifies

Adopt AdaBoost sorter train feature and select, and a method is abandoned in use, selects the three-dimensional data of a patient to test at every turn, remaining data is trained, the acute stage be divided into Branch Retinal Artery to block each data or atrophy phase.

(3) Branch Retinal Artery blocks the segmentation of acute stage

Branch Retinal Artery obstruction acute stage, shows as interior retina regional area brightness and strengthens on OCT image, and this region brightened is exactly congested areas.The present invention cuts algorithm in conjunction with Bayes posterior probability and graph search-Tu and carrys out auto Segmentation congested areas, comprises following two steps: initialization and Accurate Segmentation.

A () carries out initialize partition with Bayes posterior probability to congested areas

The present invention uses Bayes posterior probability to estimate that each voxel belongs to the possibility of congested areas.The computing formula of this possibility probability is:

$P\left(occ\right|I_p)=\frac{P\left(I_P\right|occ)\·P\left(occ\right)}{P\left(I_P\right|occ)\·P\left(occ\right)+P\left(I_P\right|non)\·P\left(non\right)}---\left(1\right)$

Wherein, I _prepresent the brightness value of voxel, occ represents obstruction, and non represents it is not block; P (occ) and P (non) represents congested areas in whole image and is not the probability of congested areas; P (I _p| occ) and P (I _p| non) represent that brightness is I _pvoxel p belong to congested areas and be not the probability of congested areas.Fig. 2 is the graph of a relation of posterior probability and voxel intensity.

See Fig. 3 (a), in the training stage, interior amphiblestroid each voxel is marked as and blocks or be not block, and this mark is the hand labeled that instructs according to clinician.At test phase, the probability between given one 0 to 1 of interior amphiblestroid each voxel estimates that it is the possibility of congested areas, obtains the probability graph of whole image, as shown in Fig. 3 (b).This probability graph is used for as constraint in the cost function split.

After obtaining probability graph, obtain initialize partition result by some post-processing operation.First, mean filter is used to carry out smoothed image.Then, separated by point high for probability by threshold value 0.5, these points are regarded as the region of obstruction.Initialize partition result is obtained finally by morphologic open and close operation.By opening the little loose point outside operation removal congested areas, fill the blank spot inside congested areas by closed operation.Initialization result is as shown in Fig. 3 (c).

(b) based on figure Sou Suo ?figure cut the auto Segmentation of algorithm

This method use graph search-Tu cuts algorithm and carries out Accurate Segmentation to congested areas, and cost function is designed to:

E(f)＝E(Surfaces)+E(Regions)+E(Interactions)

(2)

Wherein, E (Surfaces) represents all costs relevant to surface segmentation, and E (Regions) represents the cost relevant to cut zone, the cost that E (Interactions) retrains between presentation surface and region.

Graph search-Tu cuts required prospect and background Seed Points is obtained automatically by Morphology Algorithm.Based on initialize partition result, obtain foreground seeds point with morphological erosion operation, then obtain background Seed Points with expansive working.

(4) Branch Retinal Artery blocks the segmentation of atrophy phase

Branch Retinal Artery blocks the atrophy phase and on OCT image, shows as interior retinal thickness reduce, and therefore carries out ratio of division based on the change of thickness to the congested areas of atrophy phase more reasonable.But due to retina itself thickness in different positions also different (thicker near macula lutea, place far away is thinner, as Suo Shi Fig. 4 (a)), easily affect judgement.The present invention proposes to use the method for thickness model to split the congested areas of atrophy phase, comprises following 3 steps: (1) image alignment.Because retinal images divides left eye and right eye, its retinal structure becomes mirror image.In order to obtain retinal thickness model in unification, in the present invention, left-eye image being overturn, making it consistent with right eye.(2) interior retinal thickness model is set up.In retina OCT image, record centered by macula lutea, the interior amphiblestroid thickness at diverse location place, form the thickness model of a 2D, the congested areas segmentation of (3) atrophy phase.Choose the retinal images of 20 normal eyes, average as retinal thickness model in the normal eye of standard with retinal thickness model in them, compare with the interior retinal thickness of atrophy phase patient (as shown in Fig. 4 (b), (c)), thus be partitioned into congested areas.

Experimental result

The present invention uses altogether the data of the patient that 23 Branch Retinal Artery block, wherein 12 example acute stages, 11 routine atrophy phases.By expert diagnosis and hand labeled congested areas as goldstandard.Specific experiment result is as follows.

A classification results that () Branch Retinal Artery blocks

The classification results that Branch Retinal Artery blocks acute stage and atrophy phase is as shown in table 2, and the overall accuracy of AdaBoost sorter is 87.0%.

The classification performance that table 2 Branch Retinal Artery blocks

(b) Branch Retinal Artery congested areas segmentation result

Adopt True Positive Rate TPVF and false positive rate FPVF as the objective indicator of appraisal procedure, be calculated as follows:

$TPVF=\frac{V_{TP}}{V_{TP}+V_{FN}}---\left(3\right)$

$FPVF=\frac{V_{FP}}{V_{FP}+V_{TN}}---\left(4\right)$

Wherein, V _tP, V _fP, V _tNand V _fNtrue positives, false positive, true negative and false-negative volume respectively.Experimental result shows, for the acute stage that Branch Retinal Artery blocks, the Average True positive rate of this method is 91.1%, and average false positive rate is 5.5%; For the atrophy phase, the Average True positive rate of this method is 90.5%, and average false positive rate is 8.7%.Part segmentation result as Fig. 5 (a), (b), shown in.

So far, a kind ofly the automatic classification of three-dimensional OCT image that Branch Retinal Artery blocks is applicable to and dividing method has realized and verified.Present invention incorporates graph search algorithm, AdaBoost sorter, Bayes posterior probability, graph search-Tu cuts algorithm and Morphology Algorithm carries out automatic classification and segmentation to Branch Retinal Artery obstruction, and classification and segmentation all there is higher accuracy, therefore this method can substitute manual classification and segmentation, and the Clinics and Practices for clinical relevant ophthalmology disease can play important booster action.

More than show and describe ultimate principle of the present invention and principal character and advantage of the present invention.The technician of the industry should understand; the present invention is not restricted to the described embodiments; what describe in above-described embodiment and instructions just illustrates principle of the present invention; without departing from the spirit and scope of the present invention; the present invention also has various changes and modifications, and these changes and improvements all fall in the claimed scope of the invention.Application claims protection domain is defined by appending claims and equivalent thereof.

Claims (6)

1., based on the method that automated classification and the segmentation Branch Retinal Artery of three-dimensional OCT image block, it is characterized in that, comprise following step:

(1) pre-service: carry out layering to retina by graph search algorithm, then evens up each layer of retina according to pigment epithelial layer;

(2) acute stage using AdaBoost sorter to block Branch Retinal Artery and atrophy phase carry out automatic classification;

(3) Branch Retinal Artery blocks the segmentation of acute stage: first adopt Bayes posterior probability to carry out initialize partition to congested areas; Then cut algorithm based on graph search-Tu and Accurate Segmentation is carried out to congested areas;

(4) Branch Retinal Artery blocks the segmentation of atrophy phase: carry out auto Segmentation by setting up interior retinal thickness model to the congested areas of atrophy phase.

2. the method for the automated classification based on three-dimensional OCT image according to claim 1 and segmentation Branch Retinal Artery obstruction, it is characterized in that, in step (1), the cost function of described graph search algorithm is defined as:

$E\left(S\right)=\underset{v\∈S}{\Σ}{c}_v+\underset{(p,q)\∈N}{\Σ}{h}_{p,q}\left(S\right(p)-S(q\left)\right)$

Wherein, v is a voxel, and S is the surface of requirement, c _vbe a cost based on edge, comprise the possibility inverse correlation of voxel v with S, (p, q) is a pair adjacent voxel, and N is the set of voxel in image, p and q all in image N, h _p,qbe the cost of shape in the upper change of p, q of surperficial S, s (p) is the position of voxel p on surperficial S.

3. the method for the automated classification based on three-dimensional OCT image according to claim 1 and segmentation Branch Retinal Artery obstruction, it is characterized in that, in step (2), the method that the acute stage using AdaBoost sorter to block Branch Retinal Artery and atrophy phase carry out automatic classification is as follows:

A () extracts amphiblestroid textural characteristics, shape facility and position feature;

B () adopts AdaBoost sorter train the feature in step (a) and select, and a method is abandoned in use, namely select the 3-D view of a patient to test at every turn, remaining image is trained, thus is divided into by each image Branch Retinal Artery to block acute stage or atrophy phase.

4. the method for the automated classification based on three-dimensional OCT image according to claim 1 and segmentation Branch Retinal Artery obstruction, it is characterized in that, in step (3), the method adopting Bayes posterior probability to carry out initialize partition to congested areas is as follows:

(3 ?1) use Bayes posterior probability to estimate that each voxel belongs to the possibility of congested areas, and the computing formula of this possibility probability is:

$P\left(occ\right|I_p)=\frac{P\left(I_P\right|occ)\·P\left(occ\right)}{P\left(I_P\right|occ)\·P\left(occ\right)+P\left(I_P\right|non)\·P\left(non\right)}---\left(1\right)$

Wherein, I _prepresent the brightness value of voxel, occ represents obstruction, and non represents it is not block; P (occ) and P (non) represents congested areas in whole image and is not the probability of congested areas; P (I _p| occ) and P (I _p| non) represent that brightness is I _pvoxel p belong to congested areas and be not the probability of congested areas; In the training stage, interior amphiblestroid each voxel is marked as and blocks or be not block, and this mark is the hand labeled that instructs according to clinician; At test phase, probability between interior given one 0 to 1 of amphiblestroid each voxel estimates that it is the possibility of congested areas, thus obtain the probability graph of whole image, described probability graph is used in the cost function that graph search-Tu when splitting cuts as constraint;

After (3 ?2) obtain described probability graph, first, mean filter is used to carry out smoothed image; Then, separated by point high for probability by threshold value, these points are regarded as the region of obstruction; Finally, initialize partition result is obtained by morphologic open and close operation.

5. the method for the automated classification based on three-dimensional OCT image according to claim 4 and segmentation Branch Retinal Artery obstruction, it is characterized in that, in step (3), algorithm is cut based on graph search-Tu, with max-flow-minimal cut algorithm, carry out auto Segmentation according to cost function is minimum to image, the cost function that graph search-Tu cuts algorithm is defined as:

E(f)＝E(Surfaces)+E(Regions)+E(Interactions)(2)

Wherein, E (Surfaces) represents all costs relevant to surface segmentation, and E (Regions) represents the cost relevant to cut zone, the cost that E (Interactions) retrains between presentation surface and region;

Based on initialize partition result, with morphological erosion operation obtain described figure Sou Suo ?figure cut foreground seeds point needed for algorithm, then with expansive working obtain described figure Sou Suo ?figure cut background Seed Points needed for algorithm.

6. the method for the automated classification based on three-dimensional OCT image according to claim 1 and segmentation Branch Retinal Artery obstruction, it is characterized in that, in step (4), as follows by setting up the method that interior retinal thickness model carries out auto Segmentation to the congested areas of atrophy phase:

(4-1) image alignment: because retinal images divides left eye and right eye, its retinal structure becomes mirror image; In order to obtain retinal thickness model in unification, left-eye image being overturn, makes it consistent with right eye;

(4-2) set up interior retinal thickness model: in retina OCT image, centered by macula lutea, the interior amphiblestroid thickness at record diverse location place, forms the thickness model of a 2D;

(4-3) atrophy phase congested areas segmentation: the retina OCT image choosing several normal eyes, average as retinal thickness model in the normal eye of standard with retinal thickness model in them, compare with retinal thickness in atrophy phase patient, thus be partitioned into congested areas.