CN103699901A - Automatic detection method for IS/OS (intermediate system/operating system) missing area in 3D (three-dimensional) OCT (optical coherence tomography) retina image based on support vector machine - Google Patents

Abstract

The invention discloses an automatic detection method of an IS/OS (intermediate system/operating system) missing area in a 3D OCT (three-dimensional optical coherence tomography) retinal image based on a support vector machine, which comprises the steps of acquiring the retinal three-dimensional image data of a subject; layering the three-dimensional image data of the retina to find out an IS/OS area of the retina; manually and manually calibrating a missing area of the retinal IS/OS area to serve as a gold standard; selecting features to extract the features of each pixel point of the retinal IS/OS region to construct a feature set; constructing a training set and a test set; training and testing the training set and the testing set by using a support vector machine to obtain a testing result of the testing set; and comparing the test result with the gold standard to obtain the accuracy information of the test result, and feeding the test result back to the retina three-dimensional image data to finish the visual display of the test result. The method is simple and feasible; so that the quantitative expression of the relationship between retinal IS/OS loss and vision IS obtained.

Description

IS/OS disappearance region automatic testing method in a kind of 3D OCT retinal images based on support vector machine

Technical field

The present invention relates to IS/OS disappearance region automatic testing method in a kind of 3D OCT retinal images based on support vector machine, belong to Biologic Medical Image processing technology field.

Background technology

Growing perfect along with retina checkout equipment, 3D OCT(optical coherence tomography, optical coherence photography) used more and more widely.3D OCT helps ophthalmologist, oculist to observe better amphiblestroid form, and the generation development of retinal disease is made to judgement more accurately.

At present, IS/OS(Inner segment/outer segment layer in retinal images, inside/outside ganglionic layer) lack the artificial visual check that method for detecting area mainly depends on doctor, greatly strengthened doctor's workload.。

In addition, many experts and scholars, by research, find that the disappearance of retina IS/OS and the disappearance of human eyesight exist certain relation both at home and abroad, but at present, do not have any documents and materials and show, someone did the research of quantitative relationship between IS/OS disappearance and anopsia.

Summary of the invention

The deficiency existing for prior art, the object of the invention is to provide IS/OS disappearance region automatic testing method in a kind of 3D OCT retinal images based on support vector machine of simple possible.

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

The present invention includes following step:

(1) by 3D OCT equipment, obtain at least ten experimenters' retina three-dimensional view data;

(2) utilize retina delamination software, described retina three-dimensional view data is carried out to layering, find retina IS/OS region;

(3) utilize CAVASS software, the disappearance region in described retina IS/OS region is demarcated manually, as goldstandard;

(4) choose at least seven kinds of features on different stage, each pixel in described retina IS/OS region is carried out to feature extraction, construction feature collection;

(5) use leaving-one method, using the feature set of retina three-dimensional view data described in one of them experimenter as test data at every turn, using all the other remaining feature sets as training set, complete the structure of training set and test set;

(6) utilize support vector machine to carry out training and testing to described training set and test set, can obtain the test result of test set;

(7) described test result and goldstandard are compared, obtain the accuracy rate information of described test result, described test result is fed back in the retina three-dimensional view data in step (1) simultaneously, can provide one and show intuitively.

Above-mentioned experimenter is provided with ten, and chooses ten kinds of features on different stage.

Above-mentioned accuracy rate information comprises accuracy rate, true positives and true negative.

The present invention, by the extraction of feature, utilizes support vector machine to train test, can complete disappearance region robotization detection and result and show, the method simple possible; In addition, the present invention is by carrying out feature extraction to the pixel in retina IS/OS region seriatim, make " retina IS/OS disappearance exists certain relation with eyesight " this qualitative results, obtained quantitative expression, indirectly, by this quantitative relationship, assist a physician the ophthalmology disease of retina IS/OS disappearance is judged in advance.

Accompanying drawing explanation

Fig. 1 is the some sectioning images (in figure, indication closed curve scope A is disappearance) in 3D OCT retinal images;

Fig. 2 is workflow diagram of the present invention;

Fig. 3 is 13 displacement vectors that direction is corresponding;

Fig. 4 is that goldstandard result shows (in figure, white portion is shown as pixel);

Fig. 5 is that the test result of respective regions shows (in figure, white portion is shown as pixel).

Embodiment

For technological means, creation characteristic that the present invention is realized, reach object and effect is easy to understand, below in conjunction with embodiment, further set forth the present invention.

Referring to Fig. 1 and Fig. 2, first retina IS/OS disappearance of the present invention detects carries out the collection of 3DOCT view data, by the layering to retinal images, completes determining of retina IS/OS region.The disappearance position of image and non-disappearance position are demarcated, obtained goldstandard.By the feature extraction to each pixel, utilize SVM to train, test, draws last testing result.

The method of the invention is to obtain retinal images on the 3D OCT equipment of analyzing, and IS/OS disappearance detects principle and mainly utilized ten kinds of features on different scale, utilizes this machine learning method of SVM to analyze.

This method by obtaining amphiblestroid 3-D view on 3D OCT equipment, use retina delamination software, find IS/OS region, under experienced expert's help, utilize CAVASS software (a Computer Assisted Visualization and Analysis Software System), mark is manually carried out in disappearance region to retina IS/OS region, as goldstandard.

To each the pixel construction feature collection in IS/OS region, selected altogether ten kinds of features (these ten kinds of features are existing feature) herein, specific as follows:

< feature group 1>

(1) normalized gray-scale value:

$I_{\mathrm{normalized}}(i,j,k)=\frac{I_{\mathrm{original}}(i,j,k)-I_{\min }}{I_{\max }-I_{\min }}\&times;255$

Wherein, I _original(i, j, k) is the original gray-scale value of object pixel (i, j, k)

I _minit is the minimum gray-scale value in this retinal image data

I _maxit is the minimum gray-scale value in this retinal image data

I _normalized(i, j, k) is the gray-scale value after object pixel (i, j, k) normalization

(2) average (piece average) in subregion (region of 5*5*5 pixel, object pixel is positioned at center):

$M_{\mathrm{block}}(i,j,k)=\frac{1}{125}\underset{l=i-2}{\overset{i+2}{\mathrm{\&Sigma;}}}\underset{m=j-2}{\overset{j+2}{\mathrm{\&Sigma;}}}\underset{n=k-2}{\overset{k+2}{\mathrm{\&Sigma;}}}{I}_{\mathrm{normalized}}(l,m,n)$

M _block(i, j, k) is the mean value of all pixel gray-scale values in the space in the 5*5*5 region centered by object pixel (i, j, k).

(3) standard deviation in subregion (region of 5*5*5 pixel, object pixel is positioned at center):

${\mathrm{STD}}_{\mathrm{block}}(i,j,k)=\sqrt{\frac{\underset{l=i-2}{\overset{i+2}{\mathrm{\&Sigma;}}}\underset{m=j-2}{\overset{j+2}{\mathrm{\&Sigma;}}}\underset{n=k-2}{\overset{k+2}{\mathrm{\&Sigma;}}}{(I_{\mathrm{normalized}}(l,m,n)-M_{\mathrm{block}}(i,j,k))}^2}{125}}$

STD _block(i, j, k) is the standard deviation of the gray-scale value of all pixels in the space in the 5*5*5 region centered by object pixel (i, j, k).

(4) entropy in subregion (region of 5*5*5 pixel, object pixel is positioned at center):

The quantity of information that entropy Description Image has, shows the complicated process of image, and image complexity is high, and entropy is larger, otherwise less

${\mathrm{ENT}}_{\mathrm{block}}=-\underset{m=0}{\overset{M-1}{\mathrm{\&Sigma;}}}p\left(r_m\right){\log }_{2}p\left(r_m\right)$

Wherein, r _mit is the gray-scale value of pixel

P(r _m) be the probability that this gray-scale value occurs

ENT _blockit is the information entropy in this 5*5*5 region.

< feature group 2>

Because data are three-dimensional, so 13 directions (referring to Fig. 3) that can be defined as follows.

1. directions X, Y-direction, Z direction, totally 3;

2. X-Y plane diagonal, Y-Z plane diagonal, X-Z plane diagonal, totally 6;

3. body diagonal direction, totally 4.

Be defined as follows shown in Fig. 3: α is the projection of this direction in X-Y direction and the angle of X-axis, β is the angle of this direction and Z axis, and D is step-length.

(1) step-length is 1 o'clock, the gray-scale value antipode of the pixel in 13 directions of object pixel:

(2) step-length is 2 o'clock, the gray-scale value antipode of the pixel in 13 directions of object pixel:

< feature group 3>

By build the region of 5*5*5 centered by object pixel, obtain three-dimensional gray level co-occurrence matrixes, and calculate corresponding specific features, what choose is that direction 1 builds gray level co-occurrence matrixes herein.After completing the structure of gray level co-occurrence matrixes, the gray level co-occurrence matrixes fundamental function that can carry by MATLAB, calculates four category features below:

(7) contrast (contrast) in subregion (region of 5*5*5 pixel, object pixel is positioned at center)

(8) autocorrelation (correlation) in subregion (region of 5*5*5 pixel, object pixel is positioned at center)

(9) energy (energy) in subregion (region of 5*5*5 pixel, object pixel is positioned at center)

(10) homogeneity (homogeneity) in subregion (region of 5*5*5 pixel, object pixel is positioned at center) so far, has completed the structure of feature set.

Because the present embodiment data set used is limited, so locate to use existing method leaving-one method (leave-one-out) to carry out training and testing data.Support vector machines is a kind of basic, novel small-sample learning method that has solid theory, utilize it to carry out training and testing and can obtain test result, add up and can obtain corresponding accuracy rate, true positives, true negative, test result also can feed back in original data and go simultaneously, provides one and shows intuitively (referring to Fig. 4 and Fig. 5).

TP: the pixel count that disappearance correctly detected

FP: flase drop is the pixel count of disappearance

TN: the pixel count that non-disappearance correctly detected

FN: the pixel count that flase drop is non-disappearance

True positives: $\mathrm{TPR}=\frac{\mathrm{TP}}{\mathrm{TP}+\mathrm{FN}}$

True negative: $\mathrm{TNR}=\frac{\mathrm{TN}}{\mathrm{TN}+\mathrm{FP}}$

Accuracy rate: $\mathrm{ACC}=\frac{\mathrm{TP}+\mathrm{TN}}{\mathrm{TP}+\mathrm{FP}+\mathrm{TN}+\mathrm{FN}}$

More than show and described 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; that in above-described embodiment and instructions, describes 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.The claimed scope of the present invention is defined by appending claims and equivalent thereof.

Claims (3)

1. an IS/OS disappearance region automatic testing method in the 3D OCT retinal images based on support vector machine, is characterized in that, comprises following step:

(1) by 3D OCT equipment, obtain at least ten experimenters' retina three-dimensional view data;

(2) utilize retina delamination software, described retina three-dimensional view data is carried out to layering, find retina IS/OS region;

(3) utilize CAVASS software, the disappearance region in described retina IS/OS region is demarcated manually, as goldstandard;

(4) choose at least seven kinds of features on different stage, each pixel in described retina IS/OS region is carried out to feature extraction, construction feature collection;

(5) use leaving-one method, using the feature set of retina three-dimensional view data described in one of them experimenter as test data at every turn, using all the other remaining feature sets as training set, complete the structure of training set and test set;

(6) utilize support vector machine to carry out training and testing to described training set and test set, can obtain the test result of test set;

(7) described test result and goldstandard are compared, obtain the accuracy rate information of described test result, described test result is fed back in the retina three-dimensional view data in step (1) simultaneously, complete the demonstration directly perceived of test result.

2. IS/OS disappearance region automatic testing method in the 3D OCT retinal images based on support vector machine according to claim 1, is characterized in that,

Described experimenter is provided with ten, and chooses ten kinds of features on different stage.

3. IS/OS disappearance region automatic testing method in the 3D OCT retinal images based on support vector machine according to claim 1, is characterized in that,

Described accuracy rate information comprises accuracy rate, true positives and true negative.

Images