CN103006175A - Method for positioning optic disk for eye fundus image on basis of PC (Phase Congruency) - Google Patents

Abstract

The invention belongs to the field of medical image processing, and relates to a method for locating an optic disc of a fundus image based on phase consistency. In this method, after selecting 4 single-channel images with prominent optic discs, the selected images are processed using the Phase Congruency (PC) function, and the processing results are enhanced by logic "AND". Window scanning and grayscale accumulation are performed on the enhanced fundus image to locate the optic disc. The invention is not affected by the noise, brightness and contrast changes in the image, and has high accuracy for the optic disc positioning of the normal fundus and mild, moderate and severe diabetic fundus images.

Description

A method of optic disc localization in fundus images based on phase consistency

technical field

This patent relates to a method for locating the optic disc in fundus images based on phase consistency. This method can determine the position of the optic disc in the fundus image, and has high accuracy for normal fundus and fundus images of mild, moderate and severe diabetic lesions. The invention belongs to the technical field of image processing and can be applied to clinical ophthalmology and the diagnosis and treatment of diseases related to fundus lesions.

Background technique

Medical image processing and analysis has always been a focus and hot issue in the field of image processing and analysis. With the powerful means of graphics and image technology, the quality and display methods of medical images have been greatly improved, which has greatly improved the level of diagnosis and treatment. Image processing technology has been introduced into the Department of Ophthalmology for many years. Through the calculation and analysis of fundus images, it can quantitatively measure important fundus tissues such as optic disc, retinal blood vessels and macular fovea, and make a clear distinction between normal and abnormal, and can detect various diseases early and accurately. A variety of eye diseases and quite a lot of systemic diseases, such as diabetes, hypertension, arteriosclerosis and so on. With the urgent need for clinical fundus detection in society, the existing fundus image detection methods have shown many shortcomings in practical application and testing, and the difficulties and challenges in this field are also increasing.

The optic disc is one of the important tissues of the fundus. It is a light red circular area with a diameter of about 1.5mm, also known as the optic nerve head. It is the part where retinal nerve fibers and retinal blood vessels enter and exit the eyeball, and receives nerves generated by visual perception cells. The impulse is further transmitted to the brain through the optic nerve to form vision. Its center is a funnel-shaped depression, called a physiological depression. Four retinal arteries and veins can be seen in the center of the optic disc, which are important guarantees for maintaining retinal nutrition. Parameters such as the shape, area, and depth of the optic disc are important indicators to measure the health of the fundus. Optic disc positioning is the basis for fundus image processing such as registration and stitching of fundus images, blood vessel segmentation, macula and lesion extraction, and optic disc segmentation. The diversity of imaging and individual differences leads to huge differences in the shape, color, and size of the optic disc. In particular, the existence of various fundus lesions makes automatic positioning of the optic disc difficult. It is of great significance for clinical ophthalmology research and the diagnosis and treatment of diseases related to fundus lesions to develop an optic disc positioning method in fundus images that can meet the accuracy, objectivity and repeatability standards required by clinical fundus detection.

So far, the research work on fundus image detection technology has achieved many results, and scholars at home and abroad have proposed many optic disc positioning methods. Such as: Hoover et al. use the method of fuzzy clustering to establish the relationship between the blood vessel and the optic disc, and then obtain a better positioning effect (Hoover A, Goldbaum M.Locating the optic nerve in a retinal image using the fuzzy convergence of the blood vessels[J ]. IEEE Transaction on Medical Imaging, 2003, 22(8): 951-958). Based on the precise extraction of the vascular network, Tobin et al. located the optic disc by analyzing its brightness, width and direction information (Tobin K W, Chaum E, Govindasamy VP, et al. Detection of anatomic structures in human retinal imagery[J].IEEE Transaction on Medical Imaging, 2007, 26(12): 1729-1739). Li Jupeng and others achieved optic disc positioning by constructing a "vessel-like" intersection network (Li Jupeng, Chen Houjin, Zhang Xinyuan. Automatic positioning of the optic nerve head based on the intersection network. Journal of Electronics and Information Technology, 2009, 31(5): 1170-1174). Based on the different properties of the optic disc in the fundus, the existing optic disc positioning methods are generally divided into two types, the method based on the characteristics of the optic disc itself and the method based on the retinal vascular structure. The first type of method locates the optic disc based on the roughly circular structure of the optic disc and its higher brightness relative to other parts of the fundus. For retinas with lesions, this method is not robust, especially for those with fundus diseases. In this case, this method cannot achieve good positioning results. In addition, most of these methods first remove fundus blood vessels through morphological preprocessing methods. However, morphological preprocessing uses globally unified and fixed morphological structural elements, and morphological transformations are inherently nonlinear and irreversible, so it is inevitable Introduce large changes to the optic disc edge features, resulting in inaccurate positioning results. The second type of method is based on the property that the optic disc is the converging point of the retinal vascular network, which needs to segment the blood vessels in advance, has high requirements for the extraction accuracy of the vascular network, and has high computational complexity, and is only suitable for retinal images with clear vascular networks. Difficult to achieve optic disc positioning under different image quality.

The background of the fundus image is complex, the contrast between the optic disc and the background is low, the illumination is uneven, and the size of the optic disc is not uniform and other factors will affect the positioning results, resulting in inaccurate positioning and even loss of the target. Therefore, there is a need for a universal and robust optic disc localization method in fundus images that is not affected by noise, brightness and contrast changes in the image.

Contents of the invention

In order to achieve the purpose that the optic disc positioning result is not affected by factors such as image noise and brightness, the present invention provides a method for positioning the optic disc of the fundus image based on phase consistency, which includes the following steps: first, in order to highlight the optic disc, the fundus image is transformed into Lab , Yuv, Yiq and Hsv four color spaces, respectively select the channel images with the strongest contrast between the disc and the background in each space; then, carry out phase consistency processing on the four selected channel images, and enhance them through logical "AND" operation Results; Finally, the optic disc was located using the method of window scanning and gray scale accumulation.

1. Select the color space channel suitable for fundus image optic disc location: the inventive method has selected the a channel of Lab space, the u channel of Yuv space, the i channel of Yiq space and the s channel of Hsv space. The optic discs of the fundus images in these 4 channels are obviously prominent and maintain good edge characteristics.

2. Perform PC processing on the four single-channel images selected in step 1, and calculate the point with the largest phase consistency in the fundus image by estimating the peak value in the local energy function. This patent selects the log Gabor wavelet function and its Hilbert transform as the filter for calculating local energy.

3. Perform logical "AND" operation on the PC results of 4 single-channel images to enhance the PC results.

4. Select an 80×80 rectangular window to scan the entire logical AND image, calculate the maximum pixel mean value in the window, and use the centroid coordinates of the window to initially locate the optic disc.

5. Carry out grayscale accumulation along the X and Y directions in the rectangular window obtained in step 4 and calculate the maximum grayscale values X _max and Y _max , set the grayscale threshold in the X direction to 0.5*X _max , and the grayscale in the Y direction The threshold is set to 0.5*Y _max , and the image part larger than the threshold is reserved, and the center coordinates of the reserved part are calculated, and the position of the optic disc is finally determined by using the coordinates.

Compared with existing methods, the beneficial effects of the present invention are:

1. The phase information is not affected by factors such as low contrast and uneven illumination. Therefore, this patent has strong robustness to the positioning of the optic disc under different image qualities.

2. This patent considers the factor that the size of the optic disc of different individuals is not uniform, and adopts two-step positioning to finally determine the position of the optic disc, ensuring the accuracy of positioning.

3. This patent can realize the automatic positioning of the optic disk of normal fundus and mild, moderate and severe diseased fundus images, and can realize the analysis function of fundus images of patients. The scope of application of digital diagnostic systems is of great significance.

Description of drawings

Fig. 1: Flowchart of the optic disc positioning method of the present invention.

Figure 2: Single-channel fundus images in different color spaces. Figure 2-1 is the a-channel fundus image in Lab space, Figure 2-2 is the u-channel fundus image in Yuv space, Figure 2-3 is the i-channel fundus image in Yiq space, and Figure 2-4 is the s-channel fundus image in Hsv space image.

Figure 3: Single-channel fundus images in different color spaces after PC processing. Figure 3-1 is the a-channel fundus image in Lab space, Figure 3-2 is the u-channel fundus image in Yuv space, Figure 3-3 is the i-channel fundus image in Yiq space, and Figure 3-4 is the s-channel fundus image in Hsv space image.

Figure 4: Image after logical AND.

Figure 5: Diagram of the results of the optic disc positioning experiment in clinical fundus images. The first column is the original image; the second column is the positioning result of the method of the present invention. Figure 5-1 is a normal fundus image, Figure 5-2 is a mildly diseased fundus image, Figure 5-3 is a moderately diseased fundus image, and Figure 5-4 is a severely diseased fundus image.

Figure 6: Using the STARE library to perform the optic disc positioning experiment, and comparing it with the positioning results of the Hoover method. The first column is the original image; the second column is the positioning result of the Hoover method; the third column is the positioning result of the method of the present invention. Figure 6-1 is a normal fundus image, and Figures 6-2 to 6-5 are diseased fundus images.

Detailed ways

The flow chart of the present invention is as shown in Figure 1, at first for highlighting the video disc, the original image is converted to four color spaces of Lab, Yuv, Yiq, and Hsv; in each space, a single-channel image is selected to carry out PC processing, and by logic "and "Computing enhances the processing results; uses the method of window scanning and grayscale accumulation to locate the optic disc. The specific implementation process of the technical solution of the present invention will be described below in conjunction with the accompanying drawings.

1. Color space channel selection:

Select the a channel of Lab space, the u channel of Yuv space, the i channel of Yiq space and the s channel of Hsv space. As shown in Figure 2, after transforming the original image into these 4 channels, the optic disc is obviously prominent and maintains good edge characteristics, which is beneficial to the subsequent positioning of the optic disc.

2. Process the fundus image based on phase consistency:

The phase consistency function is defined as follows:

$\mathrm{<span\; class="notranslate">\; PC</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; max</span>}}_{\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; \&phi;</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&Element;</span><span\; class="notranslate">\; [</span>\mathrm{<span\; class="notranslate">\; 0,2</span>}\mathrm{<span\; class="notranslate">\; \&pi;</span>}<span\; class="notranslate">\; ]</span>}\frac{{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; A</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}\mathrm{<span\; class="notranslate">\; cos</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; \&phi;</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; \&phi;</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span>}{{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; A</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}}$

为位于该点的所有傅里叶项的局部相位的加权平均。In the formula, A _n is the amplitude of the cosine component of the nth harmonic, φ _n is the phase of the nth frequency component,

is the weighted average of the local phases of all Fourier terms at that point.

The point of maximum phase consistency can be equivalent to the peak in the local energy function. The local energy can be estimated by the following formula:

$\mathrm{<span\; class="notranslate">\; LE</span>}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; E.</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span><span\; class="notranslate">\; =</span>\sqrt{{\mathrm{<span\; class="notranslate">\; f</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; h</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; =</span>\sqrt{{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; *</span>{\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; even</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; *</span>{\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; odd</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}$

For a one-dimensional signal F(x), the local energy is defined as the square root of the sum of the squares of the signal F(x) and its Hilbert transform H(x). The two parts of the local energy can be estimated from the convolution of the signal I(x) with a pair of orthogonal filters, one filter is even symmetric, M _even , and the other is odd symmetric, M _odd .

This patent selects the log Gabor wavelet function and its Hilbert transform as the filter for calculating local energy. On a linear frequency scale, the transfer function of the log Gabor function has the form:

$\mathrm{<span\; class="notranslate">\; g</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&omega;</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; e</span>}}^{\frac{<span\; class="notranslate">\; -</span>{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; log</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&omega;</span>}<span\; class="notranslate">\; /</span>{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}{\mathrm{<span\; class="notranslate">\; 2</span>}{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; log</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&beta;</span>}<span\; class="notranslate">\; /</span>{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}}$

Here ω ₀ is the center frequency of the filter. To keep the shape of the filter constant, β/ω ₀ must be consistent for different center frequencies ω ₀ .

Perform PC processing on the four single-channel images selected above, and the results are shown in Figure 3.

3. Logical "AND" operation:

Perform logic "AND" operation on the PC results of the four single-channel images, and the results are shown in Figure 4.

4. Use two-step positioning for the fundus image after logical "AND" to finally determine the optic disc position:

4.1 Initial positioning: Use a rectangular window of 80×80 to scan the entire image after logical “AND”, and calculate the pixel mean value in the window. The area with the largest pixel mean value is the position of the optic disc, and use the centroid coordinates of the window at this time to locate the optic disc.

4.2 Accurate positioning: In the rectangular window obtained in the previous step, grayscale accumulation is carried out along the X and Y directions respectively, and the maximum grayscale values X _max and Y _max are calculated, and the grayscale threshold in the X direction is set to 0.5*X _max , and the Y direction is The gray threshold is set to 0.5*Y _max , and the image part larger than the threshold is reserved, and the center coordinates of the reserved part are calculated, and the position of the optic disc is finally determined by using the coordinates.

The method of this patent is used to conduct experiments on fundus images collected clinically and in the STARE and DRIVE databases, and the results are shown in Figure 5, Figure 6 and Table 1. The method of the invention has a good effect on the positioning of the optic disc of the fundus image, can align and position the optic disc for normal fundus and different degrees of diseased fundus, and has strong robustness.

Table 1: Using the DRIVE library to perform optic disc positioning experiments, and comparing the positioning success rate with Tobin and Hoover methods.

Claims (4)

1. A fundus image optic disc positioning method based on phase consistency, comprising the following steps: (1) Select a color space single-channel image; (2) Process the selected single-channel fundus image by using the phase consistency function; (3) Carry out logical "AND" operation to the PC result of the single-channel image; (4) select the rectangular window of 80 * 80 to scan the image after the whole logic "AND", calculate the pixel mean value maximum value in the window, and use the centroid coordinates of the window to initially locate the optic disc at this time; (5) In the rectangular window obtained in step (4), carry out grayscale accumulation along the Y direction and calculate the maximum grayscale values X _max , Y _max , set the grayscale threshold in the X direction to 0.5*X _max , and set the grayscale threshold in the Y direction to The gray threshold is set to 0.5*Y _max , and the part of the image larger than the threshold is reserved, and the center coordinates of the reserved part are calculated, and the optic disc is finally precisely positioned using the coordinates. 2. the fundus image optic disc localization method according to claim 1, is characterized in that, in step (1), only selects the color space single-channel image suitable for fundus image optic disc localization, does not carry out any morphological preprocessing to image, guarantees The accuracy of optic disc edge features was improved. 3. The method for locating the optic disc of the fundus image according to claim 1, wherein in step (2), phase consistency is used to identify the features of the optic disc, without any assumptions being made to the waveform, only in the Fourier transform domain according to the phase Find feature points in a consistent order. 4. The method for locating the optic disc of the fundus image according to claim 1, wherein, in steps (4) and (5), the position of the optic disc is determined by using the two positioning methods of initial positioning and then precise positioning to improve the positioning accuracy.

Images