CN111640090A

CN111640090A - Method for evaluating fundus image quality

Info

Publication number: CN111640090A
Application number: CN202010399562.3A
Authority: CN
Inventors: 孔令敏; 陈硕; 佟萌萌; 朱姗姗
Original assignee: Ningbo Lanming Information Technology Co ltd
Current assignee: Ningbo Lanming Information Technology Co ltd
Priority date: 2020-05-12
Filing date: 2020-05-12
Publication date: 2020-09-08
Anticipated expiration: 2040-05-12
Also published as: CN111640090B

Abstract

The invention belongs to the field of medical image processing, and discloses a method for evaluating the quality of fundus images. Preprocessing an acquired fundus image, cutting redundant background around a retina image, and obtaining an area only containing a retina; then, color, focus, contrast, and illumination characteristics are respectively extracted based on the preprocessed fundus images and evaluated; finally, based on the confidence of the above-described feature weighted evaluation, the evaluation result of the fundus image is determined, and the cause of low imaging quality is analyzed. Thus, the method can be used to guide the retinal image acquisition process, helping to reduce the workload of ophthalmologists in screening and improving cost effectiveness.

Description

Method for evaluating fundus image quality

Technical Field

The invention relates to the field of medical image processing, in particular to a method for evaluating the quality of fundus images.

Background

With the development of socioeconomic development and the change of diet and living habits of Chinese people, the incidence of diabetes is on the trend of increasing year by year, and at present, China becomes the world with the largest number of diabetics. Among them, diabetic retinopathy is one of the common diabetic complications, the incidence rate of which is more than 40% in diabetic patients, and diabetic retinopathy has become one of the main causes of visual deterioration and even blindness. However, by early diagnosis and treatment of diabetic retinopathy patients, visual impairment and blindness can be prevented in more than 50% of patients. Fundus screening is therefore critical for the early diagnosis of diabetic retinopathy and the prevention of blindness.

However, the fundus camera sometimes has poor imaging quality due to defocus or insufficient illumination, and further affects subsequent image analysis and clinical diagnosis. Therefore, before further analyzing the fundus image, the fundus image quality needs to be evaluated.

Disclosure of Invention

In view of the above problems, the present invention aims to provide a method for evaluating fundus image quality. The method carries out weighted evaluation on the fundus image quality aiming at the reasons which are easy to cause fundus image quality low, such as color distortion, blurring, low contrast ratio, insufficient or too strong illumination and the like.

In order to achieve the above object, the technical solution of the present invention is as follows:

a method for evaluating the quality of eyeground image, said method comprises carrying on the preconditioning to the examinee's eyeground image that the eyeground camera gathers, cut out the redundant background around the retinal image, obtain the area comprising retina only; then, based on the preprocessed fundus images, color, focus, contrast and illumination features are respectively extracted and evaluated; and finally, performing weighted evaluation on the characteristics, calculating the confidence coefficient of the characteristics, giving an evaluation result of the fundus image and reasons for low imaging quality, and guiding the next fundus image acquisition process.

Drawings

FIG. 1 is a flow chart of the method of the present invention.

Detailed Description

The following detailed description of specific embodiments of the present invention is provided in connection with the accompanying drawings.

A method for evaluating the quality of an image of a fundus of the eye, comprising in particular the steps of:

step 1, preprocessing a fundus image of a detected person acquired by a fundus camera, cutting redundant background around the fundus image, and obtaining a region only containing retina;

in the present embodiment, step 1 converts the retina image into a grayscale image. And obtaining a mask image by using a threshold segmentation algorithm, searching the maximum radius as a circular area in the foreground area by taking the central point of the image as the center of a circle, wherein the circular area comprises all foreground areas, and finally cutting the circular area to obtain a retina image.

Step 2, respectively extracting color, focus, contrast and illumination characteristics based on the preprocessed fundus image and evaluating the color, focus, contrast and illumination characteristics;

in the present embodiment, in the color feature evaluation process in step 2, the cut RGB color image of the retina is converted into HSV color space, and a single threshold T is set on the H channel_cAnd evaluating whether the color of the retina image is distorted. When H channel is larger than threshold value T_cWhen the color of the retina image is not distorted, when the H channel is less than the threshold value T_cThe retinal image color is distorted.

In the focusing characteristic evaluation process in the step 2, firstly, blood vessel and optic disc characteristics of the preprocessed retina image are extracted, the blood vessel and the optic disc are divided, the density of the blood vessel and the optic disc is obtained through calculation, and whether the retina image is focused or not is determined through the density of the blood vessel and the optic disc. Wherein the blood vessel and optic disc density is obtained by the ratio of the segmented blood vessel and optic disc area to the area of the cropped retina image.

In the contrast characteristic evaluation process in the step 2, firstly, the retina image of the standard database is selected as a reference retina image, and the difference value between the reference retina image and the retina image to be tested is calculated to obtain an error retina image. The spatial domain of the error retinal image is then transformed to the frequency domain using a fast fourier transform. Then, contrast sensitivity function weighted modulation is applied to the frequency domain coefficients, and the frequency domain is transformed to a space domain by using fast Fourier transform. Finally, the root mean square error of the spatial error image is calculated, the lower the root mean square error value, the higher the contrast of the retinal image. And setting a threshold value for the root mean square value of the calculated spatial error image, wherein if the threshold value is larger than the threshold value, the retinal image contrast is low, and if the threshold value is smaller than the threshold value, the retinal image contrast is high.

In the illumination characteristic evaluation process in the step 2, the retina RGB color image after pretreatment is converted into HSV color space, and two thresholds T are set on the V channel_H1And T_H2To assess whether the retinal image illumination is uniform. When V channel is less than threshold T_H1When the retina image is not sufficiently exposed, the retina image is not sufficiently exposed; when V channel is larger than threshold value T_H2When the retina image is over-exposed; when V channel is at threshold T_H1And T_H2In between, the retinal image is uniformly illuminated.

Step 3, evaluating the characteristics of color, focus, contrast, illumination and the like according to w respectively₁，w₂，w₃，w₄The weights of (a) and (b) are weighted, and a quality evaluation result of the fundus image and a cause of low quality of the fundus image are given based on the confidence of the feature weighted evaluation.

Claims

1. A method for evaluating the quality of a fundus image, characterized in that features of the fundus image, such as color, focus, contrast and illumination, are weighted and evaluated, and based on the confidence of the weighted evaluation of the features, an evaluation result of the fundus image is given, comprising the steps of:

preprocessing a fundus image of a detected person acquired by a fundus camera, cutting redundant background around the fundus image, and obtaining an area only containing retinas;

secondly, respectively extracting color, focus, contrast and illumination characteristics based on the preprocessed fundus image and evaluating the color, the focus, the contrast and the illumination characteristics;

and step three, performing weighted evaluation on the characteristics, calculating the confidence coefficient of the characteristics and giving an evaluation result of the fundus image.

2. The preprocessing procedure in a method for evaluating quality of a fundus image according to claim 1, wherein said fundus image is converted into a gray scale image, and a mask image is obtained by using a threshold segmentation algorithm, a circular area is found as a maximum radius from a center point of the image in a foreground area, the circular area includes all foreground areas, and finally the retinal image is obtained by cutting out the circular area.

3. The color feature evaluation process in a method for evaluating fundus image quality according to claim 1 or 2, wherein the pre-processed retina RGB color image is converted into HSV color space, and a single threshold is set on an H-channel to evaluate whether the retina image color is distorted; further, the retinal image color is not distorted when the H channel is greater than the threshold, and is distorted when the H channel is less than the threshold.

4. The focus characteristic evaluation process in a method for evaluating fundus image quality according to claim 1 or 2, characterized in that blood vessel and optic disc characteristics of the pre-processed retinal image are first extracted, the blood vessel and optic disc are divided, blood vessel and optic disc densities are calculated, and whether the retinal image is focused or not is judged by the blood vessel and optic disc densities; further, the blood vessel and optic disc densities are the ratio of the segmented blood vessel and optic disc areas to the area of the cropped retina image, respectively.

5. The contrast characteristic evaluation process in a method for evaluating the quality of a fundus image according to claim 1 or 2, wherein a retina image of a standard database is first selected as a reference retina image, a difference between the reference retina image and a retina image to be tested is calculated, and an error retina image is obtained; then transforming the error retina image from a space domain to a frequency domain by using fast Fourier transform; then applying contrast sensitivity function weighted modulation to the frequency domain coefficient, and transforming the frequency domain coefficient to a space domain by using inverse fast Fourier transform to obtain a space error image; finally, calculating the root mean square error of the spatial error image, wherein the lower the root mean square error value is, the higher the contrast of the representative retina image is; furthermore, a threshold value is set for the root mean square value of the spatial error image, the retinal image contrast is low when the threshold value is larger than the threshold value, and the retinal image contrast is high when the threshold value is smaller than the threshold value.

6. The illumination characteristic evaluation process in a method for evaluating fundus image quality according to claim 1 or 2, wherein the pre-processed retina RGB color image is converted into HSV color space, and two thresholds T are set on the V-channel_H1And T_H2To assess whether the retinal image illumination is uniform; further, when the V channel is less than the threshold T_H1When the V channel is larger than the threshold value T, the retina image is underexposed_H2When the retinal image is overexposed, when the V channel is at the threshold value T_H1And T_H2In between, the retinal image is uniformly illuminated.

7. A method for evaluating the quality of a fundus image according to claim 2 or 3, characterized in that the results of the evaluation of the characteristics of its colour, focus, contrast and illumination are according to w respectively₁，w₂，w₃，w₄The weighting is carried out, and finally a global retinal image quality evaluation result is obtained.