阅读说明：本技术 一种基于U-Net神经网络的眼底视盘分割方法 (Eye fundus optic disk segmentation method based on U-Net neural network ) 是由 曾亚光 郭学东 熊红莲 韩定安 黄铭斌 许祥丛 王陆权 覃楚渝 刘明迪 翁祥涛 于 2021-02-05 设计创作，主要内容包括：本发明提供了本发明一实施例中的一种基于U-Net神经网络的眼底视盘分割方法,包括如下步骤：采集多张原始眼底图像；对多张原始眼底图像进行预处理；对预处理后的多张原始眼底图像的视盘区域进行标记；构建U-Net神经网络,并利用U-Net神经网络对原始眼底图像以及标记好视盘区域的眼底图像进行训练验证以获取U-Net神经网络模型；利用U-Net神经网络模型对眼底图像的视盘区域进行识别分割。本发明能够辅助医生得出准确而有效的杯盘比,辅助医生作出治疗方法,加快了对患者青光眼轻重程度的辨别速度,减少了医生主观因素导致的误诊情况的发生。(The invention provides a fundus optic disk segmentation method based on a U-Net neural network in one embodiment of the invention, which comprises the following steps: collecting a plurality of original fundus images; preprocessing a plurality of original fundus images; marking the optic disc areas of the preprocessed multiple original fundus images; constructing a U-Net neural network, and utilizing the U-Net neural network to train and verify the original fundus image and the fundus image of the marked optic disc area so as to obtain a U-Net neural network model; and identifying and segmenting the optic disc region of the fundus image by using a U-Net neural network model. The invention can assist doctors to obtain accurate and effective cup-to-tray ratio, assist doctors to make treatment methods, accelerate the speed of distinguishing the degree of glaucoma of patients and reduce misdiagnosis caused by subjective factors of doctors.)

1. An eyeground optic disk segmentation method based on a U-Net neural network is characterized by comprising the following steps:

collecting a plurality of original fundus images;

preprocessing a plurality of original fundus images;

marking the optic disc areas of the preprocessed multiple original fundus images;

constructing a U-Net neural network, and utilizing the U-Net neural network to train and verify the original fundus image and the fundus image of the marked optic disc area so as to obtain a U-Net neural network model;

and identifying and segmenting the optic disc region of the fundus image by using the U-Net neural network model.

2. A method as claimed in claim 1, wherein the specific method for preprocessing a plurality of said original fundus images comprises the following steps:

normalizing the original fundus images;

and carrying out filtering processing on the plurality of original fundus images after normalization processing.

3. The fundus optic disk segmentation method based on the U-Net neural network according to claim 2, wherein the specific method for carrying out the filtering processing on the plurality of the original fundus images after the normalization processing comprises the following steps:

performing Gaussian filtering processing on the plurality of original fundus images after normalization processing;

and performing median filtering processing on the original fundus images after the Gaussian filtering processing.

4. The method for segmenting the optic fundus disc based on the U-Net neural network according to claim 3, characterized in that the formula of the normalization processing isWherein, I (x, y) is the original fundus image before normalization processing, L (x, y) and C (x, y) are respectively the illumination intensity drift factor and contrast drift factor of the pixel point (x, y), and I' (x, y) is the original fundus image after normalization processing.

5. A method of segmentation of an ophthalmoscope disc based on a U-Net neural network as claimed in claim 4, wherein the U-Net neural network model comprises an up-sampling network portion and a down-sampling network portion.

6. The method of claim 5, wherein the upsampled network portion and the downsampled network portion each comprise 2 convolutional layers and 1 pooling layer.

7. The method of claim 6, wherein the convolution kernel of the convolutional layer has a pixel size of 3 x 3.

8. The method for segmenting the ophthalmoscope based on the U-Net neural network as claimed in claim 7, wherein the step size of the pooling layer is 2.

9. The method of claim 8, wherein the pixel size of the nucleus of the pooling layer is 2 x 2.

10. A computer-readable storage medium, characterized in that the computer-readable storage medium stores a computer program which, when executed by a processor, implements the U-Net neural network-based fundus optic disk segmentation method according to any one of claims 1 to 9 above.

Technical Field

The invention relates to the technical field of image processing, in particular to a fundus optic disk segmentation method based on a U-Net neural network.

Background

Glaucoma is a common irreversible blinding eye disease. Glaucomatous disease is generally without obvious ocular symptoms, and as the disease progresses, optimal treatment periods are often missed when larger visual field impairments occur. Therefore, the timely diagnosis and treatment of glaucoma are important ways to delay the progression of the disease. Most glaucoma patients suffer from optic nerve damage which is a slowly progressing damage to the optic nerve caused by elevated intraocular pressure due to impaired circulation of aqueous humor within the eyeball, exceeding the intraocular pressure limit to which the optic nerve can tolerate. Glaucoma is mainly manifested by the amplification of the central bright area (optic cup) of the optic disc, the most commonly used clinical diagnostic index is the optic cup optic disc vertical ratio (cup-to-disc ratio for short), and the larger cup-to-disc ratio indicates a larger glaucoma disease risk.

The existing glaucoma diagnosis method generally comprises the steps of acquiring an original fundus image of a patient, judging the approximate cup-to-disc ratio according to the original fundus image by a doctor, and diagnosing the glaucoma of the patient according to the degree of the cup-to-disc ratio.

The diagnosis method has certain subjectivity, depends on medical professional knowledge and experience of an ultrasonic doctor, and is easy to cause misdiagnosis.

Disclosure of Invention

Based on the above, in order to solve the problem that the existing glaucoma diagnosis method has certain subjectivity and is dependent on medical professional knowledge and experience of an ultrasonic doctor, and misdiagnosis is easily caused, the invention provides a fundus optic disc segmentation method based on a U-Net neural network, and the specific technical scheme is as follows:

an eyeground optic disk segmentation method based on a U-Net neural network comprises the following steps:

collecting a plurality of original fundus images;

preprocessing a plurality of original fundus images;

marking the optic disc areas of the preprocessed multiple original fundus images;

constructing a U-Net neural network, and utilizing the U-Net neural network training to train and verify the original fundus image and the fundus image of the marked optic disc area so as to obtain a U-Net neural network model;

and identifying and segmenting the optic disc region of the fundus image by using the U-Net neural network model.

According to the eye fundus optic disk segmentation method based on the U-Net neural network, the optic disk area of the preprocessed original eye fundus image is marked, then the U-Net neural network training is utilized to train and verify the original eye fundus image and the eye fundus image marked with the optic disk area so as to obtain the U-Net neural network model, and finally the U-Net neural network model is utilized to identify and segment the optic disk area of the eye fundus image, so that a doctor can be assisted to obtain an accurate and effective cup-to-disk ratio, the doctor is assisted to make a treatment method, the speed of distinguishing the glaucoma weight degree of a patient is increased, and the occurrence of misdiagnosis caused by subjective factors of the doctor is reduced.

Further, a specific method for preprocessing a plurality of original fundus images includes the steps of:

normalizing the original fundus images;

and carrying out filtering processing on the plurality of original fundus images after normalization processing.

Further, the specific method for performing filtering processing on the plurality of normalized original fundus images includes the following steps:

performing Gaussian filtering processing on the plurality of original fundus images after normalization processing;

and performing median filtering processing on the original fundus images after the Gaussian filtering processing.

Further, the formula of the normalization process isWherein, I (x, y) is the original fundus image before normalization processing, L (x, y) and C (x, y) are respectively the illumination intensity drift factor and contrast drift factor of the pixel point (x, y), and I' (x, y) is the original fundus image after normalization processing.

Further, the U-Net neural network model comprises an up-sampling network part and a down-sampling network part.

Further, the upsampling network portion and the downsampling network portion each include 2 convolutional layers and 1 pooling layer.

Further, the convolution kernel of the convolutional layer has a pixel size of 3 × 3.

Further, the step size of the pooling layer is 2.

Further, the pixel size of the kernel of the pooling layer is 2 × 2.

Accordingly, the present invention also provides a computer-readable storage medium storing a computer program which, when executed by a processor, implements the U-Net neural network-based fundus optic disk segmentation method as described above.

Drawings

The invention will be further understood from the following description in conjunction with the accompanying drawings. The components in the figures are not necessarily to scale, emphasis instead being placed upon illustrating the principles of the embodiments. Like reference numerals designate corresponding parts throughout the different views.

FIG. 1 is a schematic overall flow chart of a fundus optic disk segmentation method based on a U-Net neural network in one embodiment of the present invention;

fig. 2 is a schematic diagram of the effect of the fundus optic disk segmentation method based on the U-Net neural network after identifying and segmenting the optic disk region of the fundus image in the embodiment of the invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail below with reference to embodiments thereof. It should be understood that the detailed description and specific examples, while indicating the scope of the invention, are intended for purposes of illustration only and are not intended to limit the scope of the invention.

It will be understood that when an element is referred to as being "secured to" another element, it can be directly on the other element or intervening elements may also be present. When an element is referred to as being "connected" to another element, it can be directly connected to the other element or intervening elements may also be present. The terms "vertical," "horizontal," "left," "right," and the like as used herein are for illustrative purposes only and do not represent the only embodiments.

Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. As used herein, the term "and/or" includes any and all combinations of one or more of the associated listed items.

The terms "first" and "second" used herein do not denote any particular order or quantity, but rather are used to distinguish one element from another.

As shown in fig. 1, in an embodiment of the present invention, a fundus optic disk segmentation method based on a U-Net neural network includes the following steps:

collecting a plurality of original fundus images;

preprocessing a plurality of original fundus images;

marking the optic disc areas of the preprocessed multiple original fundus images;

constructing a U-Net neural network, and utilizing the U-Net neural network to train and verify an original fundus image and a fundus image of a marked optic disc region to obtain a U-Net neural network model;

and identifying and segmenting the optic disc region of the fundus image by using the U-Net neural network model, as shown in figure 2.

The eye fundus optic disk segmentation method based on the U-Net neural network comprises the steps of marking the optic disk area of an original eye fundus image after preprocessing, then training and verifying the original eye fundus image and the eye fundus image marked with the optic disk area by utilizing the U-Net neural network training to obtain a U-Net neural network model, and finally rapidly and accurately identifying and segmenting the optic disk area of the eye fundus image by utilizing the U-Net neural network model, so that a doctor can be assisted to obtain an accurate and effective cup-to-disk ratio, the doctor is assisted to make a treatment scheme, the speed of distinguishing the glaucoma weight degree of a patient is increased, and misdiagnosis caused by subjective factors of the doctor is reduced.

In addition, the original fundus images and the fundus images marked with the optic disc areas are trained and verified by utilizing the U-Net neural network training to obtain the U-Net neural network model, and the original fundus images are preprocessed, so that the difficulty in identifying, positioning and segmenting the optic disc areas of the fundus images can be effectively reduced.

Specifically, the original fundus image and the fundus image of the marked optic disc area are trained and verified by the U-Net neural network, and then a U-Net convolution neural network model is obtained according to the training and verifying result.

In one embodiment, the raw fundus image is acquired by a fundus camera.

In one embodiment, a specific method for preprocessing a plurality of original fundus images comprises the following steps:

normalizing the original fundus images;

and carrying out filtering processing on the plurality of original fundus images after normalization processing.

The normalization processing is carried out on the original fundus image, so that the optic disc area of the fundus image can be highlighted, the contrast of the original fundus image can be enhanced, and the problems that the identification and positioning and the segmentation precision of the optic disc area of the fundus image are influenced by the U-Net neural network model due to the uneven illumination intensity can be effectively solved.

And filtering the plurality of original fundus images after normalization processing, so that the noise influence is removed, the edge of the fundus image is kept, and the gradient change of the fundus image and the boundary characteristic of an optic disc area are enhanced.

In one embodiment, a specific method for performing filter processing on a plurality of normalized original fundus images includes the steps of:

performing Gaussian filtering processing on the plurality of original fundus images after normalization processing;

and performing median filtering processing on the original fundus images after the Gaussian filtering processing.

Gaussian filtering is adopted to filter Gaussian noise pollution existing in the original fundus image, and median filtering is adopted to filter salt and pepper noise pollution existing in the original fundus image. And performing Gaussian filtering processing and median filtering processing on the plurality of original fundus images after normalization processing, so that Gaussian noise pollution and salt and pepper noise pollution can be filtered simultaneously, and the best noise filtering effect is achieved.

In one embodiment, the normalization process is formulated asWherein I (x, y) is the original fundus image before normalization, L (x, y) and C (x, y) are pixels, respectivelyThe illumination intensity drift factor and the contrast drift factor, I' (x, y), of the point (x, y) are the original fundus image after normalization processing. The illumination intensity drift factor and the contrast drift factor of the pixel point (x, y) in the optic disc region can be obtained through a statistic value of the image gray scale distribution characteristic in a certain neighborhood window, and the illumination intensity drift factor and the contrast drift factor of the pixel point (x, y) in the background can be obtained through a Gaussian filter.

In one embodiment, the U-Net neural network model includes an up-sampling network part and a down-sampling network part, both of which use RelU activation functions, the number of features acquired at each down-sampling is 2 times the number of features acquired at the last down-sampling, and the number of features acquired at each up-sampling is half of the number of features acquired at the last up-sampling.

In one embodiment, the upsampling network portion and the downsampling network portion each include 2 convolutional layers and 1 pooling layer.

In one embodiment, the convolution kernel of the convolutional layer has a pixel size of 3 × 3, the pooling layer has a step size of 2, and the kernel of the pooling layer has a pixel size of 2 × 2.

In one embodiment, the present invention provides a computer-readable storage medium storing a computer program which, when executed by a processor, implements the U-Net neural network-based fundus optic disk segmentation method as described above.

The technical features of the embodiments described above may be arbitrarily combined, and for the sake of brevity, all possible combinations of the technical features in the embodiments described above are not described, but should be considered as being within the scope of the present specification as long as there is no contradiction between the combinations of the technical features.

The above-mentioned embodiments only express several embodiments of the present invention, and the description thereof is more specific and detailed, but not construed as limiting the scope of the invention. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the inventive concept, which falls within the scope of the present invention. Therefore, the protection scope of the present patent shall be subject to the appended claims.