WO2020151149A1 - 微动脉瘤自动检测方法、装置及计算机可读存储介质 - Google Patents

微动脉瘤自动检测方法、装置及计算机可读存储介质 Download PDF

Info

Publication number
WO2020151149A1
WO2020151149A1 PCT/CN2019/088639 CN2019088639W WO2020151149A1 WO 2020151149 A1 WO2020151149 A1 WO 2020151149A1 CN 2019088639 W CN2019088639 W CN 2019088639W WO 2020151149 A1 WO2020151149 A1 WO 2020151149A1
Authority
WO
WIPO (PCT)
Prior art keywords
microaneurysm
fundus
image
detection
detection model
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Ceased
Application number
PCT/CN2019/088639
Other languages
English (en)
French (fr)
Inventor
刘莉红
马进
王健宗
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Ping An Technology Shenzhen Co Ltd
Original Assignee
Ping An Technology Shenzhen Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Ping An Technology Shenzhen Co Ltd filed Critical Ping An Technology Shenzhen Co Ltd
Publication of WO2020151149A1 publication Critical patent/WO2020151149A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING OR CALCULATING; COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00Image analysis
    • G06T7/0002Inspection of images, e.g. flaw detection
    • G06T7/0012Biomedical image inspection
    • GPHYSICS
    • G06COMPUTING OR CALCULATING; COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/20Special algorithmic details
    • G06T2207/20081Training; Learning
    • GPHYSICS
    • G06COMPUTING OR CALCULATING; COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/20Special algorithmic details
    • G06T2207/20084Artificial neural networks [ANN]
    • GPHYSICS
    • G06COMPUTING OR CALCULATING; COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/30Subject of image; Context of image processing
    • G06T2207/30004Biomedical image processing
    • G06T2207/30041Eye; Retina; Ophthalmic
    • GPHYSICS
    • G06COMPUTING OR CALCULATING; COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/30Subject of image; Context of image processing
    • G06T2207/30004Biomedical image processing
    • G06T2207/30101Blood vessel; Artery; Vein; Vascular

Definitions

  • This application relates to the field of artificial intelligence technology, and in particular to an automatic detection method, device and computer-readable storage medium for microaneurysms.
  • diabetic retinopathy is one of the main causes of visual impairment and blindness in the world. Diabetic patients are 25 times more likely to cause blindness than non-diabetic patients. Microaneurysm is an early sign of diabetic retinopathy. If it can be accurately detected, it can effectively evaluate and inhibit the development of the disease.
  • top-hat transform is the most common mathematical morphology operation used to detect microaneurysms. It can get suspected microaneurysms by eliminating all blood vessels on the retinal image.
  • the mathematical morphology method needs to increase the length of structural elements, but this will cause the deterioration of blood vessel segmentation, so that more stray suspected microaneurysms will be detected on the blood vessels.
  • the filter side constructs a suitable kernel function to match the microaneurysm.
  • This application provides a method, device and computer-readable storage medium for automatic detection of microaneurysms, the main purpose of which is to provide a solution for automatic detection of microaneurysms.
  • the automatic detection of microaneurysms described in this application includes:
  • the present application also provides a device that includes a memory and a processor.
  • the memory stores an automatic microaneurysm detection program that can be run on the processor. The following steps are implemented when the automatic detection program is executed by the processor:
  • this application also provides a computer-readable storage medium that stores an automatic microaneurysm detection program, and the microaneurysm automatic detection program can be used by one or more The processor executes to implement the steps of the method for automatically detecting microaneurysms as described above.
  • the microaneurysm automatic detection method, device, and computer-readable storage medium proposed in the present application collect fundus image data of the eyeball, and perform data enhancement processing operations on the collected fundus image data; detect retinal images in the fundus image data, Making training samples of the fundus microaneurysm detection model; using the training samples obtained above to perform the training of the fundus microaneurysm detection model; and using the fundus microaneurysm detection model to perform automatic microaneurysm detection. Therefore, this application can realize automatic detection of microaneurysms.
  • FIG. 1 is a schematic flowchart of an automatic detection method for microaneurysm provided by an embodiment of the application
  • FIG. 2 is a schematic diagram of the internal structure of a device provided by an embodiment of the application.
  • FIG. 3 is a schematic diagram of modules of an automatic detection program for microaneurysms in a device provided by an embodiment of the application.
  • This application provides an automatic detection method for microaneurysms.
  • FIG. 1 it is a schematic flowchart of an automatic detection method for microaneurysm provided by an embodiment of this application.
  • the method can be executed by a device, and the device can be implemented by software and/or hardware.
  • S1 Collect fundus image data of the eyeball, and perform a data enhancement processing operation on the collected fundus image data.
  • the preferred embodiment of the present application uses a digital fundus camera (such as Kowa VX-10 ⁇ ) with a 50-degree field of view (FOV) to collect fundus images of the eyeball, and the positive and negative samples collected are 1:1, that is, the fundus microaneurysms are collected separately
  • the fundus image of the human eye and the fundus image of the healthy human eye all fundus images must be centered and close to the macula. If the eyeball image collected by the digital fundus camera is not centered and not close to the macula, it needs to be collected again.
  • the resolution of the fundus image collected by the digital fundus camera in this application is 4288 ⁇ 2848 pixels, and it is stored in the computer in a jpg file format.
  • the number of fundus images may be one hundred.
  • this application needs to perform enhancement processing on the collected fundus image data of the eyeball.
  • the enhancement processing includes enhancing the pixel values of human eye microaneurysms and nearby areas.
  • n k is the number of pixels with r k -level grayscale in the image
  • n is the total number of image pixels
  • microaneurysms mainly exist in the position of the retina
  • a preferred embodiment of the present application performs morphological calculations on the fundus image data obtained after the enhancement process to obtain the boundary of the retina.
  • the morphological operations include:
  • represents the corrosion operation
  • represents the expansion operation
  • the deep learning model mainly checks whether the picture has microaneurysms, and if so, marks the location of the microaneurysms.
  • the training method of the deep learning model mainly includes the following steps:
  • the up and down sampling method samples the training samples to obtain a data set. Subsampled is to reduce the image, and upsampling is to enlarge the image. Up and down sampling is to improve the model's ability to detect microaneurysms in different environments. For a microaneurysm image with a size of M*N, it is down-sampled by s times to obtain a resolution image of (M/s)*(N/s) size, where s is the common divisor of M and N. Since the microaneurysm image is in matrix form, the original image needs to be processed according to the following formula:
  • ⁇ Ii represents the sum of pixels in the microaneurysm picture.
  • the principle of upsampling is similar to that of downsampling, except that the interpolation method is used, that is, on the basis of the original microaneurysm pixels, the interpolation algorithm is used to insert new pixels.
  • the deep learning model described in this application is a convolutional neural network model.
  • This application uses Fast RCNN (Fast Region-based Convolutional Neural Networks) algorithm to detect microaneurysms.
  • Fast RCNN is an algorithm that applies the convolutional neural network model to the target detection problem. It implements the conversion of the target detection problem through the multi-task loss function border regression method.
  • the training of the deep learning model using the up and down sampled data set includes:
  • the convolutional neural network model used in this application is VGG (Visual Geometry Group).
  • the convolution kernel of VGG is smaller and has a 3x3 specification, because multiple convolutional layers with smaller convolution kernels are used instead of a larger convolution kernel.
  • the convolutional layer on the one hand, can reduce parameters, on the other hand, it is equivalent to more nonlinear mapping, which can increase the feature extraction ability of the network for microaneurysms, and the convolutional layer step size is set to 1.
  • Each layer of convolution kernel is followed by a pooling layer.
  • this application only uses the convolutional layer and pooling layer of the first 13 layers of the VGG model to extract features, and discards the subsequent fully connected layer.
  • the feature vector of the microaneurysm output in the previous step use the softmax classification for classification and judgment. If there is no feature vector of the microaneurysm in the picture, it is judged as a normal eye. When the feature of the microaneurysm is detected, it is marked as having micro Picture of an aneurysm.
  • the accurate microaneurysm area is obtained by using the bounding box regression method of the multi-task loss function.
  • the loss function is as follows:
  • L represents the loss function between the predicted microaneurysm area and the actual microaneurysm area.
  • y i,j represents the pixel location of the microaneurysm area pre-judged by the data set
  • y′ i,j represents the actual location of the microaneurysm area
  • xi,j represents the pixel value of the microaneurysm coordinate at (i,j)
  • w 10 and w 11 are used to adjust the parameters of sample imbalance after up and down sampling.
  • this application applies the model to the automatic detection of microaneurysm.
  • the application generates image blocks uniformly with 32 steps for the fundus image, and uses the fundus microaneurysm detection model for each image block to obtain the probability that the image block may be a microaneurysm, and finally calculates the probability Distribution map, determine whether there is microaneurysm in the fundus, and complete the automatic detection process.
  • This application also provides a device for performing automatic detection of microaneurysms.
  • FIG. 2 it is a schematic diagram of the internal structure of a device provided by an embodiment of this application.
  • the device 1 may be a terminal device such as a smart phone, a tablet computer, a portable computer, etc., a PC (Personal Computer, personal computer), or a server, a server group, and the like.
  • the device 1 at least includes a memory 11, a processor 12, a communication bus 13, and a network interface 14.
  • the memory 11 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (for example, SD or DX memory, etc.), magnetic memory, magnetic disk, optical disk, and the like.
  • the memory 11 may be an internal storage unit of the device 1 in some embodiments, such as a hard disk of the device 1. In other embodiments, the memory 11 may also be an external storage device of the device 1, such as a plug-in hard disk, a smart media card (SMC), and a secure digital (SD) card equipped on the device 1. Flash Card, etc. Further, the memory 11 may also include both an internal storage unit of the apparatus 1 and an external storage device.
  • the memory 11 can be used not only to store application software and various data installed in the device 1, such as the code of the microaneurysm automatic detection program 01, etc., but also to temporarily store data that has been output or will be output.
  • the processor 12 may be a central processing unit (CPU), controller, microcontroller, microprocessor, or other data processing chip in some embodiments, and is used to run the program code or processing stored in the memory 11 Data, such as the implementation of the microaneurysm automatic detection program 01, etc.
  • CPU central processing unit
  • controller microcontroller
  • microprocessor or other data processing chip in some embodiments, and is used to run the program code or processing stored in the memory 11 Data, such as the implementation of the microaneurysm automatic detection program 01, etc.
  • the communication bus 13 is used to realize the connection and communication between these components.
  • the network interface 14 may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface), and is usually used to establish a communication connection between the device 1 and other electronic devices.
  • the device 1 may further include a user interface.
  • the user interface may include a display (Display) and an input unit such as a keyboard (Keyboard).
  • the optional user interface may also include a standard wired interface and a wireless interface.
  • the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, an OLED (Organic Light-Emitting Diode, organic light-emitting diode) touch device, etc.
  • the display can also be appropriately called a display screen or a display unit, which is used to display the information processed in the device 1 and to display a visualized user interface.
  • Figure 2 only shows the device 1 with components 11-14 and the microaneurysm automatic detection program 01.
  • Figure 1 does not constitute a limitation on the device 1, and may include Fewer or more parts are shown, or some parts are combined, or different parts are arranged.
  • the microaneurysm automatic detection program 01 is stored in the memory 11; when the processor 12 executes the microaneurysm automatic detection program 01 stored in the memory 11, the following steps are implemented:
  • Step 1 Collect fundus image data of the eyeball, and perform a data enhancement processing operation on the collected fundus image data.
  • the preferred embodiment of the present application uses a digital fundus camera (such as Kowa VX-10 ⁇ ) with a 50-degree field of view (FOV) to collect fundus images of the eyeball, and the positive and negative samples collected are 1:1, that is, the fundus microaneurysms are collected separately
  • the fundus image of the human eye and the fundus image of the healthy human eye all fundus images must be centered and close to the macula. If the eyeball image collected by the digital fundus camera is not centered and not close to the macula, it needs to be collected again.
  • the resolution of the fundus image collected by the digital fundus camera in this application is 4288 ⁇ 2848 pixels, and it is stored in the computer in a jpg file format.
  • the number of fundus images may be one hundred.
  • this application needs to perform enhancement processing on the collected fundus image data of the eyeball.
  • the enhancement processing includes enhancing the pixel values of human eye microaneurysms and nearby areas.
  • n k is the number of pixels with r k -level grayscale in the image
  • n is the total number of image pixels
  • Step 2 Detect the retinal image in the fundus image data, and make a training sample of the fundus microaneurysm detection model.
  • microaneurysms mainly exist in the position of the retina
  • a preferred embodiment of the present application performs morphological calculations on the fundus image data obtained after the enhancement process to obtain the boundary of the retina.
  • the morphological operations include:
  • represents the corrosion operation
  • represents the expansion operation
  • Step 3 Use the training samples obtained above to train the fundus microaneurysm detection model.
  • the deep learning model mainly checks whether the picture has microaneurysms, and if so, marks the location of the microaneurysms.
  • the training method of the deep learning model mainly includes the following steps:
  • the up and down sampling method samples the training samples to obtain a data set. Subsampled is to reduce the image, and upsampling is to enlarge the image. Up and down sampling is to improve the model's ability to detect microaneurysms in different environments. For a microaneurysm image with a size of M*N, down-sample it by s times to obtain a resolution image of (M/s)*(N/s) size, where s is the common divisor of M and N. Since the microaneurysm image is in matrix form, the original image needs to be processed according to the following formula:
  • ⁇ Ii represents the sum of pixels in the microaneurysm picture.
  • the principle of upsampling is similar to that of downsampling, except that the interpolation method is used, that is, on the basis of the original microaneurysm pixels, the interpolation algorithm is used to insert new pixels.
  • the deep learning model described in this application is a convolutional neural network model.
  • This application uses Fast RCNN (Fast Region-based Convolutional Neural Networks) algorithm to detect microaneurysms.
  • Fast RCNN is an algorithm that applies the convolutional neural network model to the target detection problem. It implements the conversion of the target detection problem through the multi-task loss function border regression method.
  • the training of the deep learning model using the up and down sampled data set includes:
  • the convolutional neural network model used in this application is VGG (Visual Geometry Group).
  • the convolution kernel of VGG is smaller and has a 3x3 specification, because multiple convolutional layers with smaller convolution kernels are used instead of a larger convolution kernel.
  • the convolutional layer on the one hand, can reduce parameters, on the other hand, it is equivalent to more nonlinear mapping, which can increase the feature extraction ability of the network for microaneurysms, and the convolutional layer step size is set to 1.
  • Each layer of convolution kernel is followed by a pooling layer.
  • this application only uses the convolutional layer and pooling layer of the first 13 layers of the VGG model to extract features, and discards the subsequent fully connected layer.
  • the feature vector of the microaneurysm output in the previous step use the softmax classification for classification and judgment. If there is no feature vector of the microaneurysm in the picture, it is judged as a normal eye. When the feature of the microaneurysm is detected, it is marked as having micro Picture of an aneurysm.
  • the accurate microaneurysm area is obtained by using the bounding box regression method of the multi-task loss function.
  • the loss function is as follows:
  • L represents the loss function between the predicted microaneurysm area and the actual microaneurysm area.
  • y i,j represents the pixel position of the microaneurysm area pre-judged by the data set
  • y′ i,j represents the actual microaneurysm area location
  • xi,j represents the pixel value of the microaneurysm coordinate at (i,j)
  • w 10 and w 11 are used to adjust the parameters of sample imbalance after up and down sampling.
  • Step 4 Use the fundus microaneurysm detection model to perform automatic microaneurysm detection.
  • this application applies the model to the automatic detection of microaneurysm.
  • the application generates image blocks uniformly with 32 steps for the fundus image, and uses the fundus microaneurysm detection model for each image block to obtain the probability that the image block may be a microaneurysm, and finally calculates the probability Distribution map, determine whether there is microaneurysm in the fundus, and complete the automatic detection process.
  • the microaneurysm automatic detection program 01 can also be divided into one or more modules, and one or more modules are stored in the memory 11 and processed by one or more modules.
  • This application is executed by the processor (the processor 12 in this embodiment).
  • the module referred to in this application refers to a series of computer program instruction segments that can complete specific functions, and is used to describe the microaneurysm automatic detection program in the device In the implementation process.
  • the microaneurysm automatic detection program 01 can be divided into a data acquisition module 10, a sample The data production module 20, the model training module 30, and the microaneurysm detection module 40.
  • a data acquisition module 10 a sample The data production module 20
  • the model training module 30 the model training module 30
  • the microaneurysm detection module 40 Illustratively:
  • the data collection module 10 is used to collect fundus image data of the eyeball, and perform data enhancement processing operations on the collected fundus image data.
  • the execution of the data enhancement processing operation described in this application includes performing two adaptive histogram equalization preprocessing on the fundus image data, including:
  • n k is the number of pixels with r k -level grayscale in the image
  • n is the total number of image pixels
  • the graph of the relationship between r k and P r (r k ) in the rectangular coordinate system is called the histogram of the image.
  • the sample data preparation module 20 is used for detecting the retinal image in the fundus image data, and preparing training samples of the fundus microaneurysm detection model.
  • the detecting the retinal image in the fundus image data and making a training sample of the fundus microaneurysm detection model includes:
  • morphological operations on the fundus image data obtained after data enhancement processing to obtain the retinal boundary, wherein the morphological operations include:
  • represents the corrosion operation
  • represents the expansion operation
  • the morphological operation is used to continue to detect the retinal boundary until the threshold ⁇ is met.
  • the model training module 30 is used to perform the training of the fundus microaneurysm detection model using the training samples obtained above.
  • the method for training the fundus microaneurysm detection model includes:
  • Up and down sampling method samples the training samples to obtain a data set
  • Use the up-sampling data set to train the deep learning model including:
  • the softmax classification is used for classification and judgment. If there is no microaneurysm feature vector in the picture, it is judged as a normal eye. When the feature of microaneurysm is detected, it is marked as having microaneurysm Characteristic picture
  • the accurate microaneurysm area is obtained by the bounding box regression method of the multi-task loss function.
  • the microaneurysm detection module 40 is used to perform automatic detection of microaneurysms using the fundus microaneurysm detection model.
  • the using the fundus microaneurysm detection model to perform automatic detection of microaneurysms includes:
  • image blocks are uniformly generated in 32 steps, and the fundus microaneurysm detection model is used for each image block to obtain the probability that the image block may be a microaneurysm, and the probability distribution map is calculated to determine whether the fundus is micro Aneurysm, complete the automatic detection process.
  • an embodiment of the present application also proposes a computer-readable storage medium, the computer-readable storage medium stores a microaneurysm automatic detection program, and the microaneurysm automatic detection program can be executed by one or more processors To achieve the following operations:

Landscapes

  • Engineering & Computer Science (AREA)
  • Quality & Reliability (AREA)
  • General Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Radiology & Medical Imaging (AREA)
  • Health & Medical Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Eye Examination Apparatus (AREA)
  • Image Processing (AREA)
  • Image Analysis (AREA)

Abstract

本申请涉及一种人工智能技术,揭露了一种微动脉瘤自动检测方法,该方法包括:采集眼球的眼底图像数据,并对所采集的眼底图像数据执行数据增强处理操作;检测所述眼底图像数据中的视网膜图像,制作眼底微动脉瘤检测模型的训练样本;利用上述得到的训练样本执行眼底微动脉瘤检测模型的训练;及利用所述眼底微动脉瘤检测模型执行微动脉瘤自动检测。本申请还提出一种装置以及一种计算机可读存储介质。本申请可以实现微动脉瘤自动检测。

Description

微动脉瘤自动检测方法、装置及计算机可读存储介质
本申请要求于2019年1月23日提交中国专利局,申请号为201910063699.9、发明名称为“微动脉瘤自动检测方法、装置及计算机可读存储介质”的中国专利申请的优先权,其全部内容通过引用结合在本申请中。
技术领域
本申请涉及人工智能技术领域,尤其涉及一种微动脉瘤自动检测方法、装置及计算机可读存储介质。
背景技术
由国际糖尿病联盟的统计数据显示,全球糖尿病患者已达到3.66亿,中国糖尿病患者,已超过9240万。伴随着对糖尿病研究的不断深入,医学家们发现,由糖尿病引发的眼部并发症(如糖尿病性视网膜病变)的发病率正与日俱增。世界卫生组织的一项调查研究表明,糖尿病视网膜病变是全世界导致视力障碍和失明的主要原因之一,糖尿病患者致盲的可能性是非糖尿病患者的25倍。微动脉瘤是糖尿病视网膜病变的早期标志,如果能对其准确检测,即能够有效评估和抑制病情的发展。
目前,国内外学者针对微动脉瘤的检测提出了多种方法,大致可以分为3类:数学形态学法、滤波器法和监督学习法。Top-hat变换是用于检测微动脉瘤最常见的数学形态学运算,通过消除视网膜图像上所有血管来得到疑似微动脉瘤。但是,为了检测到较大的微动脉瘤,数学形态学法需要增加结构元素的长度,不过这会导致血管分割劣化,以至于在血管上会检测到更多的杂散疑似微动脉瘤。滤波器方通过构建一个合适的核函数来匹配微动脉瘤,如果产生强响应便得到疑似微动脉瘤。常用的滤波器包括多尺度高斯核函数滤波器和高通滤波器等。滤波器法的局限性在于需要设计合适的滤波器匹配微动脉瘤。监督学习法中神经网络和学习字典用于训练分类器,但当样本量比较大时需要较长的训练时间,因此,监督学习法基本上不可行。综合近年来有关视网膜微动脉瘤检测的文献和研究报道,可以看出当前的研究虽然取得 了很重要的进展,但是视网膜结构特征之间会相互影响,导致独立分割微动脉瘤易产生假阳性。基于以上所述,本申请提出了基于深度学习的微动脉瘤自动检测系统。
发明内容
本申请提供一种微动脉瘤自动检测方法、装置及计算机可读存储介质,其主要目的在于提供一种实现微动脉瘤自动检测方案。
本申请所述微动脉瘤自动检测包括:
采集眼球的眼底图像数据,并对所采集的眼底图像数据执行数据增强处理操作;
检测所述眼底图像数据中的视网膜图像,制作眼底微动脉瘤检测模型的训练样本;
利用上述得到的训练样本执行眼底微动脉瘤检测模型的训练;及
利用所述眼底微动脉瘤检测模型执行微动脉瘤自动检测。
此外,为实现上述目的,本申请还提供一种装置,该装置包括存储器和处理器,所述存储器中存储有可在所述处理器上运行的微动脉瘤自动检测程序,所述微动脉瘤自动检测程序被所述处理器执行时实现如下步骤:
采集眼球的眼底图像数据,并对所采集的眼底图像数据执行数据增强处理操作;
检测所述眼底图像数据中的视网膜图像,制作眼底微动脉瘤检测模型的训练样本;
利用上述得到的训练样本执行眼底微动脉瘤检测模型的训练;及
利用所述眼底微动脉瘤检测模型执行微动脉瘤自动检测。
此外,为实现上述目的,本申请还提供一种计算机可读存储介质,所述计算机可读存储介质上存储有微动脉瘤自动检测程序,所述微动脉瘤自动检测程序可被一个或者多个处理器执行,以实现如上所述的微动脉瘤自动检测方法的步骤。
本申请提出的微动脉瘤自动检测方法、装置及计算机可读存储介质采集眼球的眼底图像数据,并对所采集的眼底图像数据执行数据增强处理操作;检测所述眼底图像数据中的视网膜图像,制作眼底微动脉瘤检测模型的训练 样本;利用上述得到的训练样本执行眼底微动脉瘤检测模型的训练;及利用所述眼底微动脉瘤检测模型执行微动脉瘤自动检测。因此,本申请可以实现微动脉瘤自动检测。
附图说明
图1为本申请一实施例提供的微动脉瘤自动检测方法的流程示意图;
图2为本申请一实施例提供的装置的内部结构示意图;
图3为本申请一实施例提供的装置中微动脉瘤自动检测程序的模块示意图。
本申请目的的实现、功能特点及优点将结合实施例,参照附图做进一步说明。
具体实施方式
为了使本申请的目的、技术方案及优点更加清楚明白,以下结合附图及实施例,对本申请进行进一步详细说明。应当理解,此处所描述的具体实施例仅用以解释本申请,并不用于限定本申请。基于本申请中的实施例,本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例,都属于本申请保护的范围。
本申请的说明书和权利要求书及上述附图中的术语“第一”、“第二”、“第三”、“第四”等(如果存在)是用于区别类似的对象,而不必用于描述特定的顺序或先后次序。应该理解这样使用的数据在适当情况下可以互换,以便这里描述的实施例能够以除了在这里图示或描述的内容以外的顺序实施。此外,所述“第一”、“第二”等的描述仅用于描述目的,而不能理解为指示或暗示其相对重要性或者隐含指明所指示的技术特征的数量。由此,限定有“第一”、“第二”的特征可以明示或者隐含地包括至少一个该特征。
进一步地,术语“包括”和“具有”以及他们的任何变形,意图在于覆盖不排他的包含,例如,包含了一系列步骤或单元的过程、方法、系统、产品或设 备不必限于清楚地列出的那些步骤或单元,而是可包括没有清楚地列出的或对于这些过程、方法、产品或设备固有的其它步骤或单元。
另外,各个实施例之间的技术方案可以相互结合,但是必须是以本领域普通技术人员能够实现为基础,当技术方案的结合出现相互矛盾或无法实现时应当认为这种技术方案的结合不存在,也不在本申请要求的保护范围之内。
本申请提供一种微动脉瘤自动检测方法。
详细地,参照图1所示,为本申请一实施例提供的微动脉瘤自动检测方法的流程示意图。该方法可以由一个装置执行,该装置可以由软件和/或硬件实现。
S1、采集眼球的眼底图像数据,并对所采集的眼底图像数据执行数据增强处理操作。
本申请较佳实施例使用50度视场(FOV)的数字眼底相机(如Kowa VX-10α)采集眼球的眼底图像,采集的正负样本为1:1,即分别采集有眼底微动脉瘤的人眼的眼底图像和健康人眼的眼底图像,所有眼底图像都要求居中且靠近黄斑。如果数字眼底相机采集的眼球图片非居中且不靠近黄斑,需重新采集。较佳地,本申请使用数字眼底相机采集的眼底图像的分辨率为4288×2848像素,且以jpg文件格式存储在计算机中。本申请较佳实施例中,眼底图像的数量可以是一百张。
为了有效检测人眼视网膜微动脉瘤,为后期模型训练提供更优质的数据,本申请需要对所采集的眼球的眼底图像数据进行增强处理。
优选地,所述增强处理包括增强人眼微动脉瘤及附近区域的像素值。
人眼微动脉瘤是本申请所要检测的图像部分,为了增强该图像部分和附近区域的像素值,本申请首先对眼球的眼底图像数据进行两次自适应的直方图均衡化预处理。其中,直方图均衡化的实现原理如下:假设r代表某幅给定眼底图像中像素灰度级,作归一化处理后,r被限定为[0,1]范围内,令p r(w)为图像的灰度级的概率密度函数,对灰度级执行下式所示的变换,得到输出灰度级S:
Figure PCTCN2019088639-appb-000001
其中,w为积分的虚拟变量。但在数字图像中,灰度级是离散的。因此对于离散的灰度级,可将上式变换为下式:
Figure PCTCN2019088639-appb-000002
式中n k是图像中出现r k级灰度的像素数,n是图像像素总数,
Figure PCTCN2019088639-appb-000003
为频数。在直角坐标系中做出r k与P r(r k)的关系图形,即称为图像的直方图。由于自适应直方图均衡化是在直方图均衡化原理的基础上采用局部法对图像进行处理,所以,自适应直方图均衡化可以起到增强人眼微动脉瘤及附近区域的像素亮度值的作用。
S2、检测所述眼底图像数据中的视网膜图像,制作眼底微动脉瘤检测模型的训练样本。
人眼的构造精密复杂,如何正确的捕捉到微动脉瘤部分至关重要。由于微动脉瘤主要存在于视网膜位置,本申请较佳实施例对上述得到的增强处理后的眼底图像数据,进行形态学运算,得到视网膜边界。
所述形态学运算包括:
使用Canny算子检测视网膜图像的边缘,使用一个顶栏和底栏闭合的圆形区域,并填充该区域和空洞后进行下式的形态学开运算操作:
AB=(AΘB)θB
其中,Θ表示腐蚀操作,θ表示膨胀操作。利用形态学开运算后,分别得到腐蚀图像A和膨胀图像B,最后腐蚀图像A减去膨胀图像B,即得到视网膜的圆形边界。
检测到视网膜边界后,计算并判断视网膜边界的面积,若边界面积大于设定阈值δ,表明边界检测错误,重新采用形态学运算继续检测视网膜边界,直到满足阈值δ为止。
S3、利用上述得到的训练样本执行眼底微动脉瘤检测模型的训练。
对上述制作的训练样本进行上采样和下采样,然后分别放入一个深度学习模型进行训练。在整个训练过程中,深度学习模型主要检查图片是否有微动脉瘤,若有则标记微动脉瘤的位置。
详细地,所述深度学习模型的训练的方法主要包括如下步骤:
1、上下采样法对训练样本进行采样,得到数据集。下采样(subsampled)就是缩小图像,上采样(upsampling)就是放大图像。上下采样是为了提高模型在不同环境下,对微动脉瘤病症的侦测能力。对于一张尺寸为M*N微动脉 瘤图像,对其进行s倍下采样,即得到(M/s)*(N/s)尺寸的分辨率图像,s是M和N的公约数。由于微动脉瘤图片是矩阵形式,还需把原始图像按照如下公式,进行像素的均值处理:
P k=ΣI i/s 2
其中,ΣIi代表微动脉瘤图片的像素总和。上采样原理与下采样类似,只不过采用内插值方法,即在原有微动脉瘤像素的基础上,采用内插值算法插入新的像素。
2、利用上下采样后的数据集训练深度学习模型。本申请所述深度学习模型是卷积神经网络模型。本申请使用Fast RCNN(Fast Region-based ConvolutionalNeural Networks)算法对微动脉瘤进行检测。Fast RCNN是将卷积神经网络模型应用到目标检测问题上的算法,其通过多任务损失函数边框回归方法实现目标检测问题的转化。
所述利用上下采样后的数据集训练深度学习模型包括:
A、利用卷积神经网络提取眼底微动脉瘤病灶的特征向量。本申请使用的卷积神经网络模型是VGG(Visual Geometry Group),VGG的卷积核较小,是3x3规格,因为使用多个较小卷积核的卷积层代替一个卷积核较大的卷积层,一方面可以减少参数,另一方面相当于是进行了更多的非线性映射,可以增加网络的对微动脉瘤的特征提取能力,卷积层步长设置为1。每层卷积核后紧跟池化层。但本申请只使用VGG模型前13层的卷积层和池化层是提取特征,舍去后面的全连层。
B、针对上一步输出的微动脉瘤特征向量,利用softmax分类进行分类判别,若图片中无微动脉瘤特征向量,则判别为正常眼睛,当检查到微动脉瘤的特征,则标记为有微动脉瘤的图片。
C、对于含有微动脉瘤特征的图片,通过多任务损失函数的边框回归方法得到准确的微动脉瘤区域,其中损失函数如下:
Figure PCTCN2019088639-appb-000004
L表示预判断的微动脉瘤区域与实际微动脉瘤区域的损失函数。其中y i,j表示数据集预判断的微动脉瘤区域的像素位置,y′ i,j表示实际微动脉瘤区域位置, x i,j表示微动脉瘤坐标在(i,j)的像素值,微动瘤特征图片共n个像素,w 10和w 11用来调节上下采样后样本不平衡的参数。
S4、利用所述眼底微动脉瘤检测模型执行微动脉瘤自动检测。
眼底微动脉瘤检测模型训练好后,本申请将该模型应用于微动脉瘤自动检测中。在检测过程中,本申请对眼底图像以32步长均匀地产生图像块,对每个图像块运用所述眼底微动脉瘤检测模型得到该图像块可能是微动脉瘤的概率,最后统计出概率分布图,判断眼底是否有微动脉瘤,完成自动检测过程。
本申请还提供一种执行微动脉瘤自动检测的装置。参照图2所示,为本申请一实施例提供的装置的内部结构示意图。
在本实施例中,所述装置1可以是智能手机、平板电脑、便携计算机等终端设备,可以是PC(Personal Computer,个人电脑),也可以是服务器、服务器群组等。该装置1至少包括存储器11、处理器12,通信总线13,以及网络接口14。
其中,存储器11至少包括一种类型的可读存储介质,所述可读存储介质包括闪存、硬盘、多媒体卡、卡型存储器(例如,SD或DX存储器等)、磁性存储器、磁盘、光盘等。存储器11在一些实施例中可以是装置1的内部存储单元,例如该装置1的硬盘。存储器11在另一些实施例中也可以是装置1的外部存储设备,例如装置1上配备的插接式硬盘,智能存储卡(Smart Media Card,SMC),安全数字(Secure Digital,SD)卡,闪存卡(Flash Card)等。进一步地,存储器11还可以既包括装置1的内部存储单元也包括外部存储设备。存储器11不仅可以用于存储安装于装置1的应用软件及各类数据,例如微动脉瘤自动检测程序01的代码等,还可以用于暂时地存储已经输出或者将要输出的数据。
处理器12在一些实施例中可以是一中央处理器(Central Processing Unit,CPU)、控制器、微控制器、微处理器或其他数据处理芯片,用于运行存储器11中存储的程序代码或处理数据,例如执行微动脉瘤自动检测程序01等。
通信总线13用于实现这些组件之间的连接通信。
网络接口14可选的可以包括标准的有线接口、无线接口(如WI-FI接口),通常用于在该装置1与其他电子设备之间建立通信连接。
可选地,该装置1还可以包括用户接口,用户接口可以包括显示器(Display)、输入单元比如键盘(Keyboard),可选的用户接口还可以包括标准的有线接口、无线接口。可选地,在一些实施例中,显示器可以是LED显示器、液晶显示器、触控式液晶显示器以及OLED(Organic Light-Emitting Diode,有机发光二极管)触摸器等。其中,显示器也可以适当的称为显示屏或显示单元,用于显示在装置1中处理的信息以及用于显示可视化的用户界面。
图2仅示出了具有组件11-14以及微动脉瘤自动检测程序01的装置1,本领域技术人员可以理解的是,图1示出的结构并不构成对装置1的限定,可以包括比图示更少或者更多的部件,或者组合某些部件,或者不同的部件布置。
在图2所示的装置1实施例中,存储器11中存储有微动脉瘤自动检测程序01;处理器12执行存储器11中存储的微动脉瘤自动检测程序01时实现如下步骤:
步骤一、采集眼球的眼底图像数据,并对所采集的眼底图像数据执行数据增强处理操作。
本申请较佳实施例使用50度视场(FOV)的数字眼底相机(如Kowa VX-10α)采集眼球的眼底图像,采集的正负样本为1:1,即分别采集有眼底微动脉瘤的人眼的眼底图像和健康人眼的眼底图像,所有眼底图像都要求居中且靠近黄斑。如果数字眼底相机采集的眼球图片非居中且不靠近黄斑,需重新采集。较佳地,本申请使用数字眼底相机采集的眼底图像的分辨率为4288×2848像素,且以jpg文件格式存储在计算机中。本申请较佳实施例中,眼底图像的数量可以是一百张。
为了有效检测人眼视网膜微动脉瘤,为后期模型训练提供更优质的数据,本申请需要对所采集的眼球的眼底图像数据进行增强处理。
优选地,所述增强处理包括增强人眼微动脉瘤及附近区域的像素值。
人眼微动脉瘤是本申请所要检测的图像部分,为了增强该图像部分和附近区域的像素值,本申请首先对眼球的眼底图像数据进行两次自适应的直方 图均衡化预处理。其中,直方图均衡化的实现原理如下:假设r代表某幅给定眼底图像中像素灰度级,作归一化处理后,r被限定为[0,1]范围内,令p r(w)为图像的灰度级的概率密度函数,对灰度级执行下式所示的变换,得到输出灰度级S:
Figure PCTCN2019088639-appb-000005
其中,w为积分的虚拟变量。但在数字图像中,灰度级是离散的。因此对于离散的灰度级,可将上式变换为下式:
Figure PCTCN2019088639-appb-000006
式中n k是图像中出现r k级灰度的像素数,n是图像像素总数,
Figure PCTCN2019088639-appb-000007
为频数。在直角坐标系中做出r k与P r(r k)的关系图形,即称为图像的直方图。由于自适应直方图均衡化是在直方图均衡化原理的基础上采用局部法对图像进行处理,所以,自适应直方图均衡化可以起到增强人眼微动脉瘤及附近区域的像素亮度值的作用。
步骤二、检测所述眼底图像数据中的视网膜图像,制作眼底微动脉瘤检测模型的训练样本。
人眼的构造精密复杂,如何正确的捕捉到微动脉瘤部分至关重要。由于微动脉瘤主要存在于视网膜位置,本申请较佳实施例对上述得到的增强处理后的眼底图像数据,进行形态学运算,得到视网膜边界。
所述形态学运算包括:
使用Canny算子检测视网膜图像的边缘,使用一个顶栏和底栏闭合的圆形区域,并填充该区域和空洞后进行下式的形态学开运算操作:
AB=(AΘB)θB
其中,Θ表示腐蚀操作,θ表示膨胀操作。利用形态学开运算后,分别得到腐蚀图像A和膨胀图像B,最后腐蚀图像A减去膨胀图像B,即得到视网膜的圆形边界。
检测到视网膜边界后,计算并判断视网膜边界的面积,若边界面积大于设定阈值δ,表明边界检测错误,重新采用形态学运算继续检测视网膜边界,直到满足阈值δ为止。
步骤三、利用上述得到的训练样本执行眼底微动脉瘤检测模型的训练。
对上述制作的训练样本进行上采样和下采样,然后分别放入一个深度学习模型进行训练。在整个训练过程中,深度学习模型主要检查图片是否有微动脉瘤,若有则标记微动脉瘤的位置。
详细地,所述深度学习模型的训练的方法主要包括如下步骤:
1、上下采样法对训练样本进行采样,得到数据集。下采样(subsampled)就是缩小图像,上采样(upsampling)就是放大图像。上下采样是为了提高模型在不同环境下,对微动脉瘤病症的侦测能力。对于一张尺寸为M*N微动脉瘤图像,对其进行s倍下采样,即得到(M/s)*(N/s)尺寸的分辨率图像,s是M和N的公约数。由于微动脉瘤图片是矩阵形式,还需把原始图像按照如下公式,进行像素的均值处理:
P k=ΣI i/s 2
其中,ΣIi代表微动脉瘤图片的像素总和。上采样原理与下采样类似,只不过采用内插值方法,即在原有微动脉瘤像素的基础上,采用内插值算法插入新的像素。
2、利用上下采样后的数据集训练深度学习模型。本申请所述深度学习模型是卷积神经网络模型。本申请使用Fast RCNN(Fast Region-based ConvolutionalNeural Networks)算法对微动脉瘤进行检测。Fast RCNN是将卷积神经网络模型应用到目标检测问题上的算法,其通过多任务损失函数边框回归方法实现目标检测问题的转化。
所述利用上下采样后的数据集训练深度学习模型包括:
A、利用卷积神经网络提取眼底微动脉瘤病灶的特征向量。本申请使用的卷积神经网络模型是VGG(Visual Geometry Group),VGG的卷积核较小,是3x3规格,因为使用多个较小卷积核的卷积层代替一个卷积核较大的卷积层,一方面可以减少参数,另一方面相当于是进行了更多的非线性映射,可以增加网络的对微动脉瘤的特征提取能力,卷积层步长设置为1。每层卷积核后紧跟池化层。但本申请只使用VGG模型前13层的卷积层和池化层是提取特征,舍去后面的全连层。
B、针对上一步输出的微动脉瘤特征向量,利用softmax分类进行分类判别,若图片中无微动脉瘤特征向量,则判别为正常眼睛,当检查到微动脉瘤的特征,则标记为有微动脉瘤的图片。
C、对于含有微动脉瘤特征的图片,通过多任务损失函数的边框回归方法得到准确的微动脉瘤区域,其中损失函数如下:
Figure PCTCN2019088639-appb-000008
L表示预判断的微动脉瘤区域与实际微动脉瘤区域的损失函数。其中y i,j表示数据集预判断的微动脉瘤区域的像素位置,y′ i,j表示实际微动脉瘤区域位置,x i,j表示微动脉瘤坐标在(i,j)的像素值,微动瘤特征图片共n个像素,w 10和w 11用来调节上下采样后样本不平衡的参数。
步骤四、利用所述眼底微动脉瘤检测模型执行微动脉瘤自动检测。
眼底微动脉瘤检测模型训练好后,本申请将该模型应用于微动脉瘤自动检测中。在检测过程中,本申请对眼底图像以32步长均匀地产生图像块,对每个图像块运用所述眼底微动脉瘤检测模型得到该图像块可能是微动脉瘤的概率,最后统计出概率分布图,判断眼底是否有微动脉瘤,完成自动检测过程。
可选地,在本申请实施例中,所述微动脉瘤自动检测程序01还可以被分割为一个或者多个模块,一个或者多个模块被存储于存储器11中,并由一个或多个处理器(本实施例为处理器12)所执行以完成本申请,本申请所称的模块是指能够完成特定功能的一系列计算机程序指令段,用于描述微动脉瘤自动检测程序在所述装置中的执行过程。
例如,参照图3所示,为本申请装置一实施例中的微动脉瘤自动检测程序的程序模块示意图,该实施例中,微动脉瘤自动检测程序01可以被分割为数据采集模块10、样本数据制作模块20、模型训练模块30、及微动脉瘤检测模块40。示例性地:
所述数据采集模块10用于:采集眼球的眼底图像数据,并对所采集的眼底图像数据执行数据增强处理操作。
本申请所述执行数据增强处理操作包括对所述眼底图像数据进行两次自适应的直方图均衡化预处理,包括:
对眼底图像中像素灰度级r作归一化处理,使r被限定为[0,1]范围内;
对所述像素灰度级r执行式①所示的变换,得到输出灰度级S:
Figure PCTCN2019088639-appb-000009
其中,p r(r)为图像的灰度级的概率密度函数,w为积分的虚拟变量,将式①变换为式②:
Figure PCTCN2019088639-appb-000010
其中,n k是图像中出现r k级灰度的像素数,n是图像像素总数,
Figure PCTCN2019088639-appb-000011
为频数,在直角坐标系中做出r k与P r(r k)的关系图形称为图像的直方图。
所述样本数据制作模块20用于:检测所述眼底图像数据中的视网膜图像,制作眼底微动脉瘤检测模型的训练样本。
优选地,所述检测所述眼底图像数据中的视网膜图像,制作眼底微动脉瘤检测模型的训练样本,包括:
对上述得到的数据增强处理后的眼底图像数据,进行形态学运算,得到视网膜边界,其中,所述形态学运算包括:
使用Canny算子检测视网膜图像的边缘,使用一个顶栏和底栏闭合的圆形区域,并填充该区域和空洞后进行下式的形态学开运算操作:
AB=(AΘB)θB
其中,Θ表示腐蚀操作,θ表示膨胀操作。利用形态学开运算后,分别得到腐蚀图像A和膨胀图像B,最后腐蚀图像A减去膨胀图像B,即得到视网膜的圆形边界;
检测到视网膜边界后,计算并判断视网膜边界的面积;
若视网膜边界面积大于设定阈值δ,表明边界检测错误,重新采用形态学运算继续检测视网膜边界,直到满足阈值δ为止。
所述模型训练模块30用于:利用上述得到的训练样本执行眼底微动脉瘤检测模型的训练。
可选地,所述眼底微动脉瘤检测模型的训练的方法包括:
上下采样法对所述训练样本进行采样,得到数据集;
利用上下采样后的数据集训练深度学习模型,包括:
利用卷积神经网络提取眼底微动脉瘤病灶的特征向量;
针对所述微动脉瘤病灶的特征向量,利用softmax分类进行分类判别,若图片中无微动脉瘤特征向量,则判别为正常眼睛,当检查到微动脉瘤的特征,则标记为有微动脉瘤特征的图片;
对于有微动脉瘤特征的图片,通过多任务损失函数的边框回归方法得到准确的微动脉瘤区域。
所述微动脉瘤检测模块40用于:利用所述眼底微动脉瘤检测模型执行微动脉瘤自动检测。
可选地,所述利用所述眼底微动脉瘤检测模型执行微动脉瘤自动检测,包括:
对眼底图像以32步长均匀地产生图像块,对每个图像块运用所述眼底微动脉瘤检测模型得到该图像块可能是微动脉瘤的概率,统计出概率分布图,判断眼底是否有微动脉瘤,完成自动检测过程。
上述数据采集模块10、样本数据制作模块20、模型训练模块30及微动脉瘤检测模块40等程序模块被执行时所实现的功能或操作步骤与上述实施例大体相同,在此不再赘述。
此外,本申请实施例还提出一种计算机可读存储介质,所述计算机可读存储介质上存储有微动脉瘤自动检测程序,所述微动脉瘤自动检测程序可被一个或多个处理器执行,以实现如下操作:
采集眼球的眼底图像数据,并对所采集的眼底图像数据执行数据增强处理操作;
检测所述眼底图像数据中的视网膜图像,制作眼底微动脉瘤检测模型的训练样本;
利用上述得到的训练样本执行眼底微动脉瘤检测模型的训练;及
利用所述眼底微动脉瘤检测模型执行微动脉瘤自动检测。
本申请计算机可读存储介质具体实施方式与上述微动脉瘤自动检测装置和方法各实施例基本相同,在此不作累述。
需要说明的是,上述本申请实施例序号仅仅为了描述,不代表实施例的优劣。并且本文中的术语“包括”、“包含”或者其任何其他变体意在涵盖非排他性的包含,从而使得包括一系列要素的过程、装置、物品或者方法不仅包括那些要素,而且还包括没有明确列出的其他要素,或者是还包括为这种过程、装置、物品或者方法所固有的要素。在没有更多限制的情况下,由语句“包括一个……”限定的要素,并不排除在包括该要素的过程、装置、物品或者方法中还存在另外的相同要素。
通过以上的实施方式的描述,本领域的技术人员可以清楚地了解到上述实施例方法可借助软件加必需的通用硬件平台的方式来实现,当然也可以通过硬件,但很多情况下前者是更佳的实施方式。基于这样的理解,本申请的技术方案本质上或者说对现有技术做出贡献的部分可以以软件产品的形式体现出来,该计算机软件产品存储在如上所述的一个存储介质(如ROM/RAM、磁碟、光盘)中,包括若干指令用以使得一台终端设备(可以是手机,计算机,服务器,或者网络设备等)执行本申请各个实施例所述的方法。
以上仅为本申请的优选实施例,并非因此限制本申请的专利范围,凡是利用本申请说明书及附图内容所作的等效结构或等效流程变换,或直接或间接运用在其他相关的技术领域,均同理包括在本申请的专利保护范围内。

Claims (20)

  1. 一种微动脉瘤自动检测方法,其特征在于,所述方法包括:
    采集眼球的眼底图像数据,并对所采集的眼底图像数据执行数据增强处理操作;
    检测所述眼底图像数据中的视网膜图像,制作眼底微动脉瘤检测模型的训练样本;
    利用上述得到的训练样本执行眼底微动脉瘤检测模型的训练;及
    利用所述眼底微动脉瘤检测模型执行微动脉瘤自动检测。
  2. 如权利要求1所述的微动脉瘤自动检测方法,其特征在于,所述执行数据增强处理操作包括对所述眼底图像数据进行两次自适应的直方图均衡化预处理,包括:
    对眼底图像数据中像素灰度级r作归一化处理,使r被限定为[0,1]范围内;
    对所述像素灰度级r执行式①所示的变换,得到输出灰度级S:
    Figure PCTCN2019088639-appb-100001
    其中,p r(w)为图像的灰度级的概率密度函数,w为积分的虚拟变量,将式①变换为式②:
    Figure PCTCN2019088639-appb-100002
    其中,n k是图像中出现r k级灰度的像素数,n是图像像素总数,
    Figure PCTCN2019088639-appb-100003
    为频数,在直角坐标系中做出r k与P r(r k)的关系图形称为图像的直方图。
  3. 如权利要求2所述的微动脉瘤自动检测方法,其特征在于,所述检测所述眼底图像数据中的视网膜图像,制作眼底微动脉瘤检测模型的训练样本,包括:
    对上述得到的数据增强处理后的眼底图像数据进行形态学运算,得到视网膜边界,其中,所述形态学运算包括:
    使用Canny算子检测视网膜图像的边缘,使用一个顶栏和底栏闭合的圆形区域,并填充该区域和空洞后进行下式的形态学开运算操作:
    AB=(AΘB)θB
    其中,Θ表示腐蚀操作,θ表示膨胀操作。利用形态学开运算后,分别得到腐蚀图像A和膨胀图像B,最后腐蚀图像A减去膨胀图像B,即得到视网膜的圆形边界;
    检测到视网膜边界后,计算并判断视网膜边界的面积;
    若视网膜边界面积大于设定阈值δ,表明边界检测错误,重新采用形态学运算继续检测视网膜边界,直到满足阈值δ为止。
  4. 如权利要求1至3中任意一项所述的微动脉瘤自动检测方法,其特征在于,所述眼底微动脉瘤检测模型的训练的方法包括:
    上下采样法对所述训练样本进行采样,得到数据集;
    利用上下采样后的数据集训练深度学习模型,包括:
    利用卷积神经网络提取眼底微动脉瘤病灶的特征向量;
    针对所述微动脉瘤病灶的特征向量,利用softmax分类进行分类判别,若图片中无微动脉瘤特征向量,则判别为正常眼睛,当检查到微动脉瘤的特征,则标记为有微动脉瘤特征的图片;
    对于有微动脉瘤特征的图片,通过多任务损失函数的边框回归方法得到准确的微动脉瘤区域。
  5. 如权利要求1所述的微动脉瘤自动检测方法,其特征在于,所述利用所述眼底微动脉瘤检测模型执行微动脉瘤自动检测,包括:
    对眼底图像以32步长均匀地产生图像块,对每个图像块运用所述眼底微动脉瘤检测模型得到该图像块是微动脉瘤的概率,统计出概率分布图,判断眼底是否有微动脉瘤,完成自动检测过程。
  6. 如权利要求2或3所述的微动脉瘤自动检测方法,其特征在于,所述利用所述眼底微动脉瘤检测模型执行微动脉瘤自动检测,包括:
    对眼底图像以32步长均匀地产生图像块,对每个图像块运用所述眼底微动脉瘤检测模型得到该图像块是微动脉瘤的概率,统计出概率分布图,判断眼底是否有微动脉瘤,完成自动检测过程。
  7. 如权利要求4所述的微动脉瘤自动检测方法,其特征在于,所述利用所述眼底微动脉瘤检测模型执行微动脉瘤自动检测,包括:
    对眼底图像以32步长均匀地产生图像块,对每个图像块运用所述眼底微动脉瘤检测模型得到该图像块是微动脉瘤的概率,统计出概率分布图,判断眼底是否有微动脉瘤,完成自动检测过程。
  8. 一种微动脉瘤自动检测装置,其特征在于,所述装置包括存储器和处理器,所述存储器上存储有可在所述处理器上运行的微动脉瘤自动检测程序,所述微动脉瘤自动检测程序被所述处理器执行时实现如下步骤:
    采集眼球的眼底图像数据,并对所采集的眼底图像数据执行数据增强处理操作;
    检测所述眼底图像数据中的视网膜图像,制作眼底微动脉瘤检测模型的训练样本;
    利用上述得到的训练样本执行眼底微动脉瘤检测模型的训练;及
    利用所述眼底微动脉瘤检测模型执行微动脉瘤自动检测。
  9. 如权利要求8所述的微动脉瘤自动检测装置,其特征在于,所述执行数据增强处理操作包括对所述眼底图像数据进行两次自适应的直方图均衡化预处理,包括:
    对眼底图像中像素灰度级r作归一化处理,使r被限定为[0,1]范围内;
    对所述像素灰度级r执行式①所示的变换,得到输出灰度级S:
    Figure PCTCN2019088639-appb-100004
    其中,p r(r)为图像的灰度级的概率密度函数,w为积分的虚拟变量,将式①变换为式②:
    Figure PCTCN2019088639-appb-100005
    其中,n k是图像中出现r k级灰度的像素数,n是图像像素总数,
    Figure PCTCN2019088639-appb-100006
    为频数,在直角坐标系中做出r k与P r(r k)的关系图形称为图像的直方图。
  10. 如权利要求9所述的微动脉瘤自动检测装置,其特征在于,所述检测所述眼底图像数据中的视网膜图像,制作眼底微动脉瘤检测模型的训练样本,包括:
    对上述得到的数据增强处理后的眼底图像数据,进行形态学运算,得到视网膜边界,其中,所述形态学运算包括:
    使用Canny算子检测视网膜图像的边缘,使用一个顶栏和底栏闭合的圆形区域,并填充该区域和空洞后进行下式的形态学开运算操作:
    AB=(AΘB)θB
    其中,Θ表示腐蚀操作,θ表示膨胀操作。利用形态学开运算后,分别得到腐蚀图像A和膨胀图像B,最后腐蚀图像A减去膨胀图像B,即得到视网膜的圆形边界;
    检测到视网膜边界后,计算并判断视网膜边界的面积;
    若视网膜边界面积大于设定阈值δ,表明边界检测错误,重新采用形态学运算继续检测视网膜边界,直到满足阈值δ为止。
  11. 如权利要求8至10中任意一项所述的微动脉瘤自动检测装置,其特征在于,所述眼底微动脉瘤检测模型的训练的方法包括:
    上下采样法对所述训练样本进行采样,得到数据集;
    利用上下采样后的数据集训练深度学习模型,包括:
    利用卷积神经网络提取眼底微动脉瘤病灶的特征向量;
    针对所述微动脉瘤病灶的特征向量,利用softmax分类进行分类判别,若图片中无微动脉瘤特征向量,则判别为正常眼睛,当检查到微动脉瘤的特征,则标记为有微动脉瘤特征的图片;
    对于有微动脉瘤特征的图片,通过多任务损失函数的边框回归方法得到准确的微动脉瘤区域。
  12. 如权利要求8所述的微动脉瘤自动检测装置,其特征在于,所述利用所述眼底微动脉瘤检测模型执行微动脉瘤自动检测,包括:
    对眼底图像以32步长均匀地产生图像块,对每个图像块运用所述眼底微动脉瘤检测模型得到该图像块是微动脉瘤的概率,统计出概率分布图,判断眼底是否有微动脉瘤,完成自动检测过程。
  13. 如权利要求9或10所述的微动脉瘤自动检测装置,其特征在于,所述利用所述眼底微动脉瘤检测模型执行微动脉瘤自动检测,包括:
    对眼底图像以32步长均匀地产生图像块,对每个图像块运用所述眼底微动脉瘤检测模型得到该图像块是微动脉瘤的概率,统计出概率分布图,判断眼底是否有微动脉瘤,完成自动检测过程。
  14. 如权利要求11所述的微动脉瘤自动检测装置,其特征在于,所述利用所述眼底微动脉瘤检测模型执行微动脉瘤自动检测,包括:
    对眼底图像以32步长均匀地产生图像块,对每个图像块运用所述眼底微动脉瘤检测模型得到该图像块是微动脉瘤的概率,统计出概率分布图,判断眼底是否有微动脉瘤,完成自动检测过程。
  15. 一种计算机可读存储介质,其特征在于,所述计算机可读存储介质上存储有微动脉瘤自动检测程序,所述微动脉瘤自动检测程序可被一个或者多个处理器执行,以实现如下步骤:
    采集眼球的眼底图像数据,并对所采集的眼底图像数据执行数据增强处理操作;
    检测所述眼底图像数据中的视网膜图像,制作眼底微动脉瘤检测模型的训练样本;
    利用上述得到的训练样本执行眼底微动脉瘤检测模型的训练;及
    利用所述眼底微动脉瘤检测模型执行微动脉瘤自动检测。
  16. 如权利要求15所述的计算机可读存储介质,其特征在于,所述执行数据增强处理操作包括对所述眼底图像数据进行两次自适应的直方图均衡化预处理,包括:
    对眼底图像中像素灰度级r作归一化处理,使r被限定为[0,1]范围内;
    对所述像素灰度级r执行式①所示的变换,得到输出灰度级S:
    Figure PCTCN2019088639-appb-100007
    其中,p r(r)为图像的灰度级的概率密度函数,w为积分的虚拟变量,将式①变换为式②:
    Figure PCTCN2019088639-appb-100008
    其中,n k是图像中出现r k级灰度的像素数,n是图像像素总数,
    Figure PCTCN2019088639-appb-100009
    为频数,在直角坐标系中做出r k与P r(r k)的关系图形称为图像的直方图。
  17. 如权利要求16所述的计算机可读存储介质,其特征在于,所述检测所述眼底图像数据中的视网膜图像,制作眼底微动脉瘤检测模型的训练样本,包括:
    对上述得到的数据增强处理后的眼底图像数据,进行形态学运算,得到视网膜边界,其中,所述形态学运算包括:
    使用Canny算子检测视网膜图像的边缘,使用一个顶栏和底栏闭合的圆形区域,并填充该区域和空洞后进行下式的形态学开运算操作:
    AB=(AΘB)θB
    其中,Θ表示腐蚀操作,θ表示膨胀操作。利用形态学开运算后,分别得到腐蚀图像A和膨胀图像B,最后腐蚀图像A减去膨胀图像B,即得到视网膜的圆形边界;
    检测到视网膜边界后,计算并判断视网膜边界的面积;
    若视网膜边界面积大于设定阈值δ,表明边界检测错误,重新采用形态学运算继续检测视网膜边界,直到满足阈值δ为止。
  18. 如权利要求15至17中任意一项所述的计算机可读存储介质,其特征在于,所述眼底微动脉瘤检测模型的训练的方法包括:
    上下采样法对所述训练样本进行采样,得到数据集;
    利用上下采样后的数据集训练深度学习模型,包括:
    利用卷积神经网络提取眼底微动脉瘤病灶的特征向量;
    针对所述微动脉瘤病灶的特征向量,利用softmax分类进行分类判别,若图片中无微动脉瘤特征向量,则判别为正常眼睛,当检查到微动脉瘤的特征,则标记为有微动脉瘤特征的图片;
    对于有微动脉瘤特征的图片,通过多任务损失函数的边框回归方法得到准确的微动脉瘤区域。
  19. 如权利要求15-17任一项所述的计算机可读存储介质,其特征在于,所述利用所述眼底微动脉瘤检测模型执行微动脉瘤自动检测,包括:
    对眼底图像以32步长均匀地产生图像块,对每个图像块运用所述眼底微动脉瘤检测模型得到该图像块是微动脉瘤的概率,统计出概率分布图,判断眼底是否有微动脉瘤,完成自动检测过程。
  20. 如权利要求18所述的计算机可读存储介质,其特征在于,所述利用所述眼底微动脉瘤检测模型执行微动脉瘤自动检测,包括:
    对眼底图像以32步长均匀地产生图像块,对每个图像块运用所述眼底微动脉瘤检测模型得到该图像块是微动脉瘤的概率,统计出概率分布图,判断眼底是否有微动脉瘤,完成自动检测过程。
PCT/CN2019/088639 2019-01-23 2019-05-27 微动脉瘤自动检测方法、装置及计算机可读存储介质 Ceased WO2020151149A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910063699.9 2019-01-23
CN201910063699.9A CN109919179A (zh) 2019-01-23 2019-01-23 微动脉瘤自动检测方法、装置及计算机可读存储介质

Publications (1)

Publication Number Publication Date
WO2020151149A1 true WO2020151149A1 (zh) 2020-07-30

Family

ID=66960507

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/088639 Ceased WO2020151149A1 (zh) 2019-01-23 2019-05-27 微动脉瘤自动检测方法、装置及计算机可读存储介质

Country Status (2)

Country Link
CN (1) CN109919179A (zh)
WO (1) WO2020151149A1 (zh)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112288697A (zh) * 2020-10-23 2021-01-29 北京百度网讯科技有限公司 用于量化异常程度的方法、装置、电子设备及可读存储介质
CN113554660A (zh) * 2020-11-28 2021-10-26 上海海事大学 一种用于检测眼底微动脉瘤的检测方法
CN113706451A (zh) * 2021-07-07 2021-11-26 杭州脉流科技有限公司 颅内动脉瘤识别检测的方法、装置、系统和计算机可读存储介质
CN114445364A (zh) * 2022-01-25 2022-05-06 中南大学 眼底图像微动脉瘤区域检测方法及其成像方法
US11981403B1 (en) * 2023-06-05 2024-05-14 Wuhan University Of Technology Method and device for automatic detection of vessel draft depth

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110276802B (zh) * 2019-06-26 2021-08-17 湖南大学 医学图像中病症组织定位方法、装置与设备
CN112308114A (zh) * 2020-09-24 2021-02-02 赣州好朋友科技有限公司 白钨矿的分选方法、装置和可读存储介质
CN115049655B (zh) 2022-08-15 2022-11-11 汕头大学·香港中文大学联合汕头国际眼科中心 一种小鼠模型视网膜病灶分布分析方法

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105046693A (zh) * 2015-06-29 2015-11-11 嘉兴慧康智能科技有限公司 一种眼底图像微动脉瘤自动检测方法
CN106355584A (zh) * 2016-08-30 2017-01-25 上海交通大学 基于局部熵确定阈值的眼底图像微动脉瘤自动检测方法
CN106408562A (zh) * 2016-09-22 2017-02-15 华南理工大学 基于深度学习的眼底图像视网膜血管分割方法及系统
CN106651899A (zh) * 2016-12-09 2017-05-10 东北大学 基于Adaboost的眼底图像微动脉瘤检测系统

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2018116321A2 (en) * 2016-12-21 2018-06-28 Braviithi Technologies Private Limited Retinal fundus image processing method

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105046693A (zh) * 2015-06-29 2015-11-11 嘉兴慧康智能科技有限公司 一种眼底图像微动脉瘤自动检测方法
CN106355584A (zh) * 2016-08-30 2017-01-25 上海交通大学 基于局部熵确定阈值的眼底图像微动脉瘤自动检测方法
CN106408562A (zh) * 2016-09-22 2017-02-15 华南理工大学 基于深度学习的眼底图像视网膜血管分割方法及系统
CN106651899A (zh) * 2016-12-09 2017-05-10 东北大学 基于Adaboost的眼底图像微动脉瘤检测系统

Cited By (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112288697A (zh) * 2020-10-23 2021-01-29 北京百度网讯科技有限公司 用于量化异常程度的方法、装置、电子设备及可读存储介质
CN112288697B (zh) * 2020-10-23 2023-07-28 北京百度网讯科技有限公司 用于量化异常程度的方法、装置、电子设备及可读存储介质
CN113554660A (zh) * 2020-11-28 2021-10-26 上海海事大学 一种用于检测眼底微动脉瘤的检测方法
CN113554660B (zh) * 2020-11-28 2024-03-12 上海海事大学 一种用于检测眼底微动脉瘤的检测方法
CN113706451A (zh) * 2021-07-07 2021-11-26 杭州脉流科技有限公司 颅内动脉瘤识别检测的方法、装置、系统和计算机可读存储介质
CN114445364A (zh) * 2022-01-25 2022-05-06 中南大学 眼底图像微动脉瘤区域检测方法及其成像方法
CN114445364B (zh) * 2022-01-25 2024-05-07 中南大学 眼底图像微动脉瘤区域检测方法及其成像方法
US11981403B1 (en) * 2023-06-05 2024-05-14 Wuhan University Of Technology Method and device for automatic detection of vessel draft depth

Also Published As

Publication number Publication date
CN109919179A (zh) 2019-06-21

Similar Documents

Publication Publication Date Title
WO2020151149A1 (zh) 微动脉瘤自动检测方法、装置及计算机可读存储介质
WO2020140370A1 (zh) 眼底出血点的自动检测方法、装置及计算机可读存储介质
CN114287878B (zh) 一种基于注意力模型的糖尿病性视网膜病变病灶图像识别方法
EP3739431B1 (en) Method for determining point of gaze, contrast adjustment method and device, virtual reality apparatus, and storage medium
EP3783533A1 (en) Artificial intelligence-based ophthalmic disease diagnostic modeling method, apparatus, and system
WO2020199468A1 (zh) 图像分类方法、装置及计算机可读存储介质
US20190325215A1 (en) Methods and apparatuses for image detection, electronic devices and storage media
CN107506770A (zh) 糖尿病视网膜病变眼底照相标准图像生成方法
CN107330449A (zh) 一种糖尿病性视网膜病变体征检测方法及装置
WO2018201632A1 (zh) 用于识别眼底图像病变的人工神经网络及系统
CN108464840A (zh) 一种乳腺肿块自动检测方法及系统
WO2020103417A1 (zh) 一种bmi评测方法、装置及计算机可读存储介质
Zhao et al. Fine-grained diabetic wound depth and granulation tissue amount assessment using bilinear convolutional neural network
CN111938567B (zh) 基于深度学习的眼科参数测量方法、系统及设备
CN112750531A (zh) 一种中医自动化望诊系统、方法、设备和介质
CN107292835B (zh) 一种眼底图像视网膜血管自动矢量化的方法及装置
WO2022088665A1 (zh) 病灶分割方法、装置及存储介质
CN108614999A (zh) 基于深度学习的眼睛睁闭状态检测方法
CN108416371A (zh) 一种糖尿病性视网膜病变自动检测方法
Guo et al. Joint optic disc and cup segmentation using feature fusion and attention
Hwang et al. Smartphone-based diabetic macula edema screening with an offline artificial intelligence
CN114821756B (zh) 一种基于深度学习的眨眼数据统计方法和装置
CN106570530A (zh) 一种眼底影像中硬性渗出提取的方法
CN108921836A (zh) 一种提取眼底图像标注的方法及装置
CN118196218B (zh) 一种眼底图像处理方法、装置及设备

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19911430

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 19911430

Country of ref document: EP

Kind code of ref document: A1