WO2020147263A1 - 一种评估眼底图像质量的方法、装置及存储介质 - Google Patents

一种评估眼底图像质量的方法、装置及存储介质 Download PDF

Info

Publication number
WO2020147263A1
WO2020147263A1 PCT/CN2019/092671 CN2019092671W WO2020147263A1 WO 2020147263 A1 WO2020147263 A1 WO 2020147263A1 CN 2019092671 W CN2019092671 W CN 2019092671W WO 2020147263 A1 WO2020147263 A1 WO 2020147263A1
Authority
WO
WIPO (PCT)
Prior art keywords
fundus image
image
fundus
classification model
quality
Prior art date
Application number
PCT/CN2019/092671
Other languages
English (en)
French (fr)
Inventor
刘莉芬
Original Assignee
平安科技(深圳)有限公司
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 平安科技(深圳)有限公司 filed Critical 平安科技(深圳)有限公司
Priority to SG11202008279XA priority Critical patent/SG11202008279XA/en
Publication of WO2020147263A1 publication Critical patent/WO2020147263A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00Image analysis
    • G06T7/0002Inspection of images, e.g. flaw detection
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR 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; CALCULATING OR 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; CALCULATING OR 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; CALCULATING OR 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
    • G06T2207/30168Image quality inspection

Definitions

  • This application relates to the field of image detection technology, and in particular to a method, device and storage medium for evaluating the quality of fundus images.
  • the current quality assessment methods include: a reference-oriented method based on edges and brightness, based on the length of visible blood vessels in the macular region and Multi-scale filter bank response oriented no-reference method, based on the elliptical local blood vessel density algorithm, a comprehensive quality evaluation method that fully considers the image color, brightness, contrast, blur and other characteristics.
  • the quality problems encountered in fundus images mainly include: darkening of the macular area, overexposure, blurred images, camera lens stains, and optic discs. It is centered or ignores the optic disc, etc., and AMD screening based on fundus images requires the area where the macula is to be clearly visible, but the existing quality assessment methods are not well applicable to the clinical application scenarios of AMD screening, and cannot meet AMD screening. The real-time nature of clinical application scenarios.
  • the present application provides a method, device, and storage medium for evaluating the quality of fundus images, which can solve the problem that the applicability and real-time quality of the quality evaluation method in the prior art cannot meet the clinical application scenarios of AMD screening.
  • this application provides a method for evaluating the quality of fundus images, the method including:
  • a prompt message is issued, and the prompt information is used to remind the collection part corresponding to the abnormal image to be recollected.
  • the method further includes:
  • Data cleaning is performed on the fundus image collection, and the image size of the cleaned src image is transformed to the same size as the dst image.
  • the constructing a classification model based on the deep learning method and the fundus image collection includes:
  • the trained neural network After reaching the preset sensitivity and specificity on the test set, the best model parameters are selected, and the model corresponding to the best model parameters is used as the classification model.
  • the method before performing image cutting and black edge processing on the fundus image according to the above classification model and inputting it to the target detection network, the method further includes:
  • the locating the optic disc area and the macula area in the fundus image according to the target detection network includes:
  • the retinal blood vessels in the fundus image are segmented in a multi-scale manner to calibrate multiple blood vessel end points; the least square method is used to calibrate the multiple blood vessel end points Curve fitting is performed to obtain the macular area.
  • the method further includes:
  • At least two detection areas include a detection frame for detecting the macula area and the optic disc area;
  • the quality of the fundus image is evaluated according to the classification model, and if the macular detection frame and the optic disc detection frame appear on the fundus image at the same time, it is determined that the quality of the fundus image is qualified.
  • the target detection network is a yolo network.
  • the present application provides a device for evaluating the quality of fundus images, which has a function of implementing the method corresponding to the method for evaluating the quality of fundus images provided in the first aspect.
  • the functions can be realized by hardware, or by hardware executing corresponding software.
  • the hardware or software includes one or more modules corresponding to the above-mentioned functions, and the modules may be software and/or hardware.
  • the device includes:
  • Input and output module for obtaining a collection of fundus images to be processed
  • the processing module is used to construct a classification model based on the deep learning method and the fundus image collection; obtain real-time collected clinical patient fundus images through the input and output module; after determining that the fundus image quality is qualified according to the classification model,
  • the above classification model performs image cutting black edge processing on the fundus image, and inputs it to the target detection network;
  • a detection module configured to locate the optic disc area and the macula area in the fundus image according to the target detection network
  • a prompt message is sent through the input and output module, and the prompt information is used to remind the collection part corresponding to the abnormal image to be re-collected.
  • the processing module constructs a classification model according to the deep learning method and the fundus image set, it is further used to:
  • Data cleaning is performed on the fundus image collection.
  • processing module is specifically used for:
  • the trained neural network After reaching the preset sensitivity and specificity on the test set, the best model parameters are selected, and the model corresponding to the best model parameters is used as the classification model.
  • the processing module performs image cutting black edge processing on the fundus image according to the above classification model, and before inputting it to the target detection network through the input and output module, it is also used for:
  • the detection module is specifically used for:
  • the retinal blood vessels in the fundus image are segmented in a multi-scale manner to calibrate multiple blood vessel end points; the least square method is used to calibrate the multiple blood vessel end points Curve fitting is performed to obtain the macular area.
  • the processing module is further configured to:
  • At least two detection areas include a detection frame for detecting the macula area and the optic disc area;
  • the quality of the fundus image is evaluated according to the classification model, and if the macular detection frame and the optic disc detection frame appear on the fundus image at the same time, it is determined that the quality of the fundus image is qualified.
  • the target detection network is a yolo network.
  • Another aspect of the present application provides an apparatus for evaluating the quality of fundus images, which includes at least one connected processor, a memory, and an input and output unit, wherein the memory is used to store program codes, and the processor is used to call The program code in the memory performs the following operations:
  • Construct a classification model according to the deep learning method and the collection of fundus images obtain real-time fundus images of clinical patients collected through the input and output module; after determining that the quality of the fundus images is qualified according to the classification model, compare all the images according to the above classification model
  • the fundus image is processed by cutting the black edges of the image and input to the target detection network;
  • a prompt message is sent through the input and output unit, and the prompt information is used to remind the acquisition site corresponding to the abnormal image to be re-collected.
  • the processor is further configured to perform the following operations before constructing a classification model according to the deep learning method and the fundus image set:
  • Data cleaning is performed on the fundus image collection.
  • the processor specifically performs the following operations:
  • the trained neural network After reaching the preset sensitivity and specificity on the test set, the best model parameters are selected, and the model corresponding to the best model parameters is used as the classification model.
  • the processor is further configured to perform the following operations before performing image black-cutting processing on the fundus image according to the above classification model and inputting it to the target detection network through the input and output module:
  • the processor is configured to perform the following operations:
  • the retinal blood vessels in the fundus image are segmented in a multi-scale manner to calibrate multiple blood vessel end points; the least square method is used to calibrate the multiple blood vessel end points Curve fitting is performed to obtain the macular area.
  • the processor is further configured to perform the following operations before determining that the quality of the fundus image is qualified according to the classification model:
  • At least two detection areas include a detection frame for detecting the macula area and the optic disc area;
  • the quality of the fundus image is evaluated according to the classification model, and if the macular detection frame and the optic disc detection frame appear on the fundus image at the same time, it is determined that the quality of the fundus image is qualified.
  • Another aspect of the present application provides a non-volatile computer storage medium, which includes instructions, which when run on a computer, cause the computer to execute the methods described in the above aspects.
  • a classification model is constructed according to the deep learning method and the fundus image collection; the fundus images of clinical patients collected in real time are acquired; the quality of the fundus images is determined to be qualified according to the classification model Afterwards, perform image cutting and black border processing on the fundus image according to the above classification model, and input it to the target detection network; locate the optic disc area and the macula area in the fundus image according to the target detection network; if it is determined that the processed When there is an abnormal image with defective quality in the fundus image, a prompt message is issued, and the prompt information is used to remind the collection part corresponding to the abnormal image to be re-collected.
  • the acquisition quality of fundus images can be fed back in real time, and the stability of fundus image quality can be guaranteed.
  • FIG. 1 is a schematic flowchart of a method for evaluating the quality of fundus images in an embodiment of this application
  • FIG. 2 is a schematic diagram of a fundus image of qualified quality in an embodiment of the application
  • FIG. 3 is a schematic diagram of a fundus image with abnormal quality in an embodiment of the application.
  • FIG. 4 is a schematic diagram of a fundus image to be evaluated for quality in an embodiment of the application
  • FIG. 5 is a schematic diagram of fundus images detected based on classification models in an embodiment of the application.
  • FIG. 6 is a schematic diagram of the testing process of the classification model and the target detection network constructed based on the foregoing in an embodiment of the application;
  • Fig. 7 is a graph of test results in an embodiment of the application.
  • FIG. 8 is a schematic structural diagram of an apparatus for evaluating the quality of fundus images in an embodiment of the application.
  • FIG. 9 is a schematic diagram of a structure of an apparatus for evaluating the quality of fundus images in an embodiment of the application.
  • This application provides a method, device and storage medium for evaluating the quality of fundus images, which are used in clinical medical testing.
  • this application mainly provides the following technical solutions:
  • the quality control of the captured fundus pictures is carried out.
  • the control system can remind the clinician to re-collect the pictures, thereby ensuring the stability of the fundus picture quality and improving the follow-up The accuracy of the diagnosis.
  • the algorithm has good sensitivity and specificity, and can improve the efficiency and specificity of fundus imaging.
  • deep learning refers to a method of copying this dense neural network. By processing multiple data streams at once, computers can significantly reduce the time required to process data.
  • the application of this technique to deep learning has produced artificial neural networks. These artificial neural networks are composed of input nodes, output nodes and node layers.
  • Input node the input node used to receive data.
  • the output node is used to output the result data.
  • the node layer is used to convert the data input from the input node into content that can be used by the output node.
  • the node layer refers to multiple hidden nodes between the input node and the output node, and the node layer can also become a hidden layer. As data progresses through these hidden nodes, the neural network uses logic to decide to pass the data to the next hidden node.
  • the above-mentioned fundus image collection includes multiple clinically collected fundus images.
  • Fundus images include the macular area, optic disc (OD), retinal blood vessels, fovea and other major structural features.
  • the macula is an avascular gray circular area on the temporal side of the optic disc in the fundus image.
  • the fundus image collection can be divided into a test set and a training set.
  • the above-mentioned fundus image collection comes from a sugar network competition, from which about 4000 normal images and about 4000 images with abnormal quality are selected.
  • abnormal quality include: blurred images, ignoring the optic disc, in the middle of the image optic disc, darkening of the macula, overexposure, and severe disease.
  • the cleaned src image can be resized to be the same as the dst image the size of.
  • the src image refers to the source image
  • the dst image refers to the destination image.
  • the constructing a classification model based on the deep learning method and the fundus image collection includes:
  • Set the size of the image input to the neural network (for example, set the size of the image input to the neural network to 256*256);
  • the trained neural network After reaching the preset sensitivity and specificity on the test set, the best model parameters are selected, and the model corresponding to the best model parameters is used as the classification model.
  • the classification model before constructing the classification model, it is also possible to filter the fundus images that are defective due to environmental factors in the fundus image collection, and filter the fundus images that have abnormal angles such as the optic disc due to the shooting angle.
  • the fundus image is an image taken clinically in real time, and the fundus image may be one or more than two.
  • the target detection network refers to dividing the input image into S*S grids. If the coordinates of the center position of a detection object fall into a certain grid, then this grid is responsible for detecting the detection object.
  • the target detection network may be a yolo network, and this application does not limit the version of the target detection network.
  • the acquired fundus image may be defective due to environmental factors, and the angle of the optic disc may be abnormal due to the shooting angle. Therefore, before performing image cutting black edge processing on the fundus image according to the above classification model and inputting it to the target detection network, the method further includes:
  • the image quality classification model can be used to filter out these overexposed and overdark images.
  • locating the optic disc area and the macula area in the fundus image according to the target detection network includes:
  • the method of locating the macular region of the retinal fundus image is adopted.
  • the fundus image is preprocessed, the retinal blood vessels in the fundus image are segmented by a multi-scale method to calibrate Multiple end points of blood vessels.
  • the least square method is used to perform curve fitting on the multiple calibrated end points of blood vessels to obtain the macular region.
  • the multi-scale method refers to the estimation of three-dimensional information through two-dimensional pictures, which is an inverse process.
  • the multi-scale method can also be referred to as a single image depth estimation based on a multi-scale depth network.
  • the prompt information is used to remind the clinical operation user to re-collect the collection part corresponding to the abnormal image.
  • Fig. 2 is a schematic diagram of a fundus image of qualified quality
  • Fig. 3 is a schematic diagram of a fundus image of abnormal quality.
  • a classification model is constructed according to the deep learning method and the fundus image collection; the fundus images of clinical patients collected in real time are acquired; after the fundus image quality is determined to be qualified according to the classification model Perform image cutting and black border processing on the fundus image according to the above classification model, and input it to the target detection network; locate the optic disc area and the macula area in the fundus image according to the target detection network; if the processed all are determined
  • a prompt message is issued, and the prompt information is used to remind the acquisition site corresponding to the abnormal image to be re-collected.
  • the method further includes:
  • At least two detection areas include a detection frame for detecting the macula area and the optic disc area;
  • the quality of the fundus image is evaluated according to the classification model, and if the macular detection frame and the optic disc detection frame appear on the fundus image at the same time, it is determined that the quality of the fundus image is qualified.
  • Figure 4 is a fundus image to be evaluated for quality
  • Figure 5 is a fundus image detected based on the classification model.
  • the two detection frames are the macular detection frame and the optic disc detection frame respectively.
  • the fundus image may include abnormal conditions of the macula, glaucoma, cataract, etc.
  • images with abnormal angles such as the optic disc due to the shooting angle can be filtered out by the macular area positioning and optic disc positioning.
  • the location of the macular area refers to locating the macular area from the fundus image.
  • image processing can be performed on the fundus image according to the grayscale characteristics of the macular area, and the macular area can be located from the fundus image, or according to the macular fovea and optic nerve The relative positions of the disc and blood vessels determine the macular area image.
  • the macular region can be located by locating the macular region of the retinal fundus image, using the feature that the macular region has no blood vessels, and using a multi-scale method to segment the retinal blood vessels based on the preprocessing of the fundus image, thereby demarcating the vascular end Point, use the least square method to curve-fit the calibration points to get the macular area.
  • Optic disc localization includes image preprocessing, bilateral smoothing filtering, pattern recognition and classification.
  • the image preprocessing can use Lab brightness projection to convert the fundus image into a grayscale image
  • the bilateral smoothing filter is used to filter the retinal blood vessels in the fundus image
  • the pattern recognition and classification can be based on linear operators to remove the original image from the spatial domain. Project to the direction domain, and adjust the parameters according to the relative size of the optic disc and the image to obtain the optic disc area.
  • the method for evaluating the quality of fundus images in this application will be illustrated below with specific application scenarios. Taking the clinical test based on the classification model as an example, the collection of fundus images collected during the clinical test is divided into a training set and a test set.
  • the training set includes: 3490 fundus images with qualified quality and 4220 fundus images with abnormal quality.
  • the test set includes: 460 fundus images with acceptable quality and 424 fundus images with abnormal quality.
  • Figure 6 shows a schematic diagram of the testing process based on the above-built classification model and target detection network.
  • the fundus image collection is collected, the fundus image is resized, and the classification is constructed according to the deep learning method and the fundus image collection. If the quality of the fundus image is qualified, the macular area and the optic disc area are automatically located based on the target detection network. If the macular area and the optic disc area are located, the detection result is given, prompting no need to retake; if it is located in the macular area and the optic disc area, 1 1 or 0, according to the detection result, prompt to retake. If the quality of the fundus image is abnormal, a retake will be prompted based on the detection result, and the process ends.
  • test results are as follows:
  • a schematic structural diagram of a device 80 for evaluating the quality of fundus images can be applied to clinical medical detection.
  • the device 80 in the embodiment of the present application can implement the steps corresponding to the method for evaluating the quality of the fundus image performed in the embodiment corresponding to FIG. 1.
  • the functions implemented by the device 80 can be implemented by hardware, or implemented by hardware executing corresponding software.
  • the hardware or software includes one or more modules corresponding to the above-mentioned functions, and the modules may be software and/or hardware.
  • the device 80 may include an input and output module 801, a processing module 802, and a detection module 803.
  • the functional realization of the processing module 302 and the acquisition module 301 can refer to the operations performed in the embodiment corresponding to FIG. 1, which will not be repeated here. .
  • the processing module can be used to control the receiving and sending operations of the acquiring module 301.
  • the input and output module 801 can be used to obtain a collection of fundus images to be processed
  • the processing module 802 can be used to construct a classification model based on the deep learning method and the fundus image collection; obtain real-time collected fundus images of clinical patients through the input and output module 801; determine that the fundus image quality is qualified according to the classification model Afterwards, perform image cutting black edge processing on the fundus image according to the above classification model, and input it to the target detection network;
  • the detection module 803 is configured to locate the optic disc area and the macula area in the fundus image according to the target detection network;
  • a prompt message is sent through the input and output module 801, and the prompt message is used to remind the collection part corresponding to the abnormal image to be re-collected.
  • the processing module 802 constructs a classification model according to the deep learning method and the fundus image collection; the input and output module 801 obtains the fundus images of clinical patients collected in real time; according to the classification model, it is determined that the quality of the fundus images is qualified , Perform image cutting black edge processing on the fundus image according to the above classification model, and input it to the target detection network through the input and output module 801; locate the optic disc area and the macula area in the fundus image according to the target detection network If it is determined that there is an abnormal image with defective quality in the processed fundus image, the input and output module 801 sends out prompt information, the prompt information is used to remind the collection part corresponding to the abnormal image to be re-collected.
  • the quality of fundus images can be fed back in real time to ensure the stability of fundus image quality, and it can also provide a preliminary data screening function for fundus AMD screening and improve the accuracy of AMD screening.
  • the input and output module 801 obtains the fundus image set to be processed, before the processing module constructs a classification model according to the deep learning method and the fundus image set, it is further used to:
  • Data cleaning is performed on the fundus image collection.
  • processing module 802 is specifically configured to:
  • the trained neural network After reaching the preset sensitivity and specificity on the test set, the best model parameters are selected, and the model corresponding to the best model parameters is used as the classification model.
  • the processing module 802 performs image blackening processing on the fundus image according to the above classification model, and before inputting it to the target detection network through the input and output module, it is also used to:
  • the detection module 803 is specifically configured to:
  • the retinal blood vessels in the fundus image are segmented in a multi-scale manner to calibrate multiple blood vessel end points; the least square method is used to calibrate the multiple blood vessel end points Curve fitting is performed to obtain the macular area.
  • the processing module 802 determines that the quality of the fundus image is qualified according to the classification model, the processing module is further configured to:
  • At least two detection areas include a detection frame for detecting the macula area and the optic disc area;
  • the quality of the fundus image is evaluated according to the classification model, and if the macular detection frame and the optic disc detection frame appear on the fundus image at the same time, it is determined that the quality of the fundus image is qualified.
  • the device for evaluating the quality of fundus images in the embodiments of the present application has been separately introduced above from the perspective of modular functional entities.
  • the following describes a device for evaluating the quality of fundus images from the perspective of hardware, as shown in FIG. 9, which includes : A processor, a memory, an input-output unit, and a computer program stored in the memory and running on the processor.
  • the computer program may be a program corresponding to the method for evaluating the quality of fundus images in the embodiment corresponding to FIG. 1.
  • the processor executes the computer program to implement the embodiment corresponding to FIG.
  • the computer program may be a program corresponding to the method for evaluating fundus image quality in the embodiment corresponding to FIG. 1.
  • the so-called processor can be a central processing unit (Central Processing Unit, CPU), it can also be other general-purpose processors, digital signal processors (Digital Signal Processor, DSP), application specific integrated circuits (Application Specific Integrated Circuit, ASIC), ready-made Field-Programmable Gate Array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.
  • the general-purpose processor may be a microprocessor, or the processor may also be any conventional processor, etc.
  • the processor is the control center of the computer device, and various interfaces and lines are used to connect various parts of the entire computer device.
  • the memory may be used to store the computer program and/or module, and the processor implements the computer by running or executing the computer program and/or module stored in the memory, and calling data stored in the memory.
  • the memory may mainly include a storage program area and a storage data area.
  • the storage program area may store an operating system, an application program required by at least one function (such as a sound playback function, an image playback function, etc.), etc.; the storage data area may store Data created based on the use of mobile phones (such as audio data, video data, etc.), etc.
  • the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disks, memory, plug-in hard disks, Smart Media Card (SMC), Secure Digital (SD) cards , Flash Card, at least one magnetic disk storage device, flash memory device, or other volatile solid-state storage device.
  • non-volatile memory such as hard disks, memory, plug-in hard disks, Smart Media Card (SMC), Secure Digital (SD) cards , Flash Card, at least one magnetic disk storage device, flash memory device, or other volatile solid-state storage device.
  • the transceiver can also be replaced by a receiver and a transmitter, and can be the same or different physical entities. When they are the same physical entity, they can be collectively referred to as transceivers.
  • the memory may be integrated in the processor, or may be provided separately from the processor.
  • the transceiver can be an input and output unit.

Landscapes

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

Abstract

本申请涉及图像检测技术领域,提供一种评估眼底图像质量的方法、装置及存储介质,所述方法包括:获取待处理的眼底图像集合;根据深度学习方法和所述眼底图像集合构建分类模型;获取实时采集的临床病人的眼底图像;根据所述分类模型确定所述眼底图像质量合格后,根据上述分类模型对所述眼底图像进行图像切黑边处理,并输入到目标检测网络;根据所述目标检测网络定位出所述眼底图像中的视盘区域和黄斑区域;若确定处理后的所述眼底图像中存在质量缺陷的异常图像时,发出提示信息,所述提示信息用于提醒对所述异常图像所对应的采集部位重新采集。通过采用本方案,能够实时反馈眼底图像的采集质量,保证眼底图片质量的稳定性。

Description

一种评估眼底图像质量的方法、装置及存储介质
本申请要求于2019年1月18日提交中国专利局、申请号为201910046569.4、发明名称为“一种评估眼底图像质量的方法、装置及存储介质”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
本申请涉及图像检测技术领域,尤其涉及一种评估眼底图像质量的方法、装置及存储介质。
背景技术
随着糖尿病、高血压、青光眼等患者的增多以及眼底筛查工作的开展,采集的眼底图像数量急剧增长,给眼科医疗带来了极大的工作压力。由于清晰的眼底图像是眼底病变自动筛查的先决条件,故需要对眼底图像进行质量评估,目前采用的质量评估方法包括:基于边缘和亮度的面向参考的方法,基于黄斑区域可见血管长度和基于多尺度滤波器组响应的面向无参考的方法,基于椭圆局部血管密度算法、充分考虑图像色彩、亮度、对比度、模糊度等多个特征的综合质量评估方法。
发明人发现,以上质量评估方法都是基于DR应用场景来对眼底图像进行质量评估,由于眼底图像中遇到的质量问题主要有:黄斑区域发黑、曝光过度、图像模糊、照相机镜头污点、视盘为中心或者无视盘等,而基于眼底图像的AMD筛查则需要黄斑所在的区域清晰可见,但是现有的质量评估方法无法很好的适用于AMD筛查临床应用场景,也无法满足AMD筛查临床应用场景的实时性。
发明内容
本申请提供了一种评估眼底图像质量的方法、装置及存储介质,能够解决现有技术中质量评估方法的适用性以及实时性都无法满足AMD筛查临床应用场景的问题。
第一方面,本申请提供一种评估眼底图像质量的方法,所述方法包括:
获取待处理的眼底图像集合;
根据深度学习方法和所述眼底图像集合构建分类模型;
获取实时采集的临床病人的眼底图像;
根据所述分类模型确定所述眼底图像质量合格后,根据上述分类模型对所述眼底图像进行图像切黑边处理,并输入到目标检测网络;
根据所述目标检测网络定位出所述眼底图像中的视盘区域和黄斑区域;
若确定处理后的所述眼底图像中存在质量缺陷的异常图像时,发出提示信息,所述提示信息用于提醒对所述异常图像所对应的采集部位重新采集。
可选的,所述获取待处理的眼底图像集合之后,所述根据深度学习方法和所述眼底图像集合构建分类模型之前,所述方法还包括:
对所述眼底图像集合进行数据清洗,将清洗后的src图像进行图像大小变换到与dst图像相同的大小。
可选的,所述根据深度学习方法和所述眼底图像集合构建分类模型,包括:
选择一个神经网络作为分类模型;
设置输入所述神经网络的图像大小;
对所述眼底图像集合进行预处理,得到训练集和测试集;
使用所述训练集对所述神经网络进行训练;
对于已训练的神经网络,在所述测试集上达到预设的灵敏度和特异性后,选取最佳模型参数,将所述最佳模型参数对应的模型作为所述分类模型。
可选的,所述根据上述分类模型对该眼底图像进行图像切黑边处理,并输入到目标检测网络之前,所述方法还包括:
通过图片质量分类模型过滤所述眼底图像集合中曝光度高于预设曝光度和灰度值高于预设灰度值的眼底图像;
以及通过黄斑区定位和视神经盘定位过滤所述眼底图像集合中无视盘的眼底图像。
可选的,所述根据所述目标检测网络定位出所述眼底图像中的视盘区域和黄斑区域,包括:
将所述眼底图像转化为灰度图像,过滤所述眼底图像中的视网膜血管,基于线性算子将所述眼底图像从空间域投影到方向域,根据视神经盘和所述眼底图像的相对大小调整参数,得到所述视盘区域;
以及在对所述眼底图像进行预处理时,利用多尺度方式对所述眼底图像中的视网膜血管进行分割,以标定多个血管末梢点;采用最小二乘法对标定的所述多个血管末梢点进行曲线拟合,得到所述黄斑区域。
可选的,所述获取实时采集的临床病人的眼底图像之后,所述根据所述分类模型确定所述眼底图像质量合格之前,所述方法还包括:
设置至少两个检测区域,以及为黄斑区域和视盘区域设计偏置范围;其中,所述至少两个检测区域包括用于检测黄斑区域和视盘区域的检测框;
根据所述分类模型对所述眼底图像进行质量评估,若所述眼底图像上同时出现黄斑检测框和视盘检测框,则确定所述眼底图像的质量合格。
可选的,所述目标检测网络为yolo网络。
第二方面,本申请提供一种用于评估眼底图像质量的装置,具有实现对应于上述第一方面提供的评估眼底图像质量的的方法的功能。所述功能可以通过硬件实现,也可以通过硬件执行相应的软件实现。硬件或软件包括一个或多个与上述功能相对应的模块,所述模块可以是软件和/或硬件。
一种可能的设计中,所述装置包括:
输入输出模块,用于获取待处理的眼底图像集合;
处理模块,用于根据深度学习方法和所述眼底图像集合构建分类模型;通过所述输入输出模块获取实时采集的临床病人的眼底图像;根据所述分类模型确定所述眼底图像质量合格后,根据上述分类模型对所述眼底图像进行图像切黑边处理,并输入到目标检测网络;
检测模块,用于根据所述目标检测网络定位出所述眼底图像中的视盘区域和黄斑区域;
若确定处理后的所述眼底图像中存在质量缺陷的异常图像时,通过所述输入输出模块发出提示信息,所述提示信息用于提醒对所述异常图像所对应的采集部位重新采集。
可选的,在所述输入输出模块获取待处理的眼底图像集合之后,所述处理模块根据深度学习方法和所述眼底图像集合构建分类模型之前,还用于:
对所述眼底图像集合进行数据清洗。
可选的,所述处理模块具体用于:
选择一个神经网络作为分类模型;
设置输入所述神经网络的图像大小;
对所述眼底图像集合进行预处理,得到训练集和测试集;
使用所述训练集对所述神经网络进行训练;
对于已训练的神经网络,在所述测试集上达到预设的灵敏度和特异性后,选取最佳模型参数,将所述最佳模型参数对应的模型作为所述分类模型。
可选的,所述处理模块根据上述分类模型对该眼底图像进行图像切黑边处理,并通过所述输入输出模块输入到目标检测网络之前,还用于:
通过图片质量分类模型过滤所述眼底图像集合中曝光度高于预设曝光度和灰度值高于预设灰度值的眼底图像;
以及通过黄斑区定位和视神经盘定位过滤所述眼底图像集合中无视盘的眼底图像。
可选的,所述检测模块具体用于:
将所述眼底图像转化为灰度图像,过滤所述眼底图像中的视网膜血管,基于线性算子将所述眼底图像从空间域投影到方向域,根据视神经盘和所述眼底图像的相对大小调整参数,得到所述视盘区域;
以及在对所述眼底图像进行预处理时,利用多尺度方式对所述眼底图像中的视网膜血管进行分割,以标定多个血管末梢点;采用最小二乘法对标定的所述多个血管末梢点进行曲线拟合,得到所述黄斑区域。
可选的,在所述输入输出模块获取实时采集的临床病人的眼底图像之后,所述处理模块根据所述分类模型确定所述眼底图像质量合格之前,所述处理模块还用于:
设置至少两个检测区域,以及为黄斑区域和视盘区域设计偏置范围;其中,所述至少两个检测区域包括用于检测黄斑区域和视盘区域的检测框;
根据所述分类模型对所述眼底图像进行质量评估,若所述眼底图像上同时出现黄斑检测框和视盘检测框,则确定所述眼底图像的质量合格。
可选的,所述目标检测网络为yolo网络。
本申请又一方面提供了一种用于评估眼底图像质量的装置,其包括至少一个连接的处理器、存储器和输入输出单元,其中,所述存储器用于存储程序代码,所述处理器用于调用所述存储器中的程序代码来执行以下操作:
通过所述输入输出单元获取待处理的眼底图像集合;
根据深度学习方法和所述眼底图像集合构建分类模型;通过所述输入输出 模块获取实时采集的临床病人的眼底图像;根据所述分类模型确定所述眼底图像质量合格后,根据上述分类模型对所述眼底图像进行图像切黑边处理,并输入到目标检测网络;
根据所述目标检测网络定位出所述眼底图像中的视盘区域和黄斑区域;
若确定处理后的所述眼底图像中存在质量缺陷的异常图像时,通过所述输入输出单元发出提示信息,所述提示信息用于提醒对所述异常图像所对应的采集部位重新采集。
可选的,在所述输入输出单元获取待处理的眼底图像集合之后,所述处理器根据深度学习方法和所述眼底图像集合构建分类模型之前,还用于执行以下操作:
对所述眼底图像集合进行数据清洗。
可选的,所述处理器具体执行以下操作:
选择一个神经网络作为分类模型;
设置输入所述神经网络的图像大小;
对所述眼底图像集合进行预处理,得到训练集和测试集;
使用所述训练集对所述神经网络进行训练;
对于已训练的神经网络,在所述测试集上达到预设的灵敏度和特异性后,选取最佳模型参数,将所述最佳模型参数对应的模型作为所述分类模型。
可选的,所述处理器在根据上述分类模型对该眼底图像进行图像切黑边处理,并通过所述输入输出模块输入到目标检测网络之前,还用于执行以下操作:
通过图片质量分类模型过滤所述眼底图像集合中曝光度高于预设曝光度和灰度值高于预设灰度值的眼底图像;
以及通过黄斑区定位和视神经盘定位过滤所述眼底图像集合中无视盘的眼底图像。
可选的,所述处理器用于执行以下操作:
将所述眼底图像转化为灰度图像,过滤所述眼底图像中的视网膜血管,基于线性算子将所述眼底图像从空间域投影到方向域,根据视神经盘和所述眼底图像的相对大小调整参数,得到所述视盘区域;
以及在对所述眼底图像进行预处理时,利用多尺度方式对所述眼底图像中的视网膜血管进行分割,以标定多个血管末梢点;采用最小二乘法对标定的所 述多个血管末梢点进行曲线拟合,得到所述黄斑区域。
可选的,在所述输入输出单元获取实时采集的临床病人的眼底图像之后,所述处理器在根据所述分类模型确定所述眼底图像质量合格之前,还用于执行以下操作:
设置至少两个检测区域,以及为黄斑区域和视盘区域设计偏置范围;其中,所述至少两个检测区域包括用于检测黄斑区域和视盘区域的检测框;
根据所述分类模型对所述眼底图像进行质量评估,若所述眼底图像上同时出现黄斑检测框和视盘检测框,则确定所述眼底图像的质量合格。
本申请又一方面提供了一种非易失性计算机存储介质,其包括指令,当其在计算机上运行时,使得计算机执行上述各方面所述的方法。
相较于现有技术,本申请提供的方案中,根据深度学习方法和所述眼底图像集合构建分类模型;获取实时采集的临床病人的眼底图像;根据所述分类模型确定所述眼底图像质量合格后,根据上述分类模型对所述眼底图像进行图像切黑边处理,并输入到目标检测网络;根据所述目标检测网络定位出所述眼底图像中的视盘区域和黄斑区域;若确定处理后的所述眼底图像中存在质量缺陷的异常图像时,发出提示信息,所述提示信息用于提醒对所述异常图像所对应的采集部位重新采集。通过采用本方案,能够实时反馈眼底图像的采集质量,保证眼底图片质量的稳定性。
附图说明
图1为本申请实施例中评估眼底图像质量的方法的一种流程示意图;
图2为本申请实施例中质量合格的眼底图像的一种示意图;
图3为本申请实施例中质量异常的眼底图像的一种示意图;
图4为本申请实施例中待进行质量评估的眼底图像的一种示意图;
图5为本申请实施例中基于分类模型检测后的眼底图像的一种示意图;
图6为本申请实施例中基于上述构建的分类模型和目标检测网络的测试流程示意图;
图7为本申请实施例中测试结果的曲线图;
图8为本申请实施例中用于评估眼底图像质量的装置的一种结构示意图;
图9为本申请实施例中用于评估眼底图像质量的装置的一种结构示意图。
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
下面详细描述本发明的实施例,所述实施例的示例在附图中示出,其中自始至终相同或类似的标号表示相同或类似的元件或具有相同或类似功能的元件。下面通过参考附图描述的实施例是示例性的,仅用于解释本发明,而不能解释为对本发明的限制。
本技术领域技术人员可以理解,除非特意声明,这里使用的单数形式“一”、“一个”、“所述”和“该”也可包括复数形式。应该进一步理解的是,本发明的说明书中使用的措辞“包括”是指存在所述特征、整数、步骤、操作,但是并不排除存在或添加一个或多个其他特征、整数、步骤、操作。
本技术领域技术人员可以理解,除非另外定义,这里使用的所有术语(包括技术术语和科学术语),具有与本发明所属领域中的普通技术人员的一般理解相同的意义。还应该理解的是,诸如通用字典中定义的那些术语,应该被理解为具有与现有技术的上下文中的意义一致的意义,并且除非像这里一样被特定定义,否则不会用理想化或过于正式的含义来解释。
本领域技术人员应当理解,本发明所称的“应用”、“应用程序”、“应用软件”以及类似表述的概念,是业内技术人员所公知的相同概念,是指由一系列计算机指令及相关数据资源有机构造的适于电子运行的计算机软件。除非特别指定,这种命名本身不受编程语言种类、级别,也不受其赖以运行的操作系统或平台所限制。理所当然地,此类概念也不受任何形式的终端所限制。
本领域技术人员应当理解,本发明所称的“应用”、“应用程序”、“应用软件”以及类似表述的概念,是业内技术人员所公知的相同概念,是指由一系列计算机指令及相关数据资源有机构造的适于电子运行的计算机软件。除非特别指定,这种命名本身不受编程语言种类、级别,也不受其赖以运行的操作系统或平台所限制。理所当然地,此类概念也不受任何形式的终端所限制。
本申请提供一种评估眼底图像质量的方法、装置及存储介质,用于临床医 疗检测。
为解决上述技术问题,本申请主要提供以下技术方案:
基于深度学习和8000张临床眼底图片构建用于眼底图片质量分类的模型,首先利用分类模型过滤掉由于环境因素导致的图像质量问题的眼底图像,然后通过黄斑区定位以及视盘定位过滤掉由于拍摄角度导致的无视盘等角度异常问题的眼底图像。基于以上两点,对拍摄到的眼底图片进行质量控制,当拍摄到质量欠缺的图片时,使用该控制系统可以提醒临床操作医生对图片重新采集,从而保证了眼底图片质量的稳定性,提高后续诊断的准确性。该算法具有良好的敏感度和特异性,能够提高眼底成像的效率和特异性。
其中,深度学习是指一种复制这种密集的神经元网络的方法。通过一次处理多个数据流,计算机能够显著减少处理数据所需的时间。将这种技术应用于深度学习已经产生了人工神经网络。这些人工神经网络由输入节点、输出节点和节点层组成。
输入节点,用于接收数据的输入节点。
输出节点,用于输出结果数据。
节点层,用于将从输入节点输入的数据转换为输出节点可以使用的内容。节点层是指在输入节点和输出节点之间的多个隐藏节点,节点层也可以成为隐藏层。当数据通过这些隐藏节点前进时,神经网络使用逻辑来决定将数据传递给下一个隐藏节点。
参照图1,以下介绍本申请中的一种评估眼底图像质量的方法,所述方法包括:
101、获取待处理的眼底图像集合。
其中,上述眼底图像集合包括多张临床采集的眼底图像。眼底图像包括黄斑区、视神经盘(OD)、视网膜血管、中央凹(fovea)等主要结构的特征。其中,黄斑在眼底图像中为视神经盘颞侧的一块无血管灰色圆形区域。可将该眼底图像集合分割为测试集和训练集。
例如,上述眼底图像集合来自糖网竞赛,从中筛选出4000张左右的正常图像以及4000张左右的质量异常的图像。其中质量异常的情况包括:图像模糊、图像无视盘、图像视盘处于中间、黄斑区发黑、曝光过多以及存在严重病变。
一些实施方式中,为便于构建分类模型,获取待处理的眼底图像集合之后,还可以对收集的眼底图像集合进行数据清洗,将清洗后的src图像进行图像大小变换(resize)到与dst图像同样的大小。其中,src图像是指源图像,dst图像是指目的图像。
102、根据深度学习方法和所述眼底图像集合构建分类模型。
一些实施方式中,所述根据深度学习方法和所述眼底图像集合构建分类模型,包括:
选择一个神经网络作为分类模型;
设置输入所述神经网络的图像大小(例如,设置输入该神经网络的图像大小为256*256);
对所述眼底图像集合进行预处理,得到训练集和测试集;
使用所述训练集对所述神经网络进行训练;
对于已训练的神经网络,在所述测试集上达到预设的灵敏度和特异性后,选取最佳模型参数,将所述最佳模型参数对应的模型作为所述分类模型。
一些实施方式中,在构建所述分类模型之前,还可以过滤该眼底图像集合中由于环境因素导致有缺陷的眼底图像,以及过滤由于拍摄角度导致无视盘等角度异常的眼底图像。以提高构建分类模型的准确性和精确度,便于对采集的图像质量进行临床实时评估,提高质量评估的准确性和效率。
103、获取实时采集的临床病人的眼底图像。
该眼底图像为临床实时拍摄的图像,该眼底图像可以是一张或两张以上。
104、根据所述分类模型确定所述眼底图像质量合格后,根据上述分类模型对所述眼底图像进行图像切黑边处理,并输入到目标检测网络。
其中,目标检测网络是指将输入图像分成S*S个格子,若某个检测对象的中心位置的坐标落入到某个格子,那么这个格子就负责检测出这个检测对象。目标检测网络可以是yolo网络,本申请不对目标检测网络的版本做限定。
一些实施方式中,由于采集的眼底图像可能因环境因素导致有缺陷,以及因拍摄角度导致无视盘等角度会出现异常。因此,在根据上述分类模型对该眼底图像进行图像切黑边处理,并输入到目标检测网络之前,所述方法还包括:
通过图片质量分类模型过滤所述眼底图像集合中曝光度高于预设曝光度和灰度值高于预设灰度值的眼底图像;以及通过黄斑区定位和视神经盘定位过 滤所述眼底图像集合中无视盘的眼底图像。例如,在过滤因环境因素导致有缺陷的图像时,可通过图片质量分类模型可以过滤掉这些曝光过多和过暗的图片。
105、根据所述目标检测网络定位出所述眼底图像中的视盘区域和黄斑区域。
一些实施方式中,根据所述目标检测网络定位出所述眼底图像中的视盘区域和黄斑区域,包括:
(1)将所述眼底图像转化为灰度图像,过滤所述眼底图像中的视网膜血管,基于线性算子将所述眼底图像从空间域投影到方向域,根据视神经盘和所述眼底图像的相对大小调整参数,得到所述视盘区域。
(2)由于黄斑区域没有血管的特征,采用定位视网膜眼底图像黄斑区域的方式,在对所述眼底图像进行预处理时,利用多尺度方式对所述眼底图像中的视网膜血管进行分割,以标定多个血管末梢点。采用最小二乘法对标定的所述多个血管末梢点进行曲线拟合,得到所述黄斑区域。
其中,多尺度方式是指通过二维的图片,估计出三维的信息,是一个逆过程。多尺度方式也可以称为基于多尺度深度网络的单幅图像深度估计。
本申请中,采用最小二乘法进行曲线拟合时,根据给定的m个血管末梢点,不要求这条曲线精确地经过这些血管末梢点,而是曲线y=f(x)的近似曲线y=φ(x)。
106、若确定处理后的所述眼底图像中存在质量缺陷的异常图像时,发出提示信息。
其中,所述提示信息用于提醒临床操作用户对所述异常图像所对应的采集部位重新采集。
为便于理解,下面给出一张质量异常、一张质量正常的眼底图像的示意图,便于对比分析或者学习。图2为质量合格的一种眼底图像的示意图,图3为质量异常的一种眼底图像的示意图。当检测出图3所示的眼底图像为质量异常后,立即在检测仪器的显示界面提示当前采集的眼底图像质量异常,需要重新采集。
与现有机制相比,本申请实施例中,根据深度学习方法和所述眼底图像集合构建分类模型;获取实时采集的临床病人的眼底图像;根据所述分类模型确 定所述眼底图像质量合格后,根据上述分类模型对所述眼底图像进行图像切黑边处理,并输入到目标检测网络;根据所述目标检测网络定位出所述眼底图像中的视盘区域和黄斑区域;若确定处理后的所述眼底图像中存在质量缺陷的异常图像时,发出提示信息,所述提示信息用于提醒对所述异常图像所对应的采集部位重新采集。通过采用本方案,能够实时反馈眼底图像的采集质量,保证眼底图片质量的稳定性,还能够为眼底AMD筛查提供初步数据筛选功能,提高AMD筛查的准确率。
可选的,在本申请的一些实施例中,在获取实时采集的临床病人的眼底图像之后,所述根据所述分类模型确定所述眼底图像质量合格之前,所述方法还包括:
设置至少两个检测区域,以及为黄斑区域和视盘区域设计偏置范围;其中,所述至少两个检测区域包括用于检测黄斑区域和视盘区域的检测框;
根据所述分类模型对所述眼底图像进行质量评估,若所述眼底图像上同时出现黄斑检测框和视盘检测框,则确定所述眼底图像的质量合格。
为便于理解,以下以具体的应用场景对本申请中的定位黄斑区域和视盘区域的流程进行举例说明。如图4和图5所示,图4为待进行质量评估的眼底图像,图5为基于分类模型检测后的眼底图像。当目标检测网络检测图4时,若同时出现如图5所示的区域的两个定位框(也可称为检测框),则判断该图4所示的眼底图像的质量合格。该两个检测框分别为黄斑检测框和视盘检测框。基于该方式检测出质量合格的眼底图像后,该眼底图像中能够清晰的呈现黄斑和视盘的形状、颜色、以及血管,故本申请能够为临床判断人眼的病变状态提供高质量的眼底图像,有利于临床分析。
可选的,在本申请的一些实施例中,由于眼底图像可能会包括黄斑、青光眼、白内障等病变的异常状态,所以需要采集到清晰可见的图像。在步骤105之前,可以通过黄斑区定位和视神经盘定位过滤掉由于拍摄角度导致的无视盘等角度异常的图像。
其中,黄斑区定位是指从眼底图像中定位出黄斑区域,例如,可以根据黄斑区的灰度特征对眼底图像进行图像处理,从眼底图像中定位出黄斑区区域,或者根据黄斑中心凹与视神经盘和血管的相对位置确定黄斑区图像。
一些实施方式中,黄斑区定位可以采用定位视网膜眼底图像黄斑区域的方 式,利用黄斑区域没有血管的特征,在对眼底图像预处理的基础上利用多尺度方法对视网膜血管进行分割,进而标定血管末梢点,利用最小二乘法对标定点进行曲线拟合,得到黄斑区域。
视神经盘定位包括图像预处理、双边平滑滤波、模式识别和分类。其中图像预处理可采用Lab亮度投影,用于将眼底图像转化为灰度图像,双边平滑滤波用于过滤到眼底图像中的视网膜血管,模式识别和分类可基于线性算子将原始图像从空间域投影到方向域,以及根据视神经盘和图像的相对大小来调整参数,得到视盘区域。
为便于理解,以下以具体的应用场景对本申请中的评估眼底图像质量的方法进行举例说明。以基于分类模型进行临床测试为例,临床测试时采集的眼底图像集合分为训练集和测试集。
训练集包括:3490张质量合格的眼底图像,以及4220张质量异常的眼底图像。
测试集包括:460张质量合格的眼底图像,以及424张质量异常的眼底图像。
如图6所示的基于上述构建的分类模型和目标检测网络的测试流程示意图,图6中,采集眼底图像集合,Resize眼底图像,根据深度学习方法和眼底图像集合构建分类。若眼底图像质量合格,则基于目标检测网络自动定位黄斑区域和视盘区域,若定位到黄斑区域和视盘区域,则给出检测结果,提示无需重拍;若定位到黄斑区域和视盘区域中的1个或0个,则根据检测结果,提示重拍。若眼底图像质量异常,则根据检测结果,提示重拍,流程结束。
基于图6所示的测试流程中,测试结果如下:
假阳性fpr(1-specificity):0.07
真阳性tpr(sensitivity):0.99
测试结果的曲线图如图7所示。
上述各实施例中所提及的各项技术特征也同样适用于本申请中的图8和图9所对应的实施例,后续类似之处不再赘述。
以上对本申请中一种评估眼底图像质量的方法进行说明,以下对执行上述 用于评估眼底图像质量的装置进行描述。
如图8所示的一种用于评估眼底图像质量的装置80的结构示意图,其可应用于临床医疗检测。本申请实施例中的装置80能够实现对应于上述图1所对应的实施例中所执行的评估眼底图像质量的方法的步骤。装置80实现的功能可以通过硬件实现,也可以通过硬件执行相应的软件实现。硬件或软件包括一个或多个与上述功能相对应的模块,所述模块可以是软件和/或硬件。所述装置80可包括输入输出模块801、处理模块802和检测模块803,所述处理模块302和获取模块301的功能实现可参考图1所对应的实施例中所执行的操作,此处不作赘述。处理模块可用于控制所述获取模块301的收发操作。
一些实施方式中,所述输入输出模块801可用于获取待处理的眼底图像集合;
所述处理模块802可用于根据深度学习方法和所述眼底图像集合构建分类模型;通过所述输入输出模块801获取实时采集的临床病人的眼底图像;根据所述分类模型确定所述眼底图像质量合格后,根据上述分类模型对所述眼底图像进行图像切黑边处理,并输入到目标检测网络;
所述检测模块803,用于根据所述目标检测网络定位出所述眼底图像中的视盘区域和黄斑区域;
若确定处理后的所述眼底图像中存在质量缺陷的异常图像时,通过所述输入输出模块801发出提示信息,所述提示信息用于提醒对所述异常图像所对应的采集部位重新采集。
本申请实施例中,处理模块802根据深度学习方法和所述眼底图像集合构建分类模型;输入输出模块801获取实时采集的临床病人的眼底图像;根据所述分类模型确定所述眼底图像质量合格后,根据上述分类模型对所述眼底图像进行图像切黑边处理,并通过所述输入输出模块801输入到目标检测网络;根据所述目标检测网络定位出所述眼底图像中的视盘区域和黄斑区域;若确定处理后的所述眼底图像中存在质量缺陷的异常图像时,图该输入输出模块801发出提示信息,所述提示信息用于提醒对所述异常图像所对应的采集部位重新采集。通过采用本方案,能够实时反馈眼底图像的采集质量,保证眼底图片质量的稳定性,还能够为眼底AMD筛查提供初步数据筛选功能,提高AMD筛查的准确率。
可选的,在所述输入输出模块801获取待处理的眼底图像集合之后,所述处理模块根据深度学习方法和所述眼底图像集合构建分类模型之前,还用于:
对所述眼底图像集合进行数据清洗。
可选的,所述处理模块802具体用于:
选择一个神经网络作为分类模型;
设置输入所述神经网络的图像大小;
对所述眼底图像集合进行预处理,得到训练集和测试集;
使用所述训练集对所述神经网络进行训练;
对于已训练的神经网络,在所述测试集上达到预设的灵敏度和特异性后,选取最佳模型参数,将所述最佳模型参数对应的模型作为所述分类模型。
可选的,所述处理模块802根据上述分类模型对该眼底图像进行图像切黑边处理,并通过所述输入输出模块输入到目标检测网络之前,还用于:
通过图片质量分类模型过滤所述眼底图像集合中曝光度高于预设曝光度和灰度值高于预设灰度值的眼底图像;
以及通过黄斑区定位和视神经盘定位过滤所述眼底图像集合中无视盘的眼底图像。
可选的,所述检测模块803具体用于:
将所述眼底图像转化为灰度图像,过滤所述眼底图像中的视网膜血管,基于线性算子将所述眼底图像从空间域投影到方向域,根据视神经盘和所述眼底图像的相对大小调整参数,得到所述视盘区域;
以及在对所述眼底图像进行预处理时,利用多尺度方式对所述眼底图像中的视网膜血管进行分割,以标定多个血管末梢点;采用最小二乘法对标定的所述多个血管末梢点进行曲线拟合,得到所述黄斑区域。
可选的,在所述输入输出模块801获取实时采集的临床病人的眼底图像之后,所述处理模块802根据所述分类模型确定所述眼底图像质量合格之前,所述处理模块还用于:
设置至少两个检测区域,以及为黄斑区域和视盘区域设计偏置范围;其中,所述至少两个检测区域包括用于检测黄斑区域和视盘区域的检测框;
根据所述分类模型对所述眼底图像进行质量评估,若所述眼底图像上同时出现黄斑检测框和视盘检测框,则确定所述眼底图像的质量合格。
上面从模块化功能实体的角度分别介绍了本申请实施例中的用于评估眼底图像质量的装置,以下从硬件角度介绍一种用于评估眼底图像质量的装置,如图9所示,其包括:处理器、存储器、输入输出单元以及存储在所述存储器中并可在所述处理器上运行的计算机程序。例如,该计算机程序可以为图1所对应的实施例中评估眼底图像质量的方法对应的程序。例如,当用于评估眼底图像质量的装置实现如图8所示的用于评估眼底图像质量的装置80的功能时,所述处理器执行所述计算机程序时实现上述图8所对应的实施例中由用于评估眼底图像质量的装置80执行的评估眼底图像质量的方法中的各步骤;或者,所述处理器执行所述计算机程序时实现上述图8所对应的实施例的用于评估眼底图像质量的装置80中各模块的功能。又例如,该计算机程序可以为图1所对应的实施例中评估眼底图像质量的方法对应的程序。
所称处理器可以是中央处理单元(Central Processing Unit,CPU),还可以是其他通用处理器、数字信号处理器(Digital Signal Processor,DSP)、专用集成电路(Application Specific Integrated Circuit,ASIC)、现成可编程门阵列(Field-Programmable Gate Array,FPGA)或者其他可编程逻辑器件、分立门或者晶体管逻辑器件、分立硬件组件等。通用处理器可以是微处理器或者该处理器也可以是任何常规的处理器等,所述处理器是所述计算机装置的控制中心,利用各种接口和线路连接整个计算机装置的各个部分。
所述存储器可用于存储所述计算机程序和/或模块,所述处理器通过运行或执行存储在所述存储器内的计算机程序和/或模块,以及调用存储在存储器内的数据,实现所述计算机装置的各种功能。所述存储器可主要包括存储程序区和存储数据区,其中,存储程序区可存储操作系统、至少一个功能所需的应用程序(比如声音播放功能、图像播放功能等)等;存储数据区可存储根据手机的使用所创建的数据(比如音频数据、视频数据等)等。此外,存储器可以包括高速随机存取存储器,还可以包括非易失性存储器,例如硬盘、内存、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)、至少一个磁盘存储器件、闪存器件、或其他易失性固态存储器件。
所述收发器也可以用接收器和发送器代替,可以为相同或者不同的物理实 体。为相同的物理实体时,可以统称为收发器。所述存储器可以集成在所述处理器中,也可以与所述处理器分开设置。该收发器可以为输入输出单元。
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一个存储介质(如ROM/RAM)中,包括若干指令用以使得一台终端(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。
上面结合附图对本申请的实施例进行了描述,但是本申请并不局限于上述的具体实施方式,上述的具体实施方式仅仅是示意性的,而不是限制性的,本领域的普通技术人员在本申请的启示下,在不脱离本申请宗旨和权利要求所保护的范围情况下,还可做出很多形式,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,这些均属于本申请的保护之内。

Claims (20)

  1. 一种评估眼底图像质量的方法,所述方法包括:
    获取待处理的眼底图像集合;
    根据深度学习方法和所述眼底图像集合构建分类模型;
    获取实时采集的临床病人的眼底图像;
    根据所述分类模型确定所述眼底图像质量合格后,根据上述分类模型对所述眼底图像进行图像切黑边处理,并输入到目标检测网络;
    根据所述目标检测网络定位出所述眼底图像中的视盘区域和黄斑区域;
    若确定处理后的所述眼底图像中存在质量缺陷的异常图像时,发出提示信息,所述提示信息用于提醒对所述异常图像所对应的采集部位重新采集。
  2. 根据权利要求1所述的方法,所述获取待处理的眼底图像集合之后,所述根据深度学习方法和所述眼底图像集合构建分类模型之前,所述方法还包括:
    对所述眼底图像集合进行数据清洗。
  3. 根据权利要求2所述的方法,所述根据深度学习方法和所述眼底图像集合构建分类模型,包括:
    选择一个神经网络作为分类模型;
    设置输入所述神经网络的图像大小;
    对所述眼底图像集合进行预处理,得到训练集和测试集;
    使用所述训练集对所述神经网络进行训练;
    对于已训练的神经网络,在所述测试集上达到预设的灵敏度和特异性后,选取最佳模型参数,将所述最佳模型参数对应的模型作为所述分类模型。
  4. 根据权利要求3所述的方法,所述根据上述分类模型对该眼底图像进行图像切黑边处理,并输入到目标检测网络之前,所述方法还包括:
    通过图片质量分类模型过滤所述眼底图像集合中曝光度高于预设曝光度和灰度值高于预设灰度值的眼底图像;
    以及通过黄斑区定位和视神经盘定位过滤所述眼底图像集合中无视盘的眼底图像。
  5. 根据权利要求4所述的方法,所述根据所述目标检测网络定位出所述眼底图像中的视盘区域和黄斑区域,包括:
    将所述眼底图像转化为灰度图像,过滤所述眼底图像中的视网膜血管,基于线性算子将所述眼底图像从空间域投影到方向域,根据视神经盘和所述眼底图像的相对大小调整参数,得到所述视盘区域;
    以及在对所述眼底图像进行预处理时,利用多尺度方式对所述眼底图像中的视网膜血管进行分割,以标定多个血管末梢点;采用最小二乘法对标定的所述多个血管末梢点进行曲线拟合,得到所述黄斑区域。
  6. 根据权利要求4或5所述的方法,所述获取实时采集的临床病人的眼底图像之后,所述根据所述分类模型确定所述眼底图像质量合格之前,所述方法还包括:
    设置至少两个检测区域,以及为黄斑区域和视盘区域设计偏置范围;其中,所述至少两个检测区域包括用于检测黄斑区域和视盘区域的检测框;
    根据所述分类模型对所述眼底图像进行质量评估,若所述眼底图像上同时出现黄斑检测框和视盘检测框,则确定所述眼底图像的质量合格。
  7. 根据权利要求4所述的方法,所述目标检测网络为yolo网络。
  8. 一种用于评估眼底图像质量的装置,所述装置包括:
    输入输出模块,用于获取待处理的眼底图像集合;
    处理模块,用于根据深度学习方法和所述眼底图像集合构建分类模型;通过所述输入输出模块获取实时采集的临床病人的眼底图像;根据所述分类模型确定所述眼底图像质量合格后,根据上述分类模型对所述眼底图像进行图像切黑边处理,并输入到目标检测网络;
    检测模块,用于根据所述目标检测网络定位出所述眼底图像中的视盘区域和黄斑区域;
    若确定处理后的所述眼底图像中存在质量缺陷的异常图像时,通过所述输入输出模块发出提示信息,所述提示信息用于提醒对所述异常图像所对应的采集部位重新采集。
  9. 根据权利要求8所述的装置,在所述输入输出模块获取待处理的眼底图像集合之后,所述处理模块根据深度学习方法和所述眼底图像集合构建分类模型之前,还用于:
    对所述眼底图像集合进行数据清洗。
  10. 根据权利要求9所述的装置,所述处理模块具体用于:
    选择一个神经网络作为分类模型;
    设置输入所述神经网络的图像大小;
    对所述眼底图像集合进行预处理,得到训练集和测试集;
    使用所述训练集对所述神经网络进行训练;
    对于已训练的神经网络,在所述测试集上达到预设的灵敏度和特异性后,选取最佳模型参数,将所述最佳模型参数对应的模型作为所述分类模型。
  11. 根据权利要求10所述的装置,所述处理模块根据上述分类模型对该眼底图像进行图像切黑边处理,并通过所述输入输出模块输入到目标检测网络之前,还用于:
    通过图片质量分类模型过滤所述眼底图像集合中曝光度高于预设曝光度和灰度值高于预设灰度值的眼底图像;
    以及通过黄斑区定位和视神经盘定位过滤所述眼底图像集合中无视盘的眼底图像。
  12. 根据权利要求11所述的装置,所述检测模块具体用于:
    将所述眼底图像转化为灰度图像,过滤所述眼底图像中的视网膜血管,基于线性算子将所述眼底图像从空间域投影到方向域,根据视神经盘和所述眼底图像的相对大小调整参数,得到所述视盘区域;
    以及在对所述眼底图像进行预处理时,利用多尺度方式对所述眼底图像中的视网膜血管进行分割,以标定多个血管末梢点;采用最小二乘法对标定的所述多个血管末梢点进行曲线拟合,得到所述黄斑区域。
  13. 根据权利要求11或12所述的装置,在所述输入输出模块获取实时采集的临床病人的眼底图像之后,所述处理模块根据所述分类模型确定所述眼底图像质量合格之前,所述处理模块还用于:
    设置至少两个检测区域,以及为黄斑区域和视盘区域设计偏置范围;其中,所述至少两个检测区域包括用于检测黄斑区域和视盘区域的检测框;
    根据所述分类模型对所述眼底图像进行质量评估,若所述眼底图像上同时出现黄斑检测框和视盘检测框,则确定所述眼底图像的质量合格。
  14. 一种用于评估眼底图像质量的装置,所述装置包括:
    至少一个处理器、存储器和输入输出单元;
    其中,所述存储器用于存储程序代码,所述处理器用于调用所述存储器中 存储的程序代码来执行以下操作:
    通过所述输入输出单元获取待处理的眼底图像集合;
    根据深度学习方法和所述眼底图像集合构建分类模型;通过所述输入输出模块获取实时采集的临床病人的眼底图像;根据所述分类模型确定所述眼底图像质量合格后,根据上述分类模型对所述眼底图像进行图像切黑边处理,并输入到目标检测网络;
    根据所述目标检测网络定位出所述眼底图像中的视盘区域和黄斑区域;
    若确定处理后的所述眼底图像中存在质量缺陷的异常图像时,通过所述输入输出单元发出提示信息,所述提示信息用于提醒对所述异常图像所对应的采集部位重新采集。
  15. 根据权利要求14所述的装置,其特征在于,在所述输入输出单元获取待处理的眼底图像集合之后,所述处理器根据深度学习方法和所述眼底图像集合构建分类模型之前,还用于执行以下操作:
    对所述眼底图像集合进行数据清洗。
  16. 根据权利要求15所述的装置,所述处理器具体执行以下操作:
    选择一个神经网络作为分类模型;
    设置输入所述神经网络的图像大小;
    对所述眼底图像集合进行预处理,得到训练集和测试集;
    使用所述训练集对所述神经网络进行训练;
    对于已训练的神经网络,在所述测试集上达到预设的灵敏度和特异性后,选取最佳模型参数,将所述最佳模型参数对应的模型作为所述分类模型。
  17. 根据权利要求16所述的装置,所述处理器在根据上述分类模型对该眼底图像进行图像切黑边处理,并通过所述输入输出模块输入到目标检测网络之前,还用于执行以下操作:
    通过图片质量分类模型过滤所述眼底图像集合中曝光度高于预设曝光度和灰度值高于预设灰度值的眼底图像;
    以及通过黄斑区定位和视神经盘定位过滤所述眼底图像集合中无视盘的眼底图像。
  18. 根据权利要求17所述的装置,所述处理器用于执行以下操作:
    将所述眼底图像转化为灰度图像,过滤所述眼底图像中的视网膜血管,基 于线性算子将所述眼底图像从空间域投影到方向域,根据视神经盘和所述眼底图像的相对大小调整参数,得到所述视盘区域;
    以及在对所述眼底图像进行预处理时,利用多尺度方式对所述眼底图像中的视网膜血管进行分割,以标定多个血管末梢点;采用最小二乘法对标定的所述多个血管末梢点进行曲线拟合,得到所述黄斑区域。
  19. 根据权利要求17或18所述的装置,在所述输入输出单元获取实时采集的临床病人的眼底图像之后,所述处理器在根据所述分类模型确定所述眼底图像质量合格之前,还用于执行以下操作:
    设置至少两个检测区域,以及为黄斑区域和视盘区域设计偏置范围;其中,所述至少两个检测区域包括用于检测黄斑区域和视盘区域的检测框;
    根据所述分类模型对所述眼底图像进行质量评估,若所述眼底图像上同时出现黄斑检测框和视盘检测框,则确定所述眼底图像的质量合格。
  20. 一种非易失性计算机存储介质,其包括指令,当其在计算机上运行时,使得计算机执行如权利要求1-7中任一项所述的方法。
PCT/CN2019/092671 2019-01-18 2019-06-25 一种评估眼底图像质量的方法、装置及存储介质 WO2020147263A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
SG11202008279XA SG11202008279XA (en) 2019-01-18 2019-06-25 Method and device for evaluating quality of fundus image and storage medium

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910046569.4 2019-01-18
CN201910046569.4A CN110021009B (zh) 2019-01-18 2019-01-18 一种评估眼底图像质量的方法、装置及存储介质

Publications (1)

Publication Number Publication Date
WO2020147263A1 true WO2020147263A1 (zh) 2020-07-23

Family

ID=67188786

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/092671 WO2020147263A1 (zh) 2019-01-18 2019-06-25 一种评估眼底图像质量的方法、装置及存储介质

Country Status (3)

Country Link
CN (1) CN110021009B (zh)
SG (1) SG11202008279XA (zh)
WO (1) WO2020147263A1 (zh)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112508898A (zh) * 2020-11-30 2021-03-16 北京百度网讯科技有限公司 眼底图像的检测方法、装置及电子设备
CN114119588A (zh) * 2021-12-02 2022-03-01 北京大恒普信医疗技术有限公司 一种训练眼底黄斑病变区域检测模型的方法、装置及系统
CN116491893A (zh) * 2023-06-28 2023-07-28 依未科技(北京)有限公司 高度近视眼底改变评估方法及装置、电子设备及存储介质
CN116740203A (zh) * 2023-08-15 2023-09-12 山东理工职业学院 用于眼底相机数据的安全存储方法
WO2024093800A1 (zh) * 2022-11-02 2024-05-10 依未科技(北京)有限公司 影像标定方法及装置、影像处理方法及装置、电子设备及存储介质

Families Citing this family (15)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110428410B (zh) * 2019-07-31 2024-02-27 腾讯医疗健康(深圳)有限公司 眼底医学图像处理方法、装置、设备及存储介质
CN110766656B (zh) * 2019-09-19 2023-08-11 平安科技(深圳)有限公司 筛查眼底黄斑区异常的方法、装置、设备和存储介质
CN111127425B (zh) * 2019-12-23 2023-04-28 北京至真互联网技术有限公司 基于视网膜眼底图像的目标检测定位方法和装置
CN111161257A (zh) * 2019-12-31 2020-05-15 中山大学中山眼科中心 一种广域眼底图像质量控制的方法及人工智能系统
CN111311545A (zh) * 2020-01-19 2020-06-19 上海箱云物流科技有限公司 集装箱检测方法、装置及计算机可读存储介质
CN111275682A (zh) * 2020-01-19 2020-06-12 上海箱云物流科技有限公司 集装箱检测方法、装置及计算机可读存储介质
KR102306279B1 (ko) * 2020-01-29 2021-09-29 자이메드 주식회사 안저 영상판독 지원방법, 안저 영상판독 지원장치 및 이를 포함하는 안저 영상판독 지원 시스템
CN111696083B (zh) * 2020-05-20 2024-05-14 平安科技(深圳)有限公司 一种图像处理方法、装置、电子设备及存储介质
CN112017168B (zh) * 2020-08-25 2024-09-06 北京致远慧图科技有限公司 一种检测视神经纤维层缺损的方法及装置
CN112381821A (zh) * 2020-12-08 2021-02-19 北京青燕祥云科技有限公司 一种智能手持式眼底相机以及图像分析方法
CN112651938B (zh) * 2020-12-24 2023-12-19 平安科技(深圳)有限公司 视盘图像分类模型的训练方法、装置、设备及存储介质
CN114387209A (zh) * 2021-12-03 2022-04-22 依未科技(北京)有限公司 眼底结构特征确定的方法、装置、介质和设备
CN114494734A (zh) * 2022-01-21 2022-05-13 平安科技(深圳)有限公司 基于眼底图像的病变检测方法、装置、设备及存储介质
CN116309391B (zh) * 2023-02-20 2023-09-05 依未科技(北京)有限公司 图像处理方法及装置、电子设备及存储介质
CN116687338B (zh) * 2023-08-01 2023-11-03 杭州目乐医疗科技股份有限公司 眼底相机状态检测方法、眼底相机与存储介质

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20150104087A1 (en) * 2013-10-10 2015-04-16 University Of Rochester Automated Fundus Image Field Detection and Quality Assessment
US20170270653A1 (en) * 2016-03-15 2017-09-21 International Business Machines Corporation Retinal image quality assessment, error identification and automatic quality correction
CN107209933A (zh) * 2014-08-25 2017-09-26 新加坡科技研究局 用于评估视网膜图像以及从视网膜图像获得信息的方法和系统
CN107862678A (zh) * 2017-10-19 2018-03-30 宁波大学 一种眼底图像无参考质量评价方法

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
AU737411B2 (en) * 1997-04-01 2001-08-16 Johns Hopkins University, The A system for imaging an ocular fundus semi-automatically at high resolution and wide field
BR112019022447A2 (pt) * 2017-04-27 2020-06-09 Bober Miroslaw sistema e método para análise de imagem funduscópica automatizada
CN107451998B (zh) * 2017-08-08 2020-11-06 北京大恒普信医疗技术有限公司 一种眼底图像质量控制方法
CN108470359A (zh) * 2018-02-11 2018-08-31 艾视医疗科技成都有限公司 一种糖尿病性视网膜眼底图像病变检测方法
CN108717696B (zh) * 2018-05-16 2022-04-22 上海鹰瞳医疗科技有限公司 黄斑影像检测方法和设备

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20150104087A1 (en) * 2013-10-10 2015-04-16 University Of Rochester Automated Fundus Image Field Detection and Quality Assessment
CN107209933A (zh) * 2014-08-25 2017-09-26 新加坡科技研究局 用于评估视网膜图像以及从视网膜图像获得信息的方法和系统
US20170270653A1 (en) * 2016-03-15 2017-09-21 International Business Machines Corporation Retinal image quality assessment, error identification and automatic quality correction
CN107862678A (zh) * 2017-10-19 2018-03-30 宁波大学 一种眼底图像无参考质量评价方法

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112508898A (zh) * 2020-11-30 2021-03-16 北京百度网讯科技有限公司 眼底图像的检测方法、装置及电子设备
CN114119588A (zh) * 2021-12-02 2022-03-01 北京大恒普信医疗技术有限公司 一种训练眼底黄斑病变区域检测模型的方法、装置及系统
WO2024093800A1 (zh) * 2022-11-02 2024-05-10 依未科技(北京)有限公司 影像标定方法及装置、影像处理方法及装置、电子设备及存储介质
CN116491893A (zh) * 2023-06-28 2023-07-28 依未科技(北京)有限公司 高度近视眼底改变评估方法及装置、电子设备及存储介质
CN116491893B (zh) * 2023-06-28 2023-09-15 依未科技(北京)有限公司 高度近视眼底改变评估方法及装置、电子设备及存储介质
CN116740203A (zh) * 2023-08-15 2023-09-12 山东理工职业学院 用于眼底相机数据的安全存储方法
CN116740203B (zh) * 2023-08-15 2023-11-28 山东理工职业学院 用于眼底相机数据的安全存储方法

Also Published As

Publication number Publication date
CN110021009B (zh) 2023-07-21
SG11202008279XA (en) 2020-09-29
CN110021009A (zh) 2019-07-16

Similar Documents

Publication Publication Date Title
WO2020147263A1 (zh) 一种评估眼底图像质量的方法、装置及存储介质
WO2020259209A1 (zh) 眼底图像识别方法、装置、设备和存储介质
JP7229881B2 (ja) 医用画像処理装置、学習済モデル、医用画像処理方法及びプログラム
US20210390696A1 (en) Medical image processing apparatus, medical image processing method and computer-readable storage medium
Patton et al. Retinal image analysis: concepts, applications and potential
CN109741346A (zh) 感兴趣区域提取方法、装置、设备及存储介质
WO2021082691A1 (zh) 眼部oct图像病灶区域的分割方法、装置及终端设备
CN109697719B (zh) 一种图像质量评估方法、装置及计算机可读存储介质
JP2022103221A (ja) 医用画像処理装置、医用画像処理システム、医用画像処理方法及びプログラム
CN108765392B (zh) 一种基于滑动窗口的消化道内镜病变检测和识别方法
CN110766656B (zh) 筛查眼底黄斑区异常的方法、装置、设备和存储介质
WO2020042404A1 (zh) 用于远程眼底筛查的质量控制方法和系统、及存储设备
CN110136147A (zh) 一种基于U-Net模型的分割医学图像的方法、装置及存储介质
WO2021190656A1 (zh) 眼底图像黄斑中心定位方法及装置、服务器、存储介质
JP7264195B2 (ja) 画像処理方法、画像処理プログラム、画像処理装置、画像表示装置、及び画像表示方法
CN113658165A (zh) 杯盘比确定方法、装置、设备及存储介质
Huang et al. A depth-first search algorithm based otoscope application for real-time otitis media image interpretation
Majumdar et al. An automated graphical user interface based system for the extraction of retinal blood vessels using kirsch‘s template
CN116030042A (zh) 一种针对医生目诊的诊断装置、方法、设备及存储介质
WO2021139446A1 (zh) 一种抗血管内皮生长因子vegf疗效预测装置及方法
WO2019184824A1 (zh) 准直器边界检测方法的训练方法和系统
CN111539940B (zh) 超广角眼底图像生成方法及设备
Gibbon et al. PallorMetrics: Software for Automatically Quantifying Optic Disc Pallor in Fundus Photographs, and Associations With Peripapillary RNFL Thickness
CN115760806A (zh) 一种裂隙灯图像质量检测方法
CN117475270B (zh) 糖尿病视网膜病变图像的病灶识别方法及系统

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: 19910148

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: 19910148

Country of ref document: EP

Kind code of ref document: A1