WO2022227342A1 - 眼底图像黄斑区域的识别检测方法和装置及设备 - Google Patents

眼底图像黄斑区域的识别检测方法和装置及设备 Download PDF

Info

Publication number
WO2022227342A1
WO2022227342A1 PCT/CN2021/112998 CN2021112998W WO2022227342A1 WO 2022227342 A1 WO2022227342 A1 WO 2022227342A1 CN 2021112998 W CN2021112998 W CN 2021112998W WO 2022227342 A1 WO2022227342 A1 WO 2022227342A1
Authority
WO
WIPO (PCT)
Prior art keywords
macular
region
image
macular region
fundus image
Prior art date
Application number
PCT/CN2021/112998
Other languages
English (en)
French (fr)
Inventor
张冬冬
Original Assignee
北京至真互联网技术有限公司
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 北京至真互联网技术有限公司 filed Critical 北京至真互联网技术有限公司
Priority to US18/010,804 priority Critical patent/US11908137B2/en
Publication of WO2022227342A1 publication Critical patent/WO2022227342A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/18—Eye characteristics, e.g. of the iris
    • G06V40/193—Preprocessing; Feature extraction
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G06T7/0012—Biomedical image inspection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/40—Image enhancement or restoration using histogram techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/50—Image enhancement or restoration using two or more images, e.g. averaging or subtraction
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/90—Determination of colour characteristics
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/22—Image preprocessing by selection of a specific region containing or referencing a pattern; Locating or processing of specific regions to guide the detection or recognition
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30004—Biomedical image processing
    • G06T2207/30041—Eye; Retina; Ophthalmic
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V2201/00—Indexing scheme relating to image or video recognition or understanding
    • G06V2201/03—Recognition of patterns in medical or anatomical images

Definitions

  • the present invention relates to the technical field of medical image processing, in particular to a method, device and device for identifying and detecting a macular region in a fundus image.
  • Color fundus photography is one of the important means to assist fundus image detection and screening.
  • the macula directly affects people's vision.
  • the detection and classification of macular abnormalities There are few studies on the detection and classification of macular abnormalities.
  • the detection and classification of the macular region mainly includes two technical modules, namely, the fundus macular region identification and localization module and the macular region abnormality detection module.
  • the macular area identification and localization algorithm based on deep learning is relatively mature, and it only needs to mark and learn the macular area in the fundus image to achieve the effect of macular area localization.
  • the localization of the macular region by the deep learning model trained only by limited fundus images often results in inaccurate positioning or failure, which in turn affects the The results of the detection and classification of the macular region are then faced, which ultimately results in a low accuracy of the location detection results for the macular region of the fundus image.
  • the purpose of the present invention is to provide a method, device and device for identifying and detecting the macular region of the fundus image, which can effectively improve the identification and detection results of the macular region of the fundus image.
  • the present invention provides a method for identifying and detecting a macular region in a fundus image, comprising:
  • Multi-modal processing is performed on the macular image, the images obtained by the multi-modal processing are fused to obtain a fused image, and whether the macular region is qualified or not is detected according to the fused image.
  • identifying and positioning the macular area including:
  • the center coordinates of the optic disc region are obtained in combination with a multiple linear regression model.
  • center coordinates of the optic disc region in combination with a multiple linear regression model to obtain the center coordinates of the macular region, according to the formula:
  • Y is the center coordinate of the macular area
  • X is the center coordinate of the optic disc area
  • W and B are both matrix parameters.
  • performing multimodal processing on the macular image includes: performing at least one of limited contrast adaptive histogram enhancement processing, Gaussian filter enhancement processing, and HSV color space processing on the macular image.
  • whether the macular region is qualified according to the fusion image including:
  • the matching network includes a feature extraction network, a feature association network, a similarity measurement network and an attention distribution network that are cascaded in sequence.
  • the present invention also provides a device for identifying and detecting a macular region in a fundus image, comprising an image reading module, a first positioning module for the macular region, a second positioning module for the macular region, a macular region extraction module, and a macular region detection module;
  • the image reading module is configured to read the current fundus image to be positioned and detected
  • the macular region first positioning module is configured to detect the macular region in the fundus image by using a target detection model
  • the second macular region localization module is configured to detect the optic disc region in the fundus image when the macular region in the fundus image is not detected by the first macular region localization module, and based on the detected the optic disc region, identify and locate the macular region;
  • the macular region extraction module is configured to extract a macular image corresponding to the macular region from the fundus image based on the positioning result of the macular region;
  • the macular region detection module is configured to perform multi-modal processing on the macular image, fuse the images obtained by the multi-modal processing to obtain a fused image, and determine whether the macular region is qualified according to the fused image detection.
  • the second positioning module in the macular region includes: a coordinate acquisition sub-module and a macular positioning sub-module;
  • the coordinate acquisition submodule is configured to acquire the center coordinates of the optic disc region
  • the macular locating sub-module is configured to obtain the central coordinates of the macular region according to the central coordinates of the optic disc region in combination with a multiple linear regression model.
  • the macular locating sub-module is configured to obtain the central coordinates of the macular region according to the central coordinates of the optic disc region in combination with a multiple linear regression model, according to the formula:
  • Y is the center coordinate of the macular area
  • X is the center coordinate of the optic disc area
  • W and B are both matrix parameters.
  • the macular region detection module includes a multimodal amplicon module
  • the multimodal amplicon module configured to perform multimodal processing on the macular image
  • performing multimodal processing on the macular image includes: performing at least one of contrast-limited adaptive histogram enhancement processing, Gaussian filter enhancement processing, and HSV color space processing on the macular image.
  • the present invention also provides a positioning detection device for the macular region of the fundus image, comprising:
  • memory for storing processor-executable instructions
  • the processor is configured to implement the method for identifying and detecting a macular region in a fundus image according to any one of claims 1 to 5 when executing the executable instructions.
  • the present invention discloses the following technical effects:
  • the present invention detects the macular region in the fundus image by using the target detection model, and when the macular region in the fundus image is not detected, detects the optic disc region in the fundus image, and detects the macular region based on the detected optic disc region.
  • Recognition and localization realizes the recognition and localization of the macular area by comprehensively considering the positional relationship between the macular area and the optic disc area. The identification and positioning of the macular region is more accurate.
  • the macular image corresponding to the macular region is extracted from the fundus image, multimodal processing is performed on the macular image, and the images obtained by the multimodal processing are fused to obtain a fused image. The image is checked for compliance with the macular region.
  • FIG. 1 is a flowchart of a method for identifying and detecting a macular region in a fundus image according to an embodiment of the present invention
  • FIG. 2 is a flow chart of locating a macular region in a method for identifying and detecting a macular region in a fundus image according to an embodiment of the present invention
  • FIG. 3 is a schematic diagram of performing multimodal processing on a macular image during macular region detection in a method for identifying and detecting macular regions in a fundus image according to an embodiment of the present invention
  • FIG. 4 is a flow chart of detecting a macular region in a method for identifying and detecting a macular region in a fundus image according to an embodiment of the present invention
  • FIG. 5 is a schematic diagram of a network structure of a matching network used in a method for identifying and detecting a macular region in a fundus image according to an embodiment of the present invention to detect whether the macular region is qualified or not;
  • FIG. 6 is a structural block diagram of an apparatus for identifying and detecting a macular region in a fundus image according to an embodiment of the present invention
  • FIG. 7 is a structural block diagram of an apparatus for identifying and detecting a macular region of a fundus image according to an embodiment of the present invention.
  • the purpose of the present invention is to provide a method, device and device for identifying and detecting the macular region of the fundus image, which can effectively improve the identification and detection results of the macular region of the fundus image.
  • FIG. 1 shows a flowchart of a method for identifying and detecting a macular region in a fundus image according to an embodiment of the present application. As shown in Figure 1, the method includes:
  • Step S100 reading the current fundus image to be positioned and detected.
  • Step S200 the target detection model is used to detect the macular region in the fundus image.
  • the adopted target detection model may be the Faster Rcnn neural network.
  • Step S300 when the macular region in the fundus image is not detected, detect the optic disc region in the fundus image, and identify and locate the macular region based on the detected optic disc region.
  • Step S400 based on the identification and positioning result of the macular region, extract a macular image corresponding to the macular region from the fundus image image.
  • Step S500 performing multi-modality processing on the macular image, and after fusing the images obtained by the multi-modality processing to obtain a fused image, check whether the macular region is qualified or not according to the fused image.
  • the method for identifying and detecting the macular region in the fundus image detects the macular region in the fundus image by using the target detection model.
  • the optic disc in the fundus image is detected
  • the macula region is detected, and the macular region is identified and positioned based on the detected optic disc region, which realizes the macular region localization considering the positional relationship between the macular region and the optic disc region.
  • the macular region localization method based on deep learning, it completely relies on the fundus
  • the way of labeling the macular area makes the identification and positioning of the macular area of the fundus image more accurate.
  • the macular image corresponding to the macular region is extracted from the fundus image, multimodal processing is performed on the macular image, and the images obtained by the multimodal processing are fused to obtain a fused image.
  • the image is checked for compliance with the macular region.
  • the center coordinates of the optic disc region can be obtained, and then the macula can be obtained by combining the multiple linear regression model according to the center coordinates of the optic disc region.
  • the coordinates of the center of the area are obtained.
  • Y is the center coordinate of the macular area
  • X is the center coordinate of the optic disc area
  • W and B are matrix parameters.
  • W and B can be determined by the least squares method.
  • W and B are both matrix parameters
  • Y is the center coordinate of the macular area
  • X is the center coordinate of the optic disc area.
  • X 0 is a matrix vector, consisting of an X matrix vector and a 1 vector.
  • the target detection model is used when detecting the macular region and the optic disc region of the fundus image. Therefore, before the identification and positioning of the macular area and the optic disc area are performed on the fundus image to be identified and detected, the target detection model needs to be trained to make the used target detection model converge.
  • the collected sample data that is, the fundus image data
  • the marked contents include the positions of the macular area and the optic disc area in the fundus image (which can be marked with a rectangular frame), and the center coordinates of the macular area and the optic disc area. (Can be marked with coordinate points).
  • the labeled sample data is sequentially input into the target detection model (Faster Rcnn), and the target detection model is trained so that it can detect and locate the macular area and the optic disc area in the fundus image.
  • the target detection model Faster Rcnn
  • step S200 when locating the macular region of the fundus image to be located and detected, firstly through step S200 , the pre-trained target detection model is used to detect the macular region of the fundus image. At the same time, step S200' is executed to determine whether the macular region is detected. If the macular region is directly detected, then through step S300', the position of the macular region in the fundus image is directly located. If the macular area is not detected, step S310 is executed, and the trained target detection model is used to detect the optic disc area on the fundus image, and it is determined whether the optic disc area is detected. If the optic disc area is also not detected, it indicates that there is a problem in the fundus image, and it is not suitable for the identification and classification of the macular area.
  • step S340 a prompt of detection failure is issued, and the next fundus image is directly read to perform localization detection of the macular region.
  • step S320 is used to obtain the center coordinates of the optic disc area
  • the multiple linear regression model is used to calculate and determine the center coordinates of the macular area according to the located center coordinates of the optic disc area.
  • step S330 is used to use a frame body such as a rectangular frame or a circular frame with a preset corresponding size, and the center coordinates of the located macular area are taken as the center to carry out the macular area. Identify positioning.
  • the center coordinate of the determined macular area can be used as the center, and the size of the located optic disc area is 1-1.5 times the size of the macular area. Cropped. For example, the macular area can be cropped with 1.5 times the size of the located optic disc area.
  • the macular region in the fundus image After the macular region in the fundus image is located in the above-mentioned manner, the macular region can be cut out from the fundus image and output, so as to extract the macular image corresponding to the macular region.
  • the macular region can be checked for eligibility based on the obtained macular image.
  • the detection of whether the macular region is qualified refers to the detection of whether the located macular region is normal and whether there is any abnormality.
  • the method before the macular region is checked for eligibility, the method further includes performing multi-modal processing on the macular image, and fusing the images obtained by the multi-modal processing to obtain a fusion image operation.
  • performing multimodal processing on the macular image includes at least one of contrast-adaptive histogram enhancement processing, Gaussian filter enhancement processing, and HSV color space processing.
  • the multimodal processing performed on the macular image includes limited contrast adaptive histogram enhancement processing, Gaussian filter enhancement processing and HSV color space processing.
  • the multimodal amplification module includes a limited contrast adaptive histogram enhancement processing unit, a Gaussian filter enhancement processing unit and an HSV color space processing unit, which are processed by the limited contrast adaptive histogram enhancement processing unit respectively.
  • the unit, the Gaussian filter enhancement processing unit and the HSV color space processing unit perform corresponding processing on the macular image to obtain a Contrast Contrast Adaptive Histogram Enhancement (CLAHE) image, a Gaussian filter enhancement image and an HSV color space map.
  • CLAHE Contrast Contrast Adaptive Histogram Enhancement
  • the fusion unit in the multimodal augmentation module uses the original image (ie, the cropped and extracted macular image), the limited contrast adaptive histogram enhancement (CLAHE) image, the Gaussian filter enhanced image and the HSV color space map.
  • the images of the four modalities are fused to obtain the fused image as the input image.
  • the obtained fused image is input into a pre-trained matching network, and the fused image is detected by the matching network, and finally a qualified or abnormal detection result of the macular image is obtained.
  • the fusion can be performed by superimposing image channels. That is, for images of four modalities of original image (ie, cropped extracted macular image), limited contrast adaptive histogram enhancement (CLAHE) image, Gaussian filter enhanced image, and HSV color space map, each modality
  • CLAHE limited contrast adaptive histogram enhancement
  • Gaussian filter enhanced image Gaussian filter enhanced image
  • HSV color space map each modality
  • Each of the images has 3 channels, and 12 channels of data can be obtained by superimposing the image channels as the input of the matching network.
  • the matching network when the obtained fused image is input into the matching network to check whether it is qualified or not, the matching network also needs to be pre-trained to converge.
  • the training of the matching network specifically includes: step S001, constructing a macular image data set, including qualified macular images (ie, normal macular images) and abnormal macular images (such as: epimacular membrane, macular hole, age macular degeneration, central serous chorioretinopathy, central exudative chorioretinopathy and other macular degeneration images).
  • qualified macular images ie, normal macular images
  • abnormal macular images such as: epimacular membrane, macular hole, age macular degeneration, central serous chorioretinopathy, central exudative chorioretinopathy and other macular degeneration images.
  • a multi-modal amplification module is constructed, and images of four modalities of original image, limited contrast adaptive histogram enhancement (CLAHE) image, Gaussian filter enhanced image, and HSV color space map are used for fusion as model training. Enter an image.
  • CLAHE limited contrast adaptive histogram enhancement
  • Gaussian filter enhanced image Gaussian filter enhanced image
  • HSV color space map HSV color space map
  • step S003 is performed to divide the multimodal fusion image into a support data set and a target data set, and through step S004, the support data set and the target data set are sequentially input into the matching network for training learning and testing applications.
  • the network structure of the designed matching network may include a feature extraction network, a feature association network, a similarity measurement network, and an attention distribution network that are cascaded in sequence.
  • the feature extraction network can use the convolutional neural network to extract features from the support image and the target image respectively (for example, ResNet50 can be used), and the feature association network uses the recurrent neural network to analyze the internal and external correlation between the support image and the target image.
  • the similarity measurement module is not trained in the network (for example, cosine similarity can be used for measurement), and the attention distribution network is used to improve the training efficiency and performance of the model (such as: The softmax function can be used for attention distribution).
  • the network is used to learn the similarity between the target image and the support image, so as to achieve the effect of classifying the target image by using a small number of samples.
  • the matching network By inputting the fused image obtained by multimodal fusion into the trained matching network, the matching network sequentially performs feature extraction, feature association and similarity measurement and attention allocation on the fused image to obtain the final classification result.
  • the present application further provides a device for identifying and detecting a macular region in a fundus image. Since the working principle of the apparatus for identifying and detecting the macular region of the fundus image provided by the present application is the same as or similar to the principle of the method for identifying and detecting the macular region of the fundus image of the present application, the repetition will not be repeated.
  • the apparatus 100 for identifying and detecting the macular region of the fundus image includes an image reading module 110, a first positioning module 120 for the macular region, a second positioning module 130 for the macular region, a macular region extraction module 140, and a macular region detection module. module 150.
  • the image reading module 110 is configured to read the current fundus image to be positioned and detected.
  • the macular region first localization module 120 is configured to detect the macular region in the fundus image by using a target detection model.
  • the second macular area localization module 130 is configured to detect the optic disc area in the fundus image when the macular area in the fundus image is not detected by the first macular area localization module 120, and based on the detected optic disc area, locate the macular area. area for identification and positioning.
  • the macular region extraction module 140 is configured to extract a macular image corresponding to the macular region from the fundus image based on the positioning result of the macular region.
  • the macular region detection module 150 is configured to perform multi-modality processing on the macular image, fuse the images obtained by the multimodal processing to obtain a fused image, and detect whether the macular region is qualified or not according to the fused image.
  • the second macular region positioning module 130 includes: a coordinate acquisition sub-module and a macular positioning sub-module (not shown in the figure).
  • the coordinate acquisition sub-module is configured to acquire the center coordinates of the optic disc area.
  • the macular locating sub-module is configured to obtain the central coordinates of the macular region according to the central coordinates of the optic disc region in combination with the multiple linear regression model.
  • the macular locating sub-module is configured to obtain the central coordinates of the macular region according to the central coordinates of the optic disc region in combination with the multiple linear regression model, according to the formula:
  • Y is the center coordinate of the macular area
  • X is the center coordinate of the optic disc area
  • W and B are matrix parameters.
  • the macular region detection module 150 includes a multimodal amplicon module (not shown in the figure).
  • a multimodal amplicon module configured to perform multimodal processing of macular images.
  • performing multimodal processing on the macular image includes: performing at least one of limited contrast adaptive histogram enhancement processing, Gaussian filtering enhancement processing, and HSV color space processing on the macular image.
  • the macular region detection module 150 further includes a detection sub-module (not shown in the figure).
  • the detection sub-module can be realized by the designed neural network for image detection and recognition.
  • the neural network (ie, the matching network) includes a feature extraction network, a feature association network, a similarity measurement network, and an attention distribution network that are cascaded in sequence.
  • the network is used to learn the similarity between the target image and the support image, so that a small number of samples can be used to achieve the effect of classifying and detecting the target image.
  • an apparatus 200 for identifying and detecting a macular region in a fundus image includes a processor 210 and a memory 220 for storing instructions executable by the processor 210 .
  • the processor 210 is configured to implement any one of the aforementioned methods for identifying and detecting a macular region in a fundus image when executing the executable instructions.
  • the number of processors 210 may be one or more.
  • the apparatus 200 for positioning and detecting the macular region of the fundus image according to the embodiment of the present application may further include an input device 230 and an output device 240 .
  • the processor 210, the memory 220, the input device 230, and the output device 240 may be connected through a bus, or may be connected in other ways, which are not specifically limited here.
  • the memory 220 can be used to store software programs, computer-executable programs, and various modules, such as programs or modules corresponding to the method for localizing and detecting the macular region of the fundus image in the embodiments of the present application.
  • the processor 210 executes various functional applications and data processing of the apparatus 200 for positioning and detecting the macular region of the fundus image by running the software programs or modules stored in the memory 220 .
  • the input device 230 may be used to receive input numbers or signals. Wherein, the signal may be the generation of a key signal related to user setting and function control of the device/terminal/server.
  • the output device 240 may include a display device such as a display screen.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Multimedia (AREA)
  • Evolutionary Computation (AREA)
  • Radiology & Medical Imaging (AREA)
  • Quality & Reliability (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Software Systems (AREA)
  • Computing Systems (AREA)
  • Artificial Intelligence (AREA)
  • Biomedical Technology (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Molecular Biology (AREA)
  • Mathematical Physics (AREA)
  • Data Mining & Analysis (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Ophthalmology & Optometry (AREA)
  • Human Computer Interaction (AREA)
  • Databases & Information Systems (AREA)
  • Eye Examination Apparatus (AREA)

Abstract

本发明公开一种眼底图像黄斑区域的识别检测方法和装置及设备，方法包括：读取当前待定位检测的眼底图像；采用目标检测模型对眼底图像中的黄斑区域进行检测；在未检测到眼底图像中的黄斑区域时，对眼底图像中的视盘区域进行检测，并基于检测出的视盘区域对黄斑区域进行识别定位；基于黄斑区域的定位结果，由眼底图像中提取出黄斑区域所对应的黄斑图像；对黄斑图像进行多模态处理，将多模态处理得到的图像进行融合得到融合图像，根据融合图像对黄斑区域进行是否合格的检测。其实现了综合考虑黄斑区和视盘区的位置关系进行黄斑区域的识别定位，相较于基于深度学习的黄斑区域定位方法只依赖黄斑区域标注情况的方式，眼底图像的黄斑区域识别更加准确。

Description

眼底图像黄斑区域的识别检测方法和装置及设备
本申请要求于2021年04月30日提交中国专利局、申请号为202110478090.5、发明名称为“眼底图像黄斑区域的识别检测方法和装置及设备”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
本发明涉及医学图像处理技术领域，特别是涉及一种眼底图像黄斑区域的识别检测方法和装置及设备。
背景技术
彩色眼底照是辅助眼底图像检测筛查的重要手段之一。黄斑作为眼底生物结构中最重要的部分，直接影响人们的视力。对于黄斑异常的检测和分类方法目前研究较少，通常情况下，黄斑区域的检测和分类主要包括两部分技术模块，即，眼底黄斑区识别定位模块和黄斑区域异常检测模块。其中，基于深度学习的黄斑区识别定位算法相对比较成熟，只需要对眼底图像中的黄斑位置进行标注和学习，便能够达到黄斑区定位的效果。但是，由于异常的眼底图像中黄斑区域的影像内容丰富多变，仅通过有限的眼底图像训练的深度学习模型来进行黄斑区域的定位经常会出现定位不准或失效的情况，进而也就影响了后面对黄斑区域检测分类的结果，最终使得对眼底图像黄斑区域的定位检测结果的准确率偏低。
发明内容
基于此，本发明的目的是提供一种眼底图像黄斑区域的识别检测方法和装置及设备，可以有效提高眼底图像黄斑区的识别和检测结果的准确率。
为实现上述目的，本发明提供了一种眼底图像黄斑区域的识别检测方法，包括：
读取当前待定位检测的眼底图像；
采用目标检测模型对所述眼底图像中的黄斑区域进行检测；
在未检测到所述眼底图像中的黄斑区域时，对所述眼底图像中的视盘区域进行检测，并基于检测出的所述视盘区域，对所述黄斑区域进行识别定位；
基于所述黄斑区域的识别定位结果，由所述眼底图像中提取出所述黄斑区域所对应的黄斑图像；
对所述黄斑图像进行多模态处理，将多模态处理得到的图像进行融合，得到融合图像，并根据所述融合图像对所述黄斑区域进行是否合格的检测。
可选地，基于检测出的所述视盘区域，对所述黄斑区域进行识别定位，包括：
获取所述视盘区域的中心坐标；
根据所述视盘区域的中心坐标，结合多元线性回归模型得到所述黄斑区域的中心坐标。
可选地，根据所述视盘区域的中心坐标，结合多元线性回归模型得到所述黄斑区域的中心坐标时，根据公式：
Y＝WX+B
计算得到所述黄斑区域的中心坐标；
其中，Y为所述黄斑区域的中心坐标，X为所述视盘区域的中心坐标，W和B均为矩阵参数。
可选地，对所述黄斑图像进行多模态处理包括：对所述黄斑图像进行限制对比度自适应直方图增强处理、高斯滤波增强处理、HSV色彩空间处理中的至少一种。
可选地，根据所述融合图像对所述黄斑区域进行是否合格的检测，包括：
将所述融合图像输入至预先训练好的匹配网络中，由所述匹配网络对所述融合图像进行检测；
其中，所述匹配网络包括依次级联的特征提取网络、特征关联网络、相似性度量网络和注意力分配网络。
本发明还提供了一种眼底图像黄斑区域的识别检测装置，包括图像读取模块、黄斑区域第一定位模块、黄斑区域第二定位模块、黄斑区域提取模块和黄斑区域检测模块；
所述图像读取模块，被配置为读取当前待定位检测的眼底图像；
所述黄斑区域第一定位模块，被配置为采用目标检测模型对所述眼底 图像中的黄斑区域进行检测；
所述黄斑区域第二定位模块，被配置为在所述黄斑区域第一定位模块未检测到所述眼底图像中的黄斑区域时，对所述眼底图像中的视盘区域进行检测，并基于检测出的所述视盘区域，对所述黄斑区域进行识别定位；
所述黄斑区域提取模块，被配置为基于所述黄斑区域的定位结果，由所述眼底图像中提取出所述黄斑区域所对应的黄斑图像；
所述黄斑区域检测模块，被配置为对所述黄斑图像进行多模态处理，将多模态处理得到的图像进行融合，得到融合图像，并根据所述融合图像对所述黄斑区域进行是否合格的检测。
可选地，所述黄斑区域第二定位模块包括：坐标获取子模块和黄斑定位子模块；
所述坐标获取子模块，被配置为获取所述视盘区域的中心坐标；
所述黄斑定位子模块，被配置为根据所述视盘区域的中心坐标，结合多元线性回归模型得到所述黄斑区域的中心坐标。
可选地，所述黄斑定位子模块，被配置为根据所述视盘区域的中心坐标，结合多元线性回归模型得到所述黄斑区域的中心坐标时，根据公式：
Y＝WX+B
计算得到所述黄斑区域的中心坐标；
其中，Y为所述黄斑区域的中心坐标，X为所述视盘区域的中心坐标，W和B均为矩阵参数。
可选地，所述黄斑区域检测模块包括多模态扩增子模块；
所述多模态扩增子模块，被配置为对所述黄斑图像进行多模态处理；
其中，对所述黄斑图像进行多模态处理包括：对所述黄斑图像进行限制对比度自适应直方图增强处理、高斯滤波增强处理、HSV色彩空间处理中的至少一种。
本发明还提供了一种眼底图像黄斑区域的定位检测设备，包括：
处理器；
用于存储处理器可执行指令的存储器；
其中，所述处理器被配置为执行所述可执行指令时实现权利要求1至5中任意一项所述的眼底图像黄斑区域的识别检测方法。
根据本发明提供的具体实施例，本发明公开了以下技术效果：
本发明通过采用目标检测模型对眼底图像中的黄斑区域进行检测，在未检测到眼底图像中的黄斑区域时，对眼底图像中的视盘区域进行检测，并基于检测出的视盘区域对黄斑区域进行识别定位，实现了综合考虑黄斑区和视盘区的位置关系进行黄斑区域的识别定位，相较于基于深度学习的黄斑区域定位方法完全依赖眼底黄斑区域的标注情况的方式，使得在进行眼底图像的黄斑区域的识别定位时更加准确。同时，在检测出黄斑区域后，由眼底图像中提取出黄斑区域所对应的黄斑图像，对黄斑图像进行多模态处理，将多模态处理得到的图像进行融合得到融合图像后，再根据融合图像对黄斑区域进行是否合格的检测。其通过对黄斑图像进行多模态融合处理，从而能够采用基于少样本量学习的方式进行建模，以达到对黄斑图像进行合格和异常分类的效果，极大地提高了黄斑图像的识别能力。
说明书附图
为了更清楚地说明本发明实施例或现有技术中的技术方案，下面将对实施例中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本发明的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动性的前提下，还可以根据这些附图获得其他的附图。
图1为本发明实施例眼底图像黄斑区域的识别检测方法的流程图；
图2为本发明实施例眼底图像黄斑区域的识别检测方法中进行黄斑区域的定位的流程图；
图3为本发明实施例眼底图像黄斑区域的识别检测方法中进行黄斑区域检测时对黄斑图像进行多模处理的示意图；
图4为本发明实施例眼底图像黄斑区域的识别检测方法中对黄斑区域检测时的流程图；
图5为本发明实施例眼底图像黄斑区域的识别检测方法对黄斑区域进行是否合格检测时所采用的匹配网络的网络结构示意图；
图6为本发明实施例眼底图像黄斑区域的识别检测装置的结构框图；
图7为本发明实施例眼底图像的黄斑区域的识别检测设备的结构框图。
具体实施方式
下面将结合本发明实施例中的附图，对本发明实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅仅是本发明一部分实施例，而不是全部的实施例。基于本发明中的实施例，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，都属于本发明保护的范围。
本发明的目的是提供一种眼底图像黄斑区域的识别检测方法和装置及设备，可以有效提高眼底图像黄斑区的识别和检测结果的准确率。
为使本发明的上述目的、特征和优点能够更加明显易懂，下面结合附图和具体实施方式对本发明作进一步详细的说明。
图1示出根据本申请一实施例的眼底图像黄斑区域的识别检测方法的流程图。如图1所示，该方法包括：
步骤S100，读取当前待定位检测的眼底图像。
步骤S200，采用目标检测模型对眼底图像中的黄斑区域进行检测。此处，需要说明的是，所采用的目标检测模型可以为Faster Rcnn神经网络。
步骤S300，在未检测到眼底图像中的黄斑区域时，对眼底图像中的视盘区域进行检测，并基于所检测出的视盘区域，对黄斑区域进行识别定位。
步骤S400，基于黄斑区域的识别定位结果，由眼底图图像中提取出黄斑区域所对应的黄斑图像。
步骤S500，对黄斑图像进行多模态处理，将多模态处理得到的图像进行融合得到融合图像后，根据融合图像对黄斑区域进行是否合格的检测。
由此，本申请实施例的眼底图像黄斑区域的识别检测方法，通过采用目标检测模型对眼底图像中的黄斑区域进行检测，在未检测到眼底图像中的黄斑区域时，对眼底图像中的视盘区域进行检测，并基于检测出的视盘区域对黄斑区域进行识别定位，实现了综合考虑黄斑区和视盘区的位置关系进行黄斑区域的定位，相较于基于深度学习的黄斑区域定位方法完全依赖眼底黄斑区域的标注情况的方式，使得在进行眼底图像的黄斑区域的识别定位时更加准确。同时，在检测出黄斑区域后，由眼底图像中提取出黄 斑区域所对应的黄斑图像，对黄斑图像进行多模态处理，将多模态处理得到的图像进行融合得到融合图像后，再根据融合图像对黄斑区域进行是否合格的检测。其通过对黄斑图像进行多模态融合处理，从而能够采用基于少样本量学习的方式进行建模，以达到对黄斑图像进行合格和异常分类的效果，极大地提高了黄斑图像的识别能力。
其中，在一种可能的实现方式中，在基于检测出的视盘区域对黄斑区域进行识别定位时，可以通过获取视盘区域的中心坐标，然后根据视盘区域的中心坐标，结合多元线性回归模型得到黄斑区域的中心坐标的方式得到。
具体的，在根据视盘区域的中心坐标，结合多元线性回归模型得到黄斑区域的中心坐标时，可以根据公式：
Y＝WX+B
计算得到黄斑区域的中心坐标。其中，Y为黄斑区域的中心坐标，X为视盘区域的中心坐标，W和B均为矩阵参数。此处，需要说明的是，W和B的取值可以通过最小二乘法确定。
即，在采用最小二乘法确定W和B的取值时，可以通过以下矩阵求解公式来实现：
Figure PCTCN2021112998-appb-000001
Figure PCTCN2021112998-appb-000002
其中，根据前面所述可知，W和B均为矩阵参数，Y为黄斑区域的中心坐标，X为视盘区域的中心坐标。X 0为矩阵向量，由X矩阵向量和1向量组成。
同时，还需要说明的是，在通过上述方式，采用多元线性回归模型进行黄斑区域的中心坐标的确定时，还可以分别针对左眼和右眼构建不同的多元线性回归模型，从而基于当前待处理的眼底图像为左眼或右眼的属性采用对应的多元线性回归模型进行黄斑区域中心坐标的确定。
此外，还需要指出的是，根据前面所述，由于在本申请实施例的方法中，对眼底图像进行黄斑区域和视盘区域的检测时采用的是目标检测模型。因此，在对待识别检测的眼底图像进行黄斑区域和视盘区域的识别定位之前，还需要先对目标检测模型进行训练，以使所使用的目标检测模型 收敛。
即，对收集到的样本数据(即，眼底图像数据)进行标注，标注的内容包括黄斑区和视盘区所在眼底图像中的位置(可以用矩形框标注)，以及黄斑区域和视盘区域的中心坐标(可以用坐标点标注)。
然后，将标注后的各样本数据依次输入至目标检测模型(Faster Rcnn)中，对目标检测模型进行训练，使其能够对眼底图像中的黄斑区和视盘区进行检测定位。
其中，参阅图2，在对待定位检测的眼底图像进行黄斑区域的定位时，首先通过步骤S200，采用预先训练好的目标检测模型对眼底图像进行黄斑区域的检测。同时，执行步骤S200’，判断是否检测出黄斑区域。如果直接检测到黄斑区域，则通过步骤S300’，直接定位出黄斑区域在眼底图像中的位置。如果没有检测到黄斑区域，则执行步骤S310，采用训练好的目标检测模型对眼底图像进行视盘区域的检测，并判断是否检测出视盘区域。如果视盘区域也没有检测出来的话，则表明该眼底图像存在问题，不适宜进行黄斑区域的识别分类。此时则通过步骤S340，发出检测失败的提示，并直接读取下一张眼底图像进行黄斑区域的定位检测。如果检测出视盘区域，则通过步骤S320，获取视盘区域的中心坐标，并使用多元线性回归模型，根据定位出的视盘区域的中心坐标进行黄斑区域的中心坐标的计算确定。由此，定位出黄斑区域的中心坐标后，再通过步骤S330，采用预先设置相应尺寸的矩形框或圆形框等框体，以所定位出的黄斑区域的中心坐标为中心，进行黄斑区域的识别定位。
在一种可能的实现方式中，在根据视盘区域的中心坐标定位出黄斑区域的中心坐标后，以预设尺寸大小的矩形框或圆形框等框体进行黄斑区域的识别定位时，可以直接采用所定位出的视盘区域的尺寸大小进行黄斑区域的定位。
其中，采用所定位出的视盘区域的尺寸大小进行黄斑区域的定位时，可以以所确定的黄斑区域的中心坐标为中心，所定位出的视盘区域的1倍—1.5倍大小尺寸进行黄斑区域的裁剪。如：可以以所定位出的视盘区域的1.5倍大小尺寸进行黄斑区域的裁剪。
在通过上述方式定位出眼底图像中的黄斑区域后，即可将该黄斑区域 由眼底图像中剪切并输出，从而提取得到黄斑区域所对应的黄斑图像。
进一步的，通过上述任一方式得到黄斑图像后，即可基于所得到的黄斑图像对黄斑区域进行是否合格的检测。此处，本领域技术人员可以理解的是，对黄斑区域进行是否合格的检测，指的是对所定位出的黄斑区域是否正常，有无异常情况的检测。
其中，根据前面所述，在本申请实施例的方法中，对黄斑区域进行是否合格的检测之前，还包括对黄斑图像进行多模态处理，将多模态处理得到的图像进行融合得到融合图像的操作。
在一种可能的实现方式中，对黄斑图像进行多模态处理包括限制对比度自适应直方图增强处理、高斯滤波增强处理、HSV色彩空间处理中的至少一种。优选的，参阅图3，对黄斑图像进行的多模态处理包括限制对比度自适应直方图增强处理、高斯滤波增强处理和HSV色彩空间处理。
即，通过构建多模态扩增模块，多模态扩增模块包括限制对比度自适应直方图增强处理单元、高斯滤波增强处理单元和HSV色彩空间处理单元，分别由限制对比度自适应直方图增强处理单元、高斯滤波增强处理单元和HSV色彩空间处理单元对黄斑图像进行相应的处理，得到限制对比度自适应直方图增强(CLAHE)图像、高斯滤波增强图像和HSV色彩空间图。
然后，再由多模态扩增模块中的融合单元采用原始图像(即，剪切提取出来的黄斑图像)、限制对比度自适应直方图增强(CLAHE)图像、高斯滤波增强图像和HSV色彩空间图四种模态的图像进行融合，得到融合图像以作为输入图像。进而再将得到的融合图像输入至预先训练好的匹配网络中，由匹配网络对融合图像进行检测，最终得到黄斑图像合格或异常的检测结果。
其中，在将四种模态下的图像进行融合时，可以通过图像通道叠加的方式进行融合。即，对于原始图像(即，剪切提取出来的黄斑图像)、限制对比度自适应直方图增强(CLAHE)图像、高斯滤波增强图像和HSV色彩空间图四种模态的图像，每一种模态的图像均具有3个通道，通过图像通道叠加可得到12个通道的数据作为匹配网络的输入。
此处，应当指出的是，在将得到的融合图像输入至匹配网络中进行是 否合格的检测时，匹配网络也是需要预先训练至收敛的。
其中，参阅图4，对匹配网络的训练具体包括：步骤S001，构建黄斑图像数据集，包括合格黄斑图像(即，正常的黄斑图像)以及异常黄斑图像(如：黄斑前膜、黄斑裂孔、年龄性黄斑变性、中心性浆液性脉络膜视网膜病变、中心性渗出性脉络膜视网膜病变等黄斑病变图像)。
进而通过步骤S002，构建多模态扩增模块，采用原始图像、限制对比度自适应直方图增强(CLAHE)图像、高斯滤波增强图像、HSV色彩空间图四种模态的图像进行融合作为模型训练的输入图像。
然后，执行步骤S003，将多模融合的图像分为支撑数据集和目标数据集，并通过步骤S004，将支撑数据集和目标数据集依次输入至匹配网络中进行训练学习以及测试应用。
进一步的，参阅图5，在本申请实施例的方法中，所设计采用的匹配网络的网络结构可以为包括依次级联的特征提取网络、特征关联网络、相似性度量网络和注意力分配网络。其中，特征提取网络可以使用基于卷积神经网络分别对支撑图像和目标图像进行特征提取(如：可以采用ResNet50)，特征关联网络使用循环神经网络来解析支撑图像和目标图像内部和外部的关联性(如：可以采用16层LSTM单元网络)，相似性度量模块在网络中不进行训练(如：可以采用余弦相似性进行度量)，注意力分配网络用于提高模型的训练效率和性能(如：可以采用softmax函数进行注意力分配)。
通过该网络用于学习目标图像与支撑图像的相似性，进而达到使用少量样本达到对目标图像进行分类的效果。
通过将多模融合得到的融合图像输入至训练好的匹配网络中，由匹配网络对融合图像依次进行特征提取、特征关联和相似性度量以及注意力分配后即可得到最终的分类结果。
相应的，基于前面任一所述的眼底图像黄斑区域的识别检测方法，本申请还提供了一种眼底图像黄斑区域的识别检测装置。由于本申请提供的眼底图像黄斑区域的识别检测装置的工作原理，与本申请的眼底图像黄斑区域的识别检测方法的原理相同或相似，因此重复之处不再赘述。
参阅图6，本申请提供的眼底图像黄斑区域的识别检测装置100，包 括图像读取模块110、黄斑区域第一定位模块120、黄斑区域第二定位模块130、黄斑区域提取模块140和黄斑区域检测模块150。其中，图像读取模块110，被配置为读取当前待定位检测的眼底图像。黄斑区域第一定位模块120，被配置为采用目标检测模型对眼底图像中的黄斑区域进行检测。黄斑区域第二定位模块130，被配置为在黄斑区域第一定位模块120未检测到眼底图像中的黄斑区域时，对眼底图像中的视盘区域进行检测，并基于检测出的视盘区域，对黄斑区域进行识别定位。黄斑区域提取模块140，被配置为基于黄斑区域的定位结果，由眼底图像中提取出黄斑区域所对应的黄斑图像。黄斑区域检测模块150，被配置为对黄斑图像进行多模态处理，将多模态处理得到的图像进行融合，得到融合图像，并根据融合图像对黄斑区域进行是否合格的检测。
在一种可能的实现方式中，黄斑区域第二定位模块130包括：坐标获取子模块和黄斑定位子模块(图中未示出)。其中，坐标获取子模块，被配置为获取视盘区域的中心坐标。黄斑定位子模块，被配置为根据视盘区域的中心坐标，结合多元线性回归模型得到黄斑区域的中心坐标。
在一种可能的实现方式中，黄斑定位子模块，被配置为根据视盘区域的中心坐标，结合多元线性回归模型得到黄斑区域的中心坐标时，根据公式：
Y＝WX+B
计算得到黄斑区域的中心坐标。其中，Y为黄斑区域的中心坐标，X为视盘区域的中心坐标，W和B均为矩阵参数。
在一种可能的实现方式中，黄斑区域检测模块150包括多模态扩增子模块(图中未示出)。多模态扩增子模块，被配置为对黄斑图像进行多模态处理。其中，对黄斑图像进行多模态处理包括：对黄斑图像进行限制对比度自适应直方图增强处理、高斯滤波增强处理、HSV色彩空间处理中的至少一种。
进一步的，在本申请实施例的眼底图像黄斑区域的定位检测装置100中，黄斑区域检测模块150还包括检测子模块(图中未示出)。此处，需要指出的是，检测子模块可以通过所设计的图像检测识别的神经网络来实现。
具体的，参阅图5，该神经网络(即，匹配网络)包括依次级联的特征提取网络、特征关联网络、相似性度量网络和注意力分配网络。通过该网络用于学习目标图像与支撑图像的相似性，从而能够达到使用少量样本达到对目标图像进行分类检测的效果。
更进一步地，根据本申请的另一方面，还提供了一种眼底图像黄斑区域的识别检测设备200。参阅图7，本申请实施例的眼底图像黄斑区域的识别检测设备200包括处理器210以及用于存储处理器210可执行指令的存储器220。其中，处理器210被配置为执行可执行指令时实现前面任一所述的眼底图像黄斑区域的识别检测方法。
此处，应当指出的是，处理器210的个数可以为一个或多个。同时，在本申请实施例的眼底图像黄斑区域的定位检测设备200中，还可以包括输入装置230和输出装置240。其中，处理器210、存储器220、输入装置230和输出装置240之间可以通过总线连接，也可以通过其他方式连接，此处不进行具体限定。
存储器220作为一种计算机可读存储介质，可用于存储软件程序、计算机可执行程序和各种模块，如：本申请实施例的眼底图像黄斑区域的定位检测方法所对应的程序或模块。处理器210通过运行存储在存储器220中的软件程序或模块，从而执行眼底图像黄斑区域的定位检测设备200的各种功能应用及数据处理。
输入装置230可用于接收输入的数字或信号。其中，信号可以为产生与设备/终端/服务器的用户设置以及功能控制有关的键信号。输出装置240可以包括显示屏等显示设备。
本说明书中各个实施例采用递进的方式描述，每个实施例重点说明的都是与其他实施例的不同之处，各个实施例之间相同相似部分互相参见即可。
本文中应用了具体个例对本发明的原理及实施方式进行了阐述，以上实施例的说明只是用于帮助理解本发明的方法及其核心思想；同时，对于本领域的一般技术人员，依据本发明的思想，在具体实施方式及应用范围上均会有改变之处。综上所述，本说明书内容不应理解为对本发明的限制。

Claims (10)

  1. 一种眼底图像黄斑区域的识别检测方法，其特征在于，包括：
    读取当前待定位检测的眼底图像；
    采用目标检测模型对所述眼底图像中的黄斑区域进行检测；
    在未检测到所述眼底图像中的黄斑区域时，对所述眼底图像中的视盘区域进行检测，并基于检测出的所述视盘区域，对所述黄斑区域进行识别定位；
    基于所述黄斑区域的识别定位结果，由所述眼底图像中提取出所述黄斑区域所对应的黄斑图像；
    对所述黄斑图像进行多模态处理，将多模态处理得到的图像进行融合，得到融合图像，并根据所述融合图像对所述黄斑区域进行是否合格的检测。
  2. 根据权利要求1所述的眼底图像黄斑区域的识别检测方法，其特征在于，基于检测出的所述视盘区域，对所述黄斑区域进行识别定位，包括：
    获取所述视盘区域的中心坐标；
    根据所述视盘区域的中心坐标，结合多元线性回归模型得到所述黄斑区域的中心坐标。
  3. 根据权利要求2所述的眼底图像黄斑区域的识别检测方法，其特征在于，根据所述视盘区域的中心坐标，结合多元线性回归模型得到所述黄斑区域的中心坐标时，根据公式：
    Y＝WX+B
    计算得到所述黄斑区域的中心坐标；
    其中，Y为所述黄斑区域的中心坐标，X为所述视盘区域的中心坐标，W和B均为矩阵参数。
  4. 根据权利要求1至3任一项所述的眼底图像黄斑区域的识别检测方法，其特征在于，对所述黄斑图像进行多模态处理包括：对所述黄斑图像进行限制对比度自适应直方图增强处理、高斯滤波增强处理、HSV色彩空间处理中的至少一种。
  5. 根据权利要求1所述的眼底图像黄斑区域的识别检测方法，其特征在于，根据所述融合图像对所述黄斑区域进行是否合格的检测，包括：
    将所述融合图像输入至预先训练好的匹配网络中，由所述匹配网络对所述融合图像进行检测；
    其中，所述匹配网络包括依次级联的特征提取网络、特征关联网络、相似性度量网络和注意力分配网络。
  6. 一种眼底图像黄斑区域的识别检测装置，其特征在于，包括图像读取模块、黄斑区域第一定位模块、黄斑区域第二定位模块、黄斑区域提取模块和黄斑区域检测模块；
    所述图像读取模块，被配置为读取当前待定位检测的眼底图像；
    所述黄斑区域第一定位模块，被配置为采用目标检测模型对所述眼底图像中的黄斑区域进行检测；
    所述黄斑区域第二定位模块，被配置为在所述黄斑区域第一定位模块未检测到所述眼底图像中的黄斑区域时，对所述眼底图像中的视盘区域进行检测，并基于检测出的所述视盘区域，对所述黄斑区域进行识别定位；
    所述黄斑区域提取模块，被配置为基于所述黄斑区域的定位结果，由所述眼底图像中提取出所述黄斑区域所对应的黄斑图像；
    所述黄斑区域检测模块，被配置为对所述黄斑图像进行多模态处理，将多模态处理得到的图像进行融合，得到融合图像，并根据所述融合图像对所述黄斑区域进行是否合格的检测。
  7. 根据权利要求6所述的眼底图像黄斑区域的识别检测装置，其特征在于，所述黄斑区域第二定位模块包括：坐标获取子模块和黄斑定位子模块；
    所述坐标获取子模块，被配置为获取所述视盘区域的中心坐标；
    所述黄斑定位子模块，被配置为根据所述视盘区域的中心坐标，结合多元线性回归模型得到所述黄斑区域的中心坐标。
  8. 根据权利要求7所述的眼底图像黄斑区域的识别检测装置，其特征在于，所述黄斑定位子模块，被配置为根据所述视盘区域的中心坐标，结合多元线性回归模型得到所述黄斑区域的中心坐标时，根据公式：
    Y＝WX+B
    计算得到所述黄斑区域的中心坐标；
    其中，Y为所述黄斑区域的中心坐标，X为所述视盘区域的中心坐标， W和B均为矩阵参数。
  9. 根据权利要求6至8任一项所述的眼底图像黄斑区域的识别检测装置，其特征在于，所述黄斑区域检测模块包括多模态扩增子模块；
    所述多模态扩增子模块，被配置为对所述黄斑图像进行多模态处理；
    其中，对所述黄斑图像进行多模态处理包括：对所述黄斑图像进行限制对比度自适应直方图增强处理、高斯滤波增强处理、HSV色彩空间处理中的至少一种。
  10. 一种眼底图像黄斑区域的定位检测设备，其特征在于，包括：
    处理器；
    用于存储处理器可执行指令的存储器；
    其中，所述处理器被配置为执行所述可执行指令时实现权利要求1至5中任意一项所述的眼底图像黄斑区域的识别检测方法。
PCT/CN2021/112998 2021-04-30 2021-08-17 眼底图像黄斑区域的识别检测方法和装置及设备 WO2022227342A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US18/010,804 US11908137B2 (en) 2021-04-30 2021-08-17 Method, device and equipment for identifying and detecting macular region in fundus image

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202110478090.5 2021-04-30
CN202110478090.5A CN112991343B (zh) 2021-04-30 2021-04-30 眼底图像黄斑区域的识别检测方法和装置及设备

Publications (1)

Publication Number Publication Date
WO2022227342A1 true WO2022227342A1 (zh) 2022-11-03

Family

ID=76336709

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/112998 WO2022227342A1 (zh) 2021-04-30 2021-08-17 眼底图像黄斑区域的识别检测方法和装置及设备

Country Status (3)

Country Link
US (1) US11908137B2 (zh)
CN (1) CN112991343B (zh)
WO (1) WO2022227342A1 (zh)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112991343B (zh) 2021-04-30 2021-08-13 北京至真互联网技术有限公司 眼底图像黄斑区域的识别检测方法和装置及设备
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

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20170296049A1 (en) * 2016-04-15 2017-10-19 Canon Kabushiki Kaisha Image processing apparatus
CN111046717A (zh) * 2019-10-11 2020-04-21 平安科技（深圳）有限公司 眼底图像黄斑中心定位方法、装置、电子设备及存储介质
CN111951933A (zh) * 2020-08-07 2020-11-17 平安科技（深圳）有限公司 眼底彩照图像分级方法、装置、计算机设备及存储介质
CN112017187A (zh) * 2020-11-02 2020-12-01 平安科技（深圳）有限公司 眼底图像黄斑中心定位方法及装置、服务器、存储介质
CN112991343A (zh) * 2021-04-30 2021-06-18 北京至真互联网技术有限公司 眼底图像黄斑区域的识别检测方法和装置及设备

Family Cites Families (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP5582772B2 (ja) * 2009-12-08 2014-09-03 キヤノン株式会社 画像処理装置及び画像処理方法
US10405739B2 (en) * 2015-10-23 2019-09-10 International Business Machines Corporation Automatically detecting eye type in retinal fundus images
CN106650596A (zh) * 2016-10-10 2017-05-10 北京新皓然软件技术有限责任公司 一种眼底图像分析方法、装置及系统
CN109662686B (zh) * 2019-02-01 2022-02-25 北京致远慧图科技有限公司 一种眼底黄斑定位方法、装置、系统及存储介质
CN109784337B (zh) * 2019-03-05 2022-02-22 北京康夫子健康技术有限公司 一种黄斑区识别方法、装置及计算机可读存储介质
CN109886955A (zh) * 2019-03-05 2019-06-14 百度在线网络技术（北京）有限公司 用于处理眼底图像的方法和装置
CN110739071B (zh) * 2019-10-10 2022-05-31 北京致远慧图科技有限公司 一种视盘黄斑联合定位模型的确定方法、装置及存储介质
CN110751637A (zh) * 2019-10-14 2020-02-04 北京至真互联网技术有限公司 糖尿病视网膜病变检测系统、方法、设备和训练系统

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20170296049A1 (en) * 2016-04-15 2017-10-19 Canon Kabushiki Kaisha Image processing apparatus
CN111046717A (zh) * 2019-10-11 2020-04-21 平安科技（深圳）有限公司 眼底图像黄斑中心定位方法、装置、电子设备及存储介质
CN111951933A (zh) * 2020-08-07 2020-11-17 平安科技（深圳）有限公司 眼底彩照图像分级方法、装置、计算机设备及存储介质
CN112017187A (zh) * 2020-11-02 2020-12-01 平安科技（深圳）有限公司 眼底图像黄斑中心定位方法及装置、服务器、存储介质
CN112991343A (zh) * 2021-04-30 2021-06-18 北京至真互联网技术有限公司 眼底图像黄斑区域的识别检测方法和装置及设备

Also Published As

Publication number Publication date
CN112991343B (zh) 2021-08-13
US20230274419A1 (en) 2023-08-31
CN112991343A (zh) 2021-06-18
US11908137B2 (en) 2024-02-20

Similar Documents

Publication Publication Date Title
WO2020215672A1 (zh) 医学图像病灶检测定位方法、装置、设备及存储介质
WO2021003821A1 (zh) 一种肾小球病理切片图像的细胞检测方法、装置及设备
WO2022088628A1 (zh) 缺陷检测方法、装置、计算机设备及存储介质
CN110675487B (zh) 基于多角度二维人脸的三维人脸建模、识别方法及装置
WO2022227342A1 (zh) 眼底图像黄斑区域的识别检测方法和装置及设备
JP2020507836A (ja) 重複撮像を予測した手術アイテムの追跡
CN111091562B (zh) 一种消化道病灶大小测量方法及系统
WO2014032496A1 (zh) 一种人脸特征点定位方法、装置及存储介质
TWI669664B (zh) 眼睛狀態檢測系統及眼睛狀態檢測系統的操作方法
CN113826143A (zh) 特征点检测
CN111597884A (zh) 面部动作单元识别方法、装置、电子设备及存储介质
US10395091B2 (en) Image processing apparatus, image processing method, and storage medium identifying cell candidate area
CN108052909B (zh) 一种基于心血管oct影像的薄纤维帽斑块自动检测方法和装置
CN110580466A (zh) 婴儿踢被子行为识别方法、装置、计算机设备及存储介质
WO2022198898A1 (zh) 图像分类方法和装置及设备
CN110543823B (zh) 基于残差网络的行人再识别方法、装置和计算机设备
WO2015131710A1 (zh) 人眼定位方法及装置
CN110378333B (zh) 一种sd-oct图像黄斑中央凹中心定位方法
CN116958679A (zh) 一种基于弱监督的目标检测方法及相关设备
CN108510497B (zh) 视网膜图像病灶信息的显示方法及显示装置
CN116052230A (zh) 一种掌静脉识别方法、装置、设备及存储介质
CN113344911B (zh) 一种结石尺寸的测量方法以及测量装置
CN110751163A (zh) 目标定位方法及其装置、计算机可读存储介质和电子设备
CN114330484A (zh) 弱监督学习糖尿病视网膜病变分级与病灶识别方法及系统
CN113658107A (zh) 一种基于ct图像的肝脏病灶诊断方法及装置

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: 21938806

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: 21938806

Country of ref document: EP

Kind code of ref document: A1