CN116824116A - Super wide angle fundus image identification method, device, equipment and storage medium - Google Patents
Super wide angle fundus image identification method, device, equipment and storage medium Download PDFInfo
- Publication number
- CN116824116A CN116824116A CN202310761433.8A CN202310761433A CN116824116A CN 116824116 A CN116824116 A CN 116824116A CN 202310761433 A CN202310761433 A CN 202310761433A CN 116824116 A CN116824116 A CN 116824116A
- Authority
- CN
- China
- Prior art keywords
- information
- image
- target
- ultra
- output
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 42
- 238000000605 extraction Methods 0.000 claims abstract description 78
- 238000007781 pre-processing Methods 0.000 claims abstract description 25
- 239000013598 vector Substances 0.000 claims abstract description 15
- 238000012549 training Methods 0.000 claims description 18
- 238000004590 computer program Methods 0.000 claims description 17
- 230000003044 adaptive effect Effects 0.000 claims description 7
- 238000012545 processing Methods 0.000 abstract description 4
- 238000013135 deep learning Methods 0.000 abstract description 2
- 201000010099 disease Diseases 0.000 description 14
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 14
- 230000008569 process Effects 0.000 description 6
- 230000000007 visual effect Effects 0.000 description 6
- 238000004891 communication Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 5
- 238000011176 pooling Methods 0.000 description 4
- 208000035719 Maculopathy Diseases 0.000 description 3
- 230000009471 action Effects 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 208000030533 eye disease Diseases 0.000 description 3
- 208000002780 macular degeneration Diseases 0.000 description 3
- 210000001525 retina Anatomy 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 238000012935 Averaging Methods 0.000 description 2
- 206010012667 Diabetic glaucoma Diseases 0.000 description 2
- 206010012689 Diabetic retinopathy Diseases 0.000 description 2
- 208000010412 Glaucoma Diseases 0.000 description 2
- 208000034700 Vitreous opacities Diseases 0.000 description 2
- 238000013459 approach Methods 0.000 description 2
- 238000004422 calculation algorithm Methods 0.000 description 2
- 238000013145 classification model Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000007667 floating Methods 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 238000002604 ultrasonography Methods 0.000 description 2
- 230000001133 acceleration Effects 0.000 description 1
- 230000032683 aging Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 210000000720 eyelash Anatomy 0.000 description 1
- 210000000744 eyelid Anatomy 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
Classifications
-
- 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
-
- 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
- G06N3/0455—Auto-encoder networks; Encoder-decoder 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
-
- 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
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
-
- 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
- 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
- G06V10/84—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using probabilistic graphical models from image or video features, e.g. Markov models or Bayesian networks
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Software Systems (AREA)
- Multimedia (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Mathematical Physics (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- Databases & Information Systems (AREA)
- Probability & Statistics with Applications (AREA)
- Image Analysis (AREA)
- Eye Examination Apparatus (AREA)
Abstract
The application discloses a method, a device, equipment and a storage medium for identifying ultra-wide-angle fundus images, which relate to the technical field of image processing and deep learning and comprise the following steps: performing image preprocessing operation on the original ultra-wide angle fundus image to obtain a preprocessed image; performing feature extraction on the preprocessed image by using a pre-trained feature extraction model to obtain target coding information; decoding target coding information through a bilinear decoder model to obtain an output probability value information set; and determining the category with the maximum probability value in the probability value information set as the final prediction category corresponding to the original ultra-wide angle fundus image. According to the application, the feature vector is obtained by extracting the feature of the preprocessed image through the pre-trained feature extraction model, multiple complex algebra operation is performed by utilizing the bilinear decoder model, a plurality of probability values are output, the final recognition result is determined by comparing the sizes of the probability values, the probability of false recognition caused by manual film reading is reduced, and the precision of ultra-wide angle fundus image recognition is improved.
Description
Technical Field
The present invention relates to the field of image processing technology and the field of deep learning technology, and in particular, to a method, a device, an apparatus, and a storage medium for identifying an ultra-wide-angle fundus image.
Background
With the continuous acceleration of population aging, various fundus diseases spread more rapidly if intervention is not performed in time, and the number of current ophthalmologists and medical equipment are insufficient to meet the demands of such huge population of ocular patients.
The existing fundus disease image recognition system is mainly used for recognizing based on common fundus color Doppler ultrasound images, the retina fundus visual range of the images is narrow, and the retina fundus visual range is only 30-75 degrees generally. The narrower visual field does not provide complete fundus information and increases the probability of misrecognition. The ultra-wide angle fundus image is less in acquisition difficulty, the visual range of the fundus can reach 200 degrees, more information can be used for judging fundus diseases, and the judgment precision is higher. However, the current fundus image identification based on ultra-wide angle adopts a manual film reading mode, is greatly limited by the number of clinicians and clinical experience, and has higher probability of image identification errors and lower accuracy of image identification during manual film reading.
Disclosure of Invention
In view of the above, an object of the present application is to provide a method, apparatus, device, and storage medium for identifying an ultra-wide-angle fundus image, which can reduce the probability of erroneous identification caused by manual reading and improve the accuracy of identifying an ultra-wide-angle fundus image. The specific scheme is as follows:
in a first aspect, the application discloses a method for identifying an ultra-wide angle fundus image, which comprises the following steps:
performing image preprocessing operation on the original ultra-wide angle fundus image to obtain a preprocessed image;
extracting features of the preprocessed image by utilizing a pre-trained feature extraction model to obtain target coding information;
decoding the target coding information through a bilinear decoder model to obtain a probability value information set output by the bilinear decoder model;
and determining the category with the maximum probability value in the probability value information set as the final prediction category corresponding to the original ultra-wide angle fundus image.
Optionally, the performing an image preprocessing operation on the original ultra-wide angle fundus image to obtain a preprocessed image includes:
extracting a target region from the original ultra-wide angle fundus image based on an adaptive ROI region rough extraction component; the target area is an area formed by pixel points meeting preset pixel conditions;
Judging whether the target area meets the quality requirement of a preset area or not;
if the target area meets the preset area quality requirement, adjusting the brightness of the target area based on a preset brightness adjustment rule to obtain the preprocessed image;
if the target area does not meet the preset area quality requirement or the target area is not extracted, performing maximum ellipse approximate fitting to obtain a fitting area;
and determining the fitting area as the target area, and re-entering the step of adjusting the brightness of the target area based on a preset brightness adjustment rule to obtain the preprocessed image.
Optionally, the adjusting the brightness of the target area based on a preset brightness adjustment rule to obtain the preprocessed image includes:
acquiring the current brightness of the target area, and determining a current brightness interval corresponding to the current brightness;
determining a current brightness adjustment strategy corresponding to the current brightness interval based on a preset brightness adjustment strategy determination rule;
and adjusting the brightness of the target area by utilizing the current brightness adjustment strategy so as to obtain the preprocessed image.
Optionally, the performing maximum ellipse approximate fitting to obtain a fitting area includes:
Calculating long axis length information and short axis length information corresponding to the length and the width of the original ultra-wide angle fundus image;
determining center point position information based on the reference point information;
an elliptical region is generated based on the center point position information, the long axis length information, and the short axis length information, and the elliptical region is determined as the fitting region.
Optionally, before the feature extraction is performed on the preprocessed image by using the pre-trained feature extraction model to obtain the target coding information, the method further includes:
training an original feature extraction model by using an open source data set to obtain an initial feature extraction model with initial weight;
training the initial feature extraction model by using a super-wide-angle fundus image dataset to obtain the pre-trained feature extraction model with target weights;
correspondingly, the feature extraction of the preprocessed image by using the pre-trained feature extraction model to obtain target coding information comprises the following steps:
extracting feature information meeting preset feature requirements from the preprocessed image by using the pre-trained feature extraction model with the target weight;
and encoding the characteristic information in a vector form to obtain the target encoding information.
Optionally, the decoding the target coding information by the bilinear decoder model to obtain the probability value information set output by the bilinear decoder model includes:
inputting the target coding information into a first channel and a second channel of the bilinear decoder model;
acquiring first output information output by the first channel and second output information output by the second channel;
carrying out space dimension feature stacking on the first output information and the second output information to obtain a stacked feature layer;
and executing preset prediction operation through the stacked feature layers to output the probability value information set.
Optionally, the inputting the target coding information into the first channel and the second channel of the bilinear decoder model includes:
inputting the target coding information into the first channel of the bilinear decoder model, and performing co-scale convolution according to a convolution kernel in the first channel to obtain convolved information;
calculating a first attention weight of a current feature layer based on the convolved information and the target coding information;
if the current feature layer is the last layer, calculating the first output information output by the first channel based on the target coding information and the first attention weight of the last layer; wherein the first output information is a weighted attention feature;
Inputting the target coding information into the second channel of the bilinear decoder model, and calculating an average value of the target coding information to obtain a second attention weight;
the second output information of the second channel output is calculated based on the second attention weight and the target encoding information.
In a second aspect, the present application discloses an ultra-wide angle fundus image identifier, comprising:
the image preprocessing module is used for performing image preprocessing operation on the original ultra-wide angle fundus image so as to obtain a preprocessed image;
the feature extraction module is used for carrying out feature extraction on the preprocessed image by utilizing a pre-trained feature extraction model so as to obtain target coding information;
the decoding module is used for decoding the target coding information through the bilinear decoder model to obtain a probability value information set output by the bilinear decoder model;
and the prediction category determining module is used for determining the category with the largest probability value in the probability value information set as the final prediction category corresponding to the original ultra-wide angle fundus image.
In a third aspect, the present application discloses an electronic device, comprising:
a memory for storing a computer program;
And a processor for executing the computer program to implement the steps of the ultra-wide angle fundus image identification method as disclosed above.
In a fourth aspect, the present application discloses a computer-readable storage medium for storing a computer program; wherein the computer program when executed by the processor implements the ultra-wide angle fundus image identification method as previously disclosed.
It can be seen that the present application provides a method for identifying an ultra-wide angle fundus image, comprising: performing image preprocessing operation on the original ultra-wide angle fundus image to obtain a preprocessed image; extracting features of the preprocessed image by utilizing a pre-trained feature extraction model to obtain target coding information; decoding the target coding information through a bilinear decoder model to obtain a probability value information set output by the bilinear decoder model; and determining the category with the maximum probability value in the probability value information set as the final prediction category corresponding to the original ultra-wide angle fundus image. Therefore, the feature extraction model is used for extracting the features of the preprocessed image to obtain the feature vector, the bilinear decoder model is used for carrying out multiple complex algebraic operations on the feature vector, a plurality of probability values are output, a final recognition result is determined by comparing the sizes of the probability values, the probability of false recognition caused by manual film reading is reduced, and the precision of ultra-wide angle fundus image recognition is improved.
Drawings
In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings that are required to be used in the embodiments or the description of the prior art will be briefly described below, and it is obvious that the drawings in the following description are only embodiments of the present application, and that other drawings can be obtained according to the provided drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flow chart of a method for identifying an ultra-wide angle fundus image, disclosed by the application;
FIG. 2 is a flowchart of a method for identifying an ultra-wide angle fundus image according to the present application;
FIG. 3 is a schematic diagram of a preprocessing flow of an ultra-wide angle fundus image disclosed by the application;
FIG. 4 is a flowchart of a method for identifying an ultra-wide angle fundus image according to the present application;
FIG. 5 is a flow chart of bilinear feature decoding prediction in accordance with the present disclosure;
FIG. 6 is a schematic structural diagram of an ultra-wide angle fundus image identifier according to the present application;
fig. 7 is a block diagram of an electronic device according to the present application.
Detailed Description
The following description of the embodiments of the present application will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present application, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the application without making any inventive effort, are intended to be within the scope of the application.
Currently, the existing fundus disease image recognition system is mainly used for recognizing based on common fundus color ultra-wide-angle fundus images, and the retina fundus visible range of the ultra-wide-angle fundus images is narrow, usually only 30-75 degrees. The narrower visual field does not provide complete fundus information and increases the probability of misrecognition. The ultra-wide-angle fundus image of the ultra-wide-angle fundus is less in acquisition difficulty, the visual range of the fundus can reach 200 degrees, more information can be used for judging fundus diseases, and the judgment precision is higher. However, the current ultra-wide-angle fundus image recognition based on ultra-wide angle is realized by adopting a manual film reading mode, is greatly limited by the number of clinicians and clinical experience, and has higher probability of image recognition error and lower accuracy of ultra-wide-angle fundus image recognition during manual film reading. Therefore, the application provides the ultra-wide-angle fundus image identification method, which can reduce the probability of false identification caused by manual film reading and improve the accuracy of ultra-wide-angle fundus image identification.
The embodiment of the application discloses a super-wide-angle fundus image identification method, which is shown in fig. 1 and comprises the following steps:
step S11: and performing image preprocessing operation on the original ultra-wide angle fundus image to obtain a preprocessed image.
In this embodiment, an image preprocessing operation is performed on an original ultra-wide angle fundus image to obtain a preprocessed image. Specifically, a target region is extracted from the original ultra-wide angle fundus image based on a self-adaptive ROI region rough extraction component; the target area is an area formed by pixel points meeting preset pixel conditions; judging whether the target area meets the quality requirement of a preset area or not; if the target area meets the preset area quality requirement, adjusting the brightness of the target area based on a preset brightness adjustment rule to obtain the preprocessed image; if the target area does not meet the preset area quality requirement or the target area is not extracted, performing maximum ellipse approximate fitting to obtain a fitting area; and determining the fitting area as the target area, and re-entering the step of adjusting the brightness of the target area based on a preset brightness adjustment rule to obtain the preprocessed image. It may be understood that the target area is an area formed by pixel points satisfying a preset pixel condition, for example, if the characteristic of the ultra-wide-angle fundus image itself is a pixel value, the pixel point greater than the preset pixel threshold is determined as a pixel point sufficient to the preset pixel condition.
It can be appreciated that adjusting the brightness of the target area based on a preset brightness adjustment rule to obtain the preprocessed image includes: acquiring the current brightness of the target area, and determining a current brightness interval corresponding to the current brightness; determining a current brightness adjustment strategy corresponding to the current brightness interval based on a preset brightness adjustment strategy determination rule; and adjusting the brightness of the target area by utilizing the current brightness adjustment strategy so as to obtain the preprocessed image. Performing a maximum ellipse approximation fit to obtain a fit region, comprising: calculating long axis length information and short axis length information corresponding to the length and the width of the original ultra-wide angle fundus image; determining center point position information based on the reference point information; an elliptical region is generated based on the center point position information, the long axis length information, and the short axis length information, and the elliptical region is determined as the fitting region.
As shown in fig. 2, the present invention is divided into three main modules: an image preprocessing module (comprising adaptive ROI region coarse extraction, maximum inscribed ellipse approximate fitting and adaptive brightness adjustment); the feature extraction and coding module; and a feature decoding and disease type prediction module. Before an image preprocessing module, the original ultra-wide-angle fundus effect is obtained, a 6-class training set and a 6-class testing set are constructed, data balance operation is carried out, and each class of data volume is guaranteed to be close to 1:1. In the feature extraction and coding module, feature extraction is carried out by utilizing the ResNet50, firstly, the initial weight is obtained by pre-training on an image Net open source data set, and then the initial weight is adjusted in training of a super wide angle data set. In the feature decoding and disease type prediction module, feature vectors coded by ResNet50 are received, features are decoded through a double-branch model, and a prediction result is output. The modules are closely connected and decoupled, namely the output of the last module is the input of the next module, but each module is mutually independent when executing corresponding functions, and the whole system realizes the functions of inputting an ultra-wide-angle fundus image and outputting fundus disease seeds corresponding to the highest probability value of the ultra-wide-angle fundus image through a series of processes.
A module capable of automatically positioning and dividing the fundus area of the ultra-wide-angle fundus image is constructed in the image preprocessing module, an inscribed ellipse fitting module is added for the ultra-wide-angle fundus image with low brightness or serious distortion of individual original images, fundus area fitting is performed on the ultra-wide-angle fundus image difficult to be positioned correctly to the greatest extent, and strong usability of overall data is ensured. The module contains 3 components: an adaptive ROI region rough extraction component, a maximum inscribed ellipse approximate fitting and an adaptive brightness adjustment. It can be understood that the ultra-wide-angle fundus image of the ultra-wide-angle fundus is entered into the self-adaptive ROI region coarse extraction component, based on machine learning and a traditional image processing method, a KMENAS clustering model is firstly constructed, n_clusters of the KMENS clustering model is set to be 2, two clusters of clustering tasks are represented, KMEANS is an unsupervised clustering algorithm, according to the characteristic of the ultra-wide-angle fundus image (namely, the pixel values of fundus regions are relatively close and the pixel values of other regions are relatively changed greatly), then the characteristic is automatically captured through KMEANS, the fundus region and the other regions are divided into 2 different clusters, and fundus and background segmentation can be normally realized for the ultra-wide-angle fundus image with normal shooting quality for most of the ultra-wide-angle fundus image, so that the ROI region coarse extraction is realized. Then, according to the effect of the previous ROI region extraction, it is determined whether to perform maximum ellipse approximate fitting, as shown in fig. 3, if the ROI region quality extracted in the first step is poor (for example, the ROI region area or the position does not meet the expectation), then further processing is performed by using a second component, which calculates the long axis length/short axis length s of the ultra-wide angle fundus image according to the length and width of the original image (for example, taking half of the length of the original image as the long axis length and half of the width of the original image as the short axis length), and obtains the center point position (x, y) (usually, the preset reference point position, for example, (0, 0)) to generate a maximum inscribed ellipse mask of the fundus image, and dividing the mask region corresponding to the original image as the approximate fitting, and determining the approximate fitting region as the rough extracted ROI region (i.e., the target region). Finally, based on the brightness adjustment algorithm provided by opencv, firstly calculating the brightness of an input image, namely a target area, then judging which level the brightness value is in (the level refers to a brightness threshold value and is obtained through multiple manual measurement and calculation), and adopting brightness improvement strategies with different degrees for the brightness of different levels until the brightness of the finally output image is basically at the same level. The component mainly aims to lighten a relatively dim ultra-wide angle fundus image, simultaneously reduce the brightness of the ultra-wide angle fundus image with overhigh brightness, and finally improve the prediction precision of a follow-up model.
As shown in fig. 3, if the target area meets the quality requirement of the preset area, or the maximum inscribed ellipse approximate fitting operation is performed to obtain an approximate fitting area, the brightness of the target area is adjusted based on a preset brightness adjustment rule.
Step S12: and extracting the characteristics of the preprocessed image by utilizing a pre-trained characteristic extraction model so as to obtain target coding information.
In this embodiment, after an image preprocessing operation is performed on an original ultra-wide angle fundus image to obtain a preprocessed image, feature extraction is performed on the preprocessed image by using a feature extraction model that is trained in advance to obtain target encoding information. Specifically, extracting feature information meeting a preset feature requirement from the preprocessed image by using the pre-trained feature extraction model with the target weight; and encoding the characteristic information in a vector form to obtain the target encoding information.
It can be appreciated that, before the feature extraction is performed on the pre-processed image using the pre-trained feature extraction model, the original feature extraction model is trained using the open source data set to obtain an initial feature extraction model with initial weights; training the initial feature extraction model by using the ultra-wide angle fundus image dataset to obtain the pre-trained feature extraction model with target weights. For example, the ResNet50 feature extraction module adopts an open source model ResNet50, and the model is trained on a pre-training dataset (namely, an open source dataset) imageNet, because the dataset comprises more than 1400 ten thousand images, the number of the covered categories is 1000, the ResNet50 after training of a large amount of data has better initial weight, and the ResNet50 can show much better performance than the ResNet50 initialized by random weight in other personalized tasks. Training is carried out on the current ultra-wide angle fundus image dataset by using an initial feature extraction model with initial weight, so that the pre-trained feature extraction model with target weight is obtained, and the pre-trained feature extraction model can be efficiently extracted to fundus features, so that a subsequent module can decode the features conveniently.
It should be noted that, as shown in fig. 2 described above, when training is performed on the current ultra-wide-angle fundus image dataset using the initial feature extraction model with initial weights, the current ultra-wide-angle fundus image dataset is acquired, and the 6-classification training set and the test set are constructed while performing the data balancing operation such that the data amount of each type is close to 1:1.
Step S13: and decoding the target coding information through a bilinear decoder model to obtain a probability value information set output by the bilinear decoder model.
In this embodiment, feature extraction is performed on the preprocessed image by using a pre-trained feature extraction model to obtain target coding information, and the target coding information is decoded by using a bilinear decoder model to obtain a probability value information set output by the bilinear decoder model. Specifically, the target coding information is input to a first channel and a second channel of the bilinear decoder model; acquiring first output information output by the first channel and second output information output by the second channel; carrying out space dimension feature stacking on the first output information and the second output information to obtain a stacked feature layer; and executing preset prediction operation through the stacked feature layers to output the probability value information set.
Step S14: and determining the category with the maximum probability value in the probability value information set as the final prediction category corresponding to the original ultra-wide angle fundus image.
In this embodiment, after the target encoding information is decoded by the bilinear decoder model to obtain a probability value information set output by the bilinear decoder model, a category with the largest probability value in the probability value information set is determined as a final prediction category corresponding to the original ultra-wide angle fundus image. It can be understood that each image predicts 6 probability values corresponding to the probabilities of the 6 categories of vitreous opacity, maculopathy, diabetic retinopathy, glaucoma and other fundus diseases, and takes the category with the highest probability value as the final output category of the model. The number of the probability values predicted by each image and the categories can be set according to different conditions.
The invention has better experimental performance of real clinical data, the 6 classification model obtains 86.2% of Acc,86.1% of specificity, 97.3% of sensitivity and 86% of f1 value on the ten thousands of European Baurgh ultra-wide angle fundus image data, even if the ophthalmologist with abundant clinical experience is present, the ultra-wide angle fundus image can be identified by combining the information such as age, sex, medical history and the like of the patient, the disease category can be determined according to the identification result, and the condition that the identification result of the ultra-wide angle fundus image of a plurality of experts is inconsistent is difficult to be accurately identified by directly manually reading the film. At present, a multi-eye disease auxiliary AI system based on ultra-wide angle fundus images does not exist, most of the prior art can only execute two classification tasks, such as glaucoma/non-glaucoma, and the reality is that a patient is likely to be suffering from maculopathy, and the multi-eye disease auxiliary AI system cannot be detected in such a two-classification model, so that most of the prior art cannot be really used for clinically assisting doctor ultra-wide angle fundus image recognition. The model not only obtains higher test precision on the current ultra-wide angle fundus image data, but also obtains good effect on the fundus color Doppler ultrasound-based sugar net classification task, and proves that the model can be further fine-tuned and expanded to richer image recognition and classification scenes.
The invention performs data preprocessing on an original ultra-wide angle fundus image, and mainly aims to extract a fundus-only region from an original image containing partial eyelid, eyelash and background information. Secondly, the ResNet50 deep learning model is adopted to code image features, finer pixel-level features in the eye bottom image are extracted, and the image features are coded in a vector mode. And finally, decoding the encoded features, carrying out multiple complex algebraic operations on the feature vectors by constructing a bilinear decoder model, finally outputting 6 probability values, wherein the 6 probability values respectively correspond to the probabilities of 5 different eye diseases and normal categories, and taking the category with the maximum probability value as the output category, namely the ultra-wide angle fundus image recognition result. The multi-fundus-disease auxiliary image recognition system based on the ultra-wide-angle fundus image can automatically recognize various fundus diseases and has higher recognition accuracy, can be applied to clinical early fundus disease screening, provides reliable ultra-wide-angle fundus image recognition references for professional ophthalmologists, relieves the pressure of current first-line clinicians, can also reduce the false recognition probability caused by manual film reading, and improves film reading efficiency and ultra-wide-angle fundus image recognition accuracy.
It can be seen that the present application provides a method for identifying an ultra-wide angle fundus image, comprising: performing image preprocessing operation on the original ultra-wide angle fundus image to obtain a preprocessed image; extracting features of the preprocessed image by utilizing a pre-trained feature extraction model to obtain target coding information; decoding the target coding information through a bilinear decoder model to obtain a probability value information set output by the bilinear decoder model; and determining the category with the maximum probability value in the probability value information set as the final prediction category corresponding to the original ultra-wide angle fundus image. Therefore, the feature extraction model is used for extracting the features of the preprocessed image to obtain the feature vector, the bilinear decoder model is used for carrying out multiple complex algebraic operations on the feature vector, a plurality of probability values are output, a final recognition result is determined by comparing the sizes of the probability values, the probability of false recognition caused by manual film reading is reduced, and the precision of ultra-wide angle fundus image recognition is improved.
Referring to fig. 4, an embodiment of the present application discloses a method for identifying an ultra-wide angle fundus image, and compared with the previous embodiment, the present embodiment further describes and optimizes a technical solution.
Step S21: and performing image preprocessing operation on the original ultra-wide angle fundus image to obtain a preprocessed image.
Step S22: and extracting the characteristics of the preprocessed image by utilizing a pre-trained characteristic extraction model so as to obtain target coding information.
Step S23: and inputting the target coding information into a first channel and a second channel of the bilinear decoder model.
In this embodiment, the target encoding information is input to the first channel and the second channel of the bilinear decoder model. Specifically, inputting the target coding information into the first channel of the bilinear decoder model, and performing co-scale convolution according to a convolution kernel in the first channel to obtain convolved information; calculating a first attention weight of a current feature layer based on the convolved information and the target coding information; if the current feature layer is the last layer, calculating the first output information output by the first channel based on the target coding information and the first attention weight of the last layer; wherein the first output information is a weighted attention feature; inputting the target coding information into the second channel of the bilinear decoder model, and calculating an average value of the target coding information to obtain a second attention weight; the second output information of the second channel output is calculated based on the second attention weight and the target encoding information.
It will be appreciated that the bilinear feature decoding prediction module employs 2 parallel channel attention branches as shown in fig. 5: the AttentionNet1 is based on a convolution branch and the AttentionNet2 is based on a pooling branch, different attention generation paradigms and initial weights are adopted, important features are searched to the greatest extent, different channel weights are given to the features with different importance degrees, and the model is enabled to have important training and learning. And finally, fusing the weighted features in the space dimension, ensuring the integrity of the features, carrying out a series of subsequent convolution, pooling and full-connection layer processing to obtain a final class score vector, giving a corresponding score for each class, outputting the score in the form of a probability value through softmax operation to obtain the probability of each class, and taking the class with the largest probability value as a final prediction result.
The AttenionNet 1 element (i.e., the first channel) implements a channel attention mechanism based on a convolution approach. The features extracted by the previous module ResNet50 are subjected to same-scale convolution by adopting convolution with the size of 7 multiplied by 7, a floating point number with the size of 1 multiplied by 1 is output for each feature layer, the floating point number is determined to be the attention weight of the current feature layer, the (B, N, 7) output by the original ResNet50 is output as (B, N, 1) after the same-scale convolution, B represents the Batch size, N represents the feature channel number, W and H represent the image width and height respectively, and the attention weight of each feature layer is multiplied with the original output feature in the next stage to obtain the weighted attention feature.
I∈(B，N，W，H)；
K 1 ＝W×H；
Attention 1 ＝I*K 1 ∈(B，N，1，1)；
Out 1 ＝I×Attention 1 ∈(B，N，W，H)；
I is target coding information, attention 1 Out is the first attention weight 1 Is the first output information.
The AttenionNet 2 element (i.e., the second channel) implements a channel attention mechanism based on a global averaging pooling approach. Taking the 7 x 7 feature matrix as the attention weight of the feature, since most detail features in the important feature layer will have higher pixel values, since the weight value obtained after averaging is also larger, it can be considered that a greater degree of attention will be obtained in the subsequent training process, and similarly, the average value is smaller, and the influence of the background and other features will be ignored gradually in the subsequent training process.
I∈(B，N，W，H)；Attention2＝GlobalAvgPool(I)∈(B，N，1，1)；
Out 2 ＝I×Attention 2 ∈(B，N，W，H)；
I is target coding information, attention 2 Out is the second attention weight 2 Is the second output information.
Step S24: and acquiring first output information output by the first channel and second output information output by the second channel, and stacking the first output information and the second output information by space dimension characteristics to obtain a stacked characteristic layer.
In this embodiment, after the target encoding information is input to a first channel and a second channel of the bilinear decoder model, first output information output by the first channel and second output information output by the second channel are obtained, and spatial dimension feature stacking is performed on the first output information and the second output information, so as to obtain a feature layer after stacking. It will be appreciated that the FeatureFuse unit stacks the Out using spatial dimension characteristics 1 And Out 2 Stacked together, will become 2N feature layers.
Step S25: and executing preset prediction operation through the stacked feature layers to output the probability value information set.
In this embodiment, after the first output information and the second output information are stacked to obtain a stacked feature layer, a preset prediction operation is performed through the stacked feature layer, so as to output the probability value information set. It will be appreciated that Out 1 And Out 2 Will become 2N feature layers, and the final class probability is obtained after a subsequent series of convolutions, pooling, full join layers is performed with all weighted feature layers retained.
F＝Concatenate(Out 1 ，Out 2 )∈(B，2N，W，H)；
Score＝FC(AvgPool(ConvBlock(F)))∈(B，6，1，1)；
Out＝SoftMax(Score)∈(B，6，1，1)。
And finally, out represents the probability of the 6 categories of the B input images, wherein each image predicts 6 probability values, the 6 probability values respectively correspond to vitreous opacity, maculopathy, diabetic retinopathy, glaucoma and other fundus diseases, and the category with the largest probability value is taken as the final output category of the model.
Step S26: and determining the category with the maximum probability value in the probability value information set as the final prediction category corresponding to the original ultra-wide angle fundus image.
For the specific content of the steps S21, S22, S26, reference may be made to the corresponding content disclosed in the foregoing embodiment, and no further description is given here.
Therefore, the embodiment of the application obtains the preprocessed image by performing image preprocessing operation on the original ultra-wide angle fundus image; extracting features of the preprocessed image by utilizing a pre-trained feature extraction model to obtain target coding information; inputting the target coding information into a first channel and a second channel of the bilinear decoder model; acquiring first output information output by the first channel and second output information output by the second channel; carrying out space dimension feature stacking on the first output information and the second output information to obtain a stacked feature layer; executing preset prediction operation through the stacked feature layers to output the probability value information set; and determining the category with the maximum probability value in the probability value information set as the final prediction category corresponding to the original ultra-wide-angle fundus image, so that the probability of false recognition caused by manual film reading is reduced, and the accuracy of ultra-wide-angle fundus image recognition is improved.
Referring to fig. 6, the embodiment of the application further correspondingly discloses a super-wide-angle fundus image identification device, which comprises:
An image preprocessing module 11, configured to perform an image preprocessing operation on an original ultra-wide angle fundus image to obtain a preprocessed image;
a feature extraction module 12, configured to perform feature extraction on the preprocessed image by using a feature extraction model that is pre-trained, so as to obtain target coding information;
a decoding module 13, configured to decode the target encoded information through a bilinear decoder model, so as to obtain a probability value information set output by the bilinear decoder model;
and the prediction category determining module 14 is configured to determine a category with the largest probability value in the probability value information set as a final prediction category corresponding to the original ultra-wide angle fundus image.
It can be seen that the present application includes: performing image preprocessing operation on the original ultra-wide angle fundus image to obtain a preprocessed image; extracting features of the preprocessed image by utilizing a pre-trained feature extraction model to obtain target coding information; decoding the target coding information through a bilinear decoder model to obtain a probability value information set output by the bilinear decoder model; and determining the category with the maximum probability value in the probability value information set as the final prediction category corresponding to the original ultra-wide angle fundus image. Therefore, the feature extraction model is used for extracting the features of the preprocessed image to obtain the feature vector, the bilinear decoder model is used for carrying out multiple complex algebraic operations on the feature vector, a plurality of probability values are output, a final recognition result is determined by comparing the sizes of the probability values, the probability of false recognition caused by manual film reading is reduced, and the precision of ultra-wide angle fundus image recognition is improved.
In some embodiments, the image preprocessing module 11 specifically includes:
a target region extraction unit for extracting a target region from the original ultra-wide angle fundus image based on an adaptive ROI region rough extraction component; the target area is an area formed by pixel points meeting preset pixel conditions;
the target area quality judging unit is used for judging whether the target area meets the preset area quality requirement or not;
the current brightness acquisition unit is used for acquiring the current brightness of the target area if the target area meets the preset area quality requirement;
the brightness interval determining unit is used for determining a current brightness interval corresponding to the current brightness;
the current brightness adjustment strategy determining unit is used for determining a current brightness adjustment strategy corresponding to the current brightness interval based on a preset brightness adjustment strategy determining rule;
the brightness adjusting unit is used for adjusting the brightness of the target area by utilizing the current brightness adjusting strategy so as to obtain the preprocessed image;
the long axis and short axis determining unit is used for calculating corresponding long axis and short axis length information according to the length and width of the original ultra-wide angle fundus image if the target area does not meet the quality requirement of the preset area or the target area is not extracted;
A center point determining unit for determining center point position information based on the reference point information;
an elliptical region generating unit configured to generate an elliptical region based on the center point position information, the long axis length information, and the short axis length information, and determine the elliptical region as the fitting region;
and a target area determining unit, configured to determine the fitting area as the target area, and reenter the step of adjusting the brightness of the target area based on a preset brightness adjustment rule, so as to obtain the preprocessed image.
In some embodiments, the feature extraction module 12 specifically includes:
the initial feature extraction model acquisition unit is used for training the initial feature extraction model by using the open source data set so as to obtain an initial feature extraction model with initial weight;
a pre-trained feature extraction model acquisition unit for training the initial feature extraction model using a super-wide angle fundus image dataset to obtain the pre-trained feature extraction model with target weights;
the feature information acquisition unit is used for extracting feature information meeting the preset feature requirement from the preprocessed image by utilizing the pre-trained feature extraction model with the target weight;
And the encoding unit is used for encoding the characteristic information in a vector form to obtain the target encoding information.
In some embodiments, the decoding module 13 specifically includes:
the convolution unit is used for inputting the target coding information into the first channel of the bilinear decoder model, and carrying out co-scale convolution according to a convolution kernel in the first channel to obtain convolved information;
a first attention weight calculation unit for calculating a first attention weight of a current feature layer based on the convolved information and the target coding information;
a first output information generating unit, configured to calculate, if the current feature layer is a last layer, the first output information output by the first channel based on the target coding information and the first attention weight of a previous layer; wherein the first output information is a weighted attention feature;
a second attention weight calculation unit for inputting the target encoded information to the second channel of the bilinear decoder model, calculating an average value of the target encoded information to obtain a second attention weight;
a second output information generating unit that calculates the second output information of the second channel output based on the second attention weight and the target encoding information;
An output information obtaining unit, configured to obtain first output information output by the first channel and second output information output by the second channel;
the output information stacking unit is used for performing space dimension feature stacking on the first output information and the second output information to obtain a stacked feature layer;
and the probability value information set output unit is used for executing preset prediction operation through the stacked feature layers so as to output the probability value information set.
In some embodiments, the prediction category determination module 14 specifically includes:
and the prediction category determining unit is used for determining the category with the largest probability value in the probability value information set as the final prediction category corresponding to the original ultra-wide angle fundus image.
Further, the embodiment of the application also provides electronic equipment. Fig. 7 is a block diagram of an electronic device 20, according to an exemplary embodiment, and is not intended to limit the scope of use of the present application in any way.
Fig. 7 is a schematic structural diagram of an electronic device 20 according to an embodiment of the present application. The electronic device 20 may specifically include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input output interface 25, and a communication bus 26. Wherein the memory 22 is used for storing a computer program, and the computer program is loaded and executed by the processor 21 to implement the relevant steps in the ultra-wide angle fundus image identification method disclosed in any of the foregoing embodiments. In addition, the electronic device 20 in the present embodiment may be specifically an electronic computer.
In this embodiment, the power supply 23 is configured to provide an operating voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and an external device, and the communication protocol to be followed is any communication protocol applicable to the technical solution of the present application, which is not specifically limited herein; the input/output interface 25 is used for acquiring external input data or outputting external output data, and the specific interface type thereof may be selected according to the specific application requirement, which is not limited herein.
The memory 22 may be a carrier for storing resources, such as a read-only memory, a random access memory, a magnetic disk, or an optical disk, and the resources stored thereon may include an operating system 221, a computer program 222, and the like, and the storage may be temporary storage or permanent storage.
The operating system 221 is used for managing and controlling various hardware devices on the electronic device 20 and computer programs 222, which may be Windows Server, netware, unix, linux, etc. The computer program 222 may further include a computer program capable of performing other specific works in addition to the computer program capable of performing the ultra-wide-angle fundus image recognition method performed by the electronic apparatus 20 disclosed in any of the foregoing embodiments.
Further, the embodiment of the application also discloses a storage medium, wherein the storage medium stores a computer program, and the computer program realizes the steps of the super-wide-angle fundus image identification method disclosed in any embodiment when being loaded and executed by a processor.
In this specification, each embodiment is described in a progressive manner, and each embodiment is mainly described in a different point from other embodiments, so that the same or similar parts between the embodiments are referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant points refer to the description of the method section.
Finally, it is further noted that relational terms such as first and second, and the like are used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element.
The above description of the method, the device, the equipment and the storage medium for identifying the ultra-wide-angle fundus image provided by the invention applies specific examples to illustrate the principle and the implementation of the invention, and the description of the above examples is only used for helping to understand the method and the core idea of the invention; meanwhile, as those skilled in the art will have variations in the specific embodiments and application scope in accordance with the ideas of the present invention, the present description should not be construed as limiting the present invention in view of the above.
Claims (10)
1. A method for identifying an ultra-wide angle fundus image, comprising:
performing image preprocessing operation on the original ultra-wide angle fundus image to obtain a preprocessed image;
extracting features of the preprocessed image by utilizing a pre-trained feature extraction model to obtain target coding information;
decoding the target coding information through a bilinear decoder model to obtain a probability value information set output by the bilinear decoder model;
and determining the category with the maximum probability value in the probability value information set as the final prediction category corresponding to the original ultra-wide angle fundus image.
2. The method of claim 1, wherein performing an image preprocessing operation on the original ultra-wide angle fundus image to obtain a preprocessed image comprises:
extracting a target region from the original ultra-wide angle fundus image based on an adaptive ROI region rough extraction component; the target area is an area formed by pixel points meeting preset pixel conditions;
judging whether the target area meets the quality requirement of a preset area or not;
if the target area meets the preset area quality requirement, adjusting the brightness of the target area based on a preset brightness adjustment rule to obtain the preprocessed image;
if the target area does not meet the preset area quality requirement or the target area is not extracted, performing maximum ellipse approximate fitting to obtain a fitting area;
and determining the fitting area as the target area, and re-entering the step of adjusting the brightness of the target area based on a preset brightness adjustment rule to obtain the preprocessed image.
3. The method according to claim 2, wherein the adjusting the brightness of the target area based on a preset brightness adjustment rule to obtain the preprocessed image includes:
Acquiring the current brightness of the target area, and determining a current brightness interval corresponding to the current brightness;
determining a current brightness adjustment strategy corresponding to the current brightness interval based on a preset brightness adjustment strategy determination rule;
and adjusting the brightness of the target area by utilizing the current brightness adjustment strategy so as to obtain the preprocessed image.
4. The method of claim 2, wherein performing a maximum ellipse approximation fit to obtain a fit region comprises:
calculating long axis length information and short axis length information corresponding to the length and the width of the original ultra-wide angle fundus image;
determining center point position information based on the reference point information;
an elliptical region is generated based on the center point position information, the long axis length information, and the short axis length information, and the elliptical region is determined as the fitting region.
5. The method according to claim 1, wherein before the feature extraction is performed on the pre-processed image by using a pre-trained feature extraction model to obtain the target encoding information, further comprising:
training an original feature extraction model by using an open source data set to obtain an initial feature extraction model with initial weight;
Training the initial feature extraction model by using a super-wide-angle fundus image dataset to obtain the pre-trained feature extraction model with target weights;
correspondingly, the feature extraction of the preprocessed image by using the pre-trained feature extraction model to obtain target coding information comprises the following steps:
extracting feature information meeting preset feature requirements from the preprocessed image by using the pre-trained feature extraction model with the target weight;
and encoding the characteristic information in a vector form to obtain the target encoding information.
6. The method according to claim 1, wherein decoding the target encoded information by a bilinear decoder model to obtain a set of probability value information output by the bilinear decoder model, comprises:
inputting the target coding information into a first channel and a second channel of the bilinear decoder model;
acquiring first output information output by the first channel and second output information output by the second channel;
carrying out space dimension feature stacking on the first output information and the second output information to obtain a stacked feature layer;
And executing preset prediction operation through the stacked feature layers to output the probability value information set.
7. The method of claim 6, wherein the inputting the target encoding information into the first and second channels of the bilinear decoder model comprises:
inputting the target coding information into the first channel of the bilinear decoder model, and performing co-scale convolution according to a convolution kernel in the first channel to obtain convolved information;
calculating a first attention weight of a current feature layer based on the convolved information and the target coding information;
if the current feature layer is the last layer, calculating the first output information output by the first channel based on the target coding information and the first attention weight of the last layer; wherein the first output information is a weighted attention feature;
inputting the target coding information into the second channel of the bilinear decoder model, and calculating an average value of the target coding information to obtain a second attention weight;
the second output information of the second channel output is calculated based on the second attention weight and the target encoding information.
8. An ultra-wide angle fundus image recognition apparatus, comprising:
the image preprocessing module is used for performing image preprocessing operation on the original ultra-wide angle fundus image so as to obtain a preprocessed image;
the feature extraction module is used for carrying out feature extraction on the preprocessed image by utilizing a pre-trained feature extraction model so as to obtain target coding information;
the decoding module is used for decoding the target coding information through the bilinear decoder model to obtain a probability value information set output by the bilinear decoder model;
and the prediction category determining module is used for determining the category with the largest probability value in the probability value information set as the final prediction category corresponding to the original ultra-wide angle fundus image.
9. An electronic device, comprising:
a memory for storing a computer program;
a processor for executing the computer program to implement the steps of the ultra-wide angle fundus image identification method as recited in any one of claims 1 to 7.
10. A computer-readable storage medium storing a computer program; wherein the computer program when executed by a processor implements the ultra-wide angle fundus image identification method as claimed in any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310761433.8A CN116824116A (en) | 2023-06-26 | 2023-06-26 | Super wide angle fundus image identification method, device, equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310761433.8A CN116824116A (en) | 2023-06-26 | 2023-06-26 | Super wide angle fundus image identification method, device, equipment and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116824116A true CN116824116A (en) | 2023-09-29 |
Family
ID=88125182
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310761433.8A Pending CN116824116A (en) | 2023-06-26 | 2023-06-26 | Super wide angle fundus image identification method, device, equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116824116A (en) |
Citations (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104021571A (en) * | 2014-06-03 | 2014-09-03 | 浙江宇视科技有限公司 | Moving target automatic tracking method and device for improving user experience |
| CN108185984A (en) * | 2017-12-28 | 2018-06-22 | 中山大学 | The method that eyeground color picture carries out eyeground lesion identification |
| CN111629146A (en) * | 2020-05-27 | 2020-09-04 | 腾讯科技（深圳）有限公司 | Shooting parameter adjusting method, shooting parameter adjusting device, shooting parameter adjusting equipment and storage medium |
| CN112587129A (en) * | 2020-12-01 | 2021-04-02 | 上海影谱科技有限公司 | Human body action recognition method and device |
| WO2021135254A1 (en) * | 2019-12-31 | 2021-07-08 | 深圳云天励飞技术股份有限公司 | License plate number recognition method and apparatus, electronic device, and storage medium |
| WO2021169128A1 (en) * | 2020-02-29 | 2021-09-02 | 平安科技（深圳）有限公司 | Method and apparatus for recognizing and quantifying fundus retina vessel, and device and storage medium |
| CN113420671A (en) * | 2021-06-24 | 2021-09-21 | 杭州电子科技大学 | Saliency target detection method based on global information attention |
| CN113837067A (en) * | 2021-09-18 | 2021-12-24 | 成都数字天空科技有限公司 | Organ contour detection method and device, electronic equipment and readable storage medium |
| CN114549538A (en) * | 2022-02-24 | 2022-05-27 | 杭州电子科技大学 | Brain tumor medical image segmentation method based on spatial information and characteristic channel |
| US20220319036A1 (en) * | 2020-10-28 | 2022-10-06 | Beijing Zhenhealth Technology Co., Ltd. | Method and device for recognizing fundus image, and equipment |
| CN115797246A (en) * | 2021-09-09 | 2023-03-14 | 上海微创卜算子医疗科技有限公司 | Pathological image quality evaluation and adjustment method and system, electronic device and medium |
| CN116091900A (en) * | 2021-10-29 | 2023-05-09 | 华为技术有限公司 | Image processing method, device and storage medium |
| CN116258850A (en) * | 2022-12-12 | 2023-06-13 | 浙江大华技术股份有限公司 | Image semantic segmentation method, electronic device and computer readable storage medium |
-
2023
- 2023-06-26 CN CN202310761433.8A patent/CN116824116A/en active Pending
Patent Citations (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104021571A (en) * | 2014-06-03 | 2014-09-03 | 浙江宇视科技有限公司 | Moving target automatic tracking method and device for improving user experience |
| CN108185984A (en) * | 2017-12-28 | 2018-06-22 | 中山大学 | The method that eyeground color picture carries out eyeground lesion identification |
| WO2021135254A1 (en) * | 2019-12-31 | 2021-07-08 | 深圳云天励飞技术股份有限公司 | License plate number recognition method and apparatus, electronic device, and storage medium |
| WO2021169128A1 (en) * | 2020-02-29 | 2021-09-02 | 平安科技（深圳）有限公司 | Method and apparatus for recognizing and quantifying fundus retina vessel, and device and storage medium |
| CN111629146A (en) * | 2020-05-27 | 2020-09-04 | 腾讯科技（深圳）有限公司 | Shooting parameter adjusting method, shooting parameter adjusting device, shooting parameter adjusting equipment and storage medium |
| US20220319036A1 (en) * | 2020-10-28 | 2022-10-06 | Beijing Zhenhealth Technology Co., Ltd. | Method and device for recognizing fundus image, and equipment |
| CN112587129A (en) * | 2020-12-01 | 2021-04-02 | 上海影谱科技有限公司 | Human body action recognition method and device |
| CN113420671A (en) * | 2021-06-24 | 2021-09-21 | 杭州电子科技大学 | Saliency target detection method based on global information attention |
| CN115797246A (en) * | 2021-09-09 | 2023-03-14 | 上海微创卜算子医疗科技有限公司 | Pathological image quality evaluation and adjustment method and system, electronic device and medium |
| CN113837067A (en) * | 2021-09-18 | 2021-12-24 | 成都数字天空科技有限公司 | Organ contour detection method and device, electronic equipment and readable storage medium |
| CN116091900A (en) * | 2021-10-29 | 2023-05-09 | 华为技术有限公司 | Image processing method, device and storage medium |
| CN114549538A (en) * | 2022-02-24 | 2022-05-27 | 杭州电子科技大学 | Brain tumor medical image segmentation method based on spatial information and characteristic channel |
| CN116258850A (en) * | 2022-12-12 | 2023-06-13 | 浙江大华技术股份有限公司 | Image semantic segmentation method, electronic device and computer readable storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109584254B (en) | Heart left ventricle segmentation method based on deep full convolution neural network | |
| CN109376636B (en) | Capsule network-based eye fundus retina image classification method | |
| CN108021916B (en) | Deep learning diabetic retinopathy sorting technique based on attention mechanism | |
| WO2020215672A1 (en) | Method, apparatus, and device for detecting and locating lesion in medical image, and storage medium | |
| CN109919928B (en) | Medical image detection method and device and storage medium | |
| CN112132817B (en) | Retina blood vessel segmentation method for fundus image based on mixed attention mechanism | |
| WO2022088665A1 (en) | Lesion segmentation method and apparatus, and storage medium | |
| CN111080643A (en) | Method and device for classifying diabetes and related diseases based on fundus images | |
| CN113177916B (en) | Slight hypertension fundus identification model based on few-sample learning method | |
| US11093800B2 (en) | Method and device for identifying object and computer readable storage medium | |
| CN110503155A (en) | A kind of method and relevant apparatus, server of information classification | |
| CN112396605B (en) | Network training method and device, image recognition method and electronic equipment | |
| Xie et al. | Optic disc and cup image segmentation utilizing contour-based transformation and sequence labeling networks | |
| CN114066884A (en) | Retinal blood vessel segmentation method and device, electronic device and storage medium | |
| CN112085745A (en) | Retinal vessel image segmentation method of multi-channel U-shaped full convolution neural network based on balanced sampling splicing | |
| CN117058676B (en) | Blood vessel segmentation method, device and system based on fundus examination image | |
| Zhang et al. | Attention-based multi-model ensemble for automatic cataract detection in B-scan eye ultrasound images | |
| CN113850753A (en) | Medical image information calculation method and device, edge calculation equipment and storage medium | |
| CN112489043A (en) | Heart disease detection device, model training method, and storage medium | |
| CN113724203B (en) | Model training method and device applied to target feature segmentation in OCT image | |
| CN113762117A (en) | Training method of image processing model, image processing model and computer equipment | |
| CN111553250B (en) | Accurate facial paralysis degree evaluation method and device based on face characteristic points | |
| CN115590481B (en) | Apparatus and computer-readable storage medium for predicting cognitive impairment | |
| CN114864093B (en) | Apparatus, method and storage medium for disease prediction based on fundus image | |
| CN116824116A (en) | Super wide angle fundus image identification method, device, equipment and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |