49 results about "Retina eye" patented technology

The invention provides a retinal fundus vessel segmentation method based on a deep multi-scale attention convolutional neural network. An internationally disclosed retinal fundus vessel data set DRIVEis adopted to perform validity verification: firstly, dividing the retinal fundus vessel data set DRIVE into a training set and a test set, and adjusting the picture size to 512*512 pixels; then, enabling the training set to be subjected to four random preprocessing links to achieve a data enhancement effect; designing a model structure of the deep multi-scale attention convolutional neural network, and inputting the processed training set into the model for training; and finally, inputting the test set into the trained network, and testing the model performance. The main innovation point ofthe method is that a double attention module is designed, so that the whole model pays more attention to segmentation of small blood vessels; and a multi-scale feature fusion module is designed, so that the global feature extraction capability of the whole model on the segmented image is stronger. The segmentation accuracy of the model on a DRIVE data set is 96.87%, the sensitivity is 79.45%, thespecificity is 98.57, and the method is superior to classical UNet and an existing most advanced segmentation method.

The invention discloses a retinal fundus image cup / disc ratio automatic evaluation method. The method comprises the following steps: A, extracting an optic disk area image from a retinal fundus image;B, establishing and training a optic disc optic cup segmentation network based on the deep convolutional neural network; C, acquiring a to-be-detected optic disc area image from the to-be-detected retina fundus image according to the step A, and inputting the to-be-detected optic disc area image into an optic disc optic cup segmentation network to output a to-be-detected optic disc segmentation mask image and a to-be-detected optic cup segmentation mask image; And step D, calculating the cup-disc ratio of the retinal fundus image according to the to-be-detected optic disc segmentation mask image and the to-be-detected optic cup segmentation mask image. The method is high in operation speed, good in effect, free of manual participation, low in cost and high in universality, and can be widely applied to auxiliary screening of glaucoma.

The present disclosure provides an artificial neural network system for identifying a lesion in a retinal fundus image that comprises a pre-processing module configured to separately pre-process a target retinal fundus image and a reference retinal fundus image taken from a same person; a first neural network (12) configured to generate a first advanced feature set from the target retinal fundus image; a second neural network (22) configured to generate a second advanced feature set from the reference retinal fundus image; a feature combination module (13) configured to combine the first advanced feature set and the second advanced feature set to form a feature combination set; and a third neural network (14) configured to generate, according to the feature combination set, a diagnosis result. By using a target retinal fundus image and a reference retinal fundus image as independent input information, the artificial neural network may simulate a doctor, determining lesions on the target retinal fundus image using other retinal fundus images from the same person as a reference, thereby enhancing the diagnosis accuracy.

The invention discloses a diabetic retina eye ground image pathology detection method comprising the following steps: preprocessing a medical eye ground image so as to form a standard eye ground image; dividing areas for the eye ground image so as to form m sub-images of different eye ground areas; using a multilayer area convolution nerve network model to extract an area depth feature vector of the eye ground sub-image; using the area depth feature vector as the LSTM nerve network input, predicting correlations of different areas, and forming an eye ground image global feature vector; finally, using a whole articulamentum and softmax to realize multi-classification detection of the eye ground image. The method is based on retina eye ground images and labels that can be crawled on the internet, uses the correlation between the eye ground image area depth features and the adjacent area, and realizes the diabetic retina eye ground image pathology automatic detection via the convolution nerve network and recursion nerve network algorithms, thus effectively improving the detection accuracy and timeliness.

The invention discloses a retinal fundus image bleeding detection method. A fundus image photographed by a color digital non-mydriatic fundus camera is utilized. The method adopts the following steps that (1) the size of an original image is adjusted; (2) the viewing field of the adjusted image is positioned; (3) bleeding and blood vessel areas based on double-scale background are crudely detected; (4) the blood vessels are detected; (5) a bleeding suspicious area is iteratively positioned; (6) feature extraction is performed on the bleeding suspicious area; and (7) classified detection is performed on the bleeding suspicious area, and the bleeding marked fundus image is generated. According to the method, the fundus image acquired under different acquisition conditions can be processed, the bleeding area can be rapidly and effectively detected and the bleeding site of the processed fundus image is visual and obvious so that further diagnosis is facilitated for ophthalmologists.

The invention discloses an eye fundus image retinal vessel segmentation method based on a mixed attention mechanism, and the method comprises the following steps: S1, obtaining a retinal eye fundus image, and dividing the retinal image into a training set and a test set; S2, constructing a hybrid attention convolutional neural network, wherein the hybrid attention convolutional neural network is used for segmenting retinal vessels in the retinal fundus image; S3, training the hybrid attention convolutional neural network by using a training set, and testing the hybrid attention convolutional neural network by using a test set to obtain a trained hybrid attention convolutional neural network; and S4, inputting a to-be-segmented retinal image into the trained mixed attention convolutional neural network, wherein the mixed attention convolutional neural network outputs a retinal image blood vessel segmentation result. According to the method, a low-contrast vascular structure is effectively and accurately segmented, and the method has high robustness for interference of complex eye fundus image focuses, blood vessel center reflection phenomena and illumination imbalance phenomena.

The invention provides a novel retina eye fundus image segmenting method. The method is characterized in that the best entropy threshold value is calculated by combining multi-scale linear detection and using the gray-level-gradient co-occurrence matrix of an image. Firstly, green components, containing rich blood vessel outline information, in the retina eye fundus image are extracted, and shadow correcting, noise reducing, CLAHE and other preprocessing are performed on the green components; secondly, multi-scale and multi-direction linear detection is performed on blood vessels of the retina eye fundus image according to morphological structure characteristics of the blood vessels, and image responses of different scales are fused to obtain the characteristics of the blood vessels; finally the best entropy threshold value of the image is calculated on the basis of the gray-level-gradient co-occurrence matrix of the image, and segmentation is performed. The method is high in segmenting accuracy, capable of extracting more fine blood vessels, high in calculating speed, very good in robustness and suitable for segmentation of the normal or lesion retina eye fundus image.

The invention provides a retinal fundus image segmentation method and device, which can realize automatic, efficient and accurate segmentation of retinal fundus image blood vessels and optic disks under the condition of only a small number of labeled samples. The method comprises the steps of acquiring a training set image and a corresponding annotation image, wherein the image is a retinal fundusimage; carrying out consistent random cutting and blocking on the obtained training set image and the corresponding annotation image; constructing a U-shaped full convolutional neural network; inputting the randomly clipped and blocked training set images and the corresponding annotation images into a constructed U-shaped full convolutional neural network, and training the U-shaped full convolutional neural network; and segmenting the retinal fundus image by using the trained U-shaped full convolutional neural network. The invention relates to the field of artificial intelligence and diabeticretina diagnosis.

The invention discloses a combined retinal vessel segmentation system and method based on the fractal dimension number and Gauss filtering. The system comprises an image pre-processing unit used for improving retina image contrast and improving image quality to reduce interference of imaging problems on the subsequent processing process, a local detection unit used for extracting detailed portions of the retina image after pre-processing to improve fine vessel extraction accuracy, a global detection unit used for extracting a main portion of the retina image after pre-processing to highlight the global information of the retina vessel, a retina image fusion unit used for fusing the detailed portions of the retina image extracted by the local detection unit and the main portion of the retina image extracted by the global detection unit and improving system segmentation accuracy, and a retina image post-processing unit. The system is advantaged in that the fractal dimension number and Gauss filtering are combined, the system is suitable for segmentation of vessels of retina eyeground images, and segmentation accuracy is substantially improved.

A method for detecting disc haemorrhages in a retinal fundus image. The method includes (a) identifying a ring-shaped region of interest in the retinal fundus image encompassing the optic disc boundary; (b) removing blood vessel regions in the identified region of interest; (c) detecting candidate disc haemorrhages from the removed blood vessels regions in the identified region of interest; and (d) screening the candidate disc haemorrhages. The detected disc haemorrhages may be used to aid in the detection of glaucoma.

The invention discloses an automatic detection method of diabetic retinopathy. According to the method, firstly, medical ocular fundus images are preprocessed to generate feature vectors correspondingto all the ocular fundus images; then a clustering algorithm is applied to carry out clustering on all the ocular fundus images on the basis of the feature vectors corresponding to the ocular fundusimages, and the same are defined as ocular fundus images of different patterns; then reference feature space is established on the basis of the feature vectors, which correspond to the ocular fundus images of different lesion periods and the different patterns, to obtain feature codes thereof in the reference feature space; and finally, automatic detection on the diabetic retinopathy is realized through calculating cosine similarity between feature codes of a to-be-detected ocular fundus image and the ocular fundus images with labels. According to the method, the image clustering algorithm iscombined to establish the reference feature space of the retinal ocular fundus images of the different lesion periods and the different patterns and feature code mapping thereof on the basis of the retinal ocular fundus images and the labels which can be crawled on a network, and accuracy and timeliness of automatic detection of the diabetic retinopathy are effectively improved.

The invention discloses a retinal vessel segmentation method and system based on a retinal fundus image, and belongs to the technical field of image processing, and the method comprises the steps: obtaining a to-be-detected retinal fundus image; constructing a basic module according to the retinal fundus image features; cascading N basic modules to serve as a final network model, the to-be-detected retinal fundus image serving as input of the whole network model, and obtaining a segmentation result of retinal vessels. The foreground characteristics of the previous basic module and the originalpicture are transmitted to the next basic module together, so that the rear basic module can inherit the learning experience of the front basic module, the training process is accelerated, and the problem of data imbalance is effectively solved; the to-be-detected retinal fundus image is used as the input of the overall model S-UNet, and the obtained segmentation result of the retinal blood vessel is more accurate.

The invention discloses a n eye fundus image optic disc and macular positioning detection algorithm based on YOLO-V3 in the field of medical image recognition. The eye fundus image optic disc and macular positioning detection algorithm comprises the following steps: a, collecting and manufacturing a retina eye fundus image data set with an optic disc and a macular label; b, after the fundus imageoptic disk and macular data set is manufactured, modifying network parameters according to the target category needing to be recognized, and training a model; and c, after the model training is completed, testing multiple groups of independent data sets to realize rapid fundus image optic disc and macular positioning detection and evaluate a model detection effect. The positions of the optic discand the macula luteae in the fundus image can be simultaneously positioned in one set of system, the identification precision and speed are improved compared with those of a traditional positioning method, and the complexity of a corresponding positioning algorithm is reduced at the same time.

A method for segmenting optic disc based on color retinal fundus image of lesion focus includes the following steps that the blood vessel detection model and optic disc detection model were established, the probabilistic map of blood vessel and the probabilistic map of optic disc were obtained from the detection model of blood vessel and the detection model of optic disc respectively; the probabilistic bubble map of the fitting straight line map of main blood vessel was obtained from the probabilistic map of blood vessel; and the optic disc region was selected from the optic disc communicationarea and the center and radius of optic disc were estimated. The method of the invention can effectively avoid interference of lesions, blood vessel occlusion, brightness change and the like in an image, thereby realizing accurate segmentation of an optic disc.

The invention discloses an eye fundus image blood vessel segmentation method of a semantic and multi-scale fusion network, and belongs to the field of medical image processing and computer vision. According to the method, a semantic fusion module and a multi-scale fusion module are designed, and the semantic fusion module and the multi-scale fusion module are utilized to construct a semantic and multi-scale fusion network with a unique structure for segmenting fundus vessels of retinal images. The semantic fusion module improves the segmentation precision of capillaries by better fusing high-dimensional semantic information, and the multi-scale fusion module solves the problem that the blood vessel scale change is large through fusion of multi-scale information. Experiments prove that themethod can effectively improve the retinal fundus vessel segmentation precision. In addition, the network provided by the invention is clear in structure, easy to construct and easy to implement.

The invention discloses a retinal fundus image classification method based on an improved CNN model. The method comprises the following steps: classifying and marking acquired training images; performing image preprocessing on the training picture; establishing an improved CNN model; training the improved CNN model by adopting the training pictures to obtain a picture classifier; and classifying the retinal fundus images to be detected by adopting an image classifier to obtain a final classification result. The improved CNN model and the improved CNN classification method based on multiple tasks are excellent in performance, higher in efficiency, less in occupied resources, high in reliability and good in accuracy.

The invention relates to a method for analyzing a retina fundus image and a related product thereof. The method comprises the steps that a pre-trained arteriovenous segmentation model is used for processing a retina fundus image to obtain an artery segmentation result and a vein segmentation result, the artery segmentation result comprises a continuous artery network, and the vein segmentation result comprises a continuous vein network; acquiring position information of a positioning reference object in the retina fundus image; and determining structure attribute information of key points in the artery network and the vein network according to the position information of the positioning reference object, the artery segmentation result and the vein segmentation result. According to the technical scheme of the invention, the continuous artery network and vein network about the retina fundus image are utilized, and the position information of the positioning reference object is combined, so that the refined detection of the multi-dimensional structure attribute of the key point can be realized.

A method for locating fovea based on color retinal fundus image of lesion focus includes detecting and locating optic disc and blood vessel, acquiring blood vessel vector and blood vessel vector set,estimating horizontal ridge line passing through central fossa from blood vessel vector set, acquiring search area according to optic disc and horizontal ridge line, and acquiring the central fossa insearch area. The method of the invention determines the horizontal ridge line of the fundus image by constructing a blood vessel vector set model and calculating the vector sum of the vectors, determines a search area by utilizing the position relationship between the optic disc and the central fossa, and finally determines the central fossa positioning according to the local brightness extreme value of the search area. The method can effectively avoid the interference of the focus and illumination on the optic disc, and accurately locate the fovea.

The invention discloses a multi-modal retinal fundus image registration method and device. The method comprises the steps: extracting a second feature point set in a first feature point set referenceimage in a floating image according to a scale invariant feature conversion algorithm; obtaining a first feature difference matrix according to the second shape context features of the first feature point set and the second feature point set, and obtaining a second feature difference matrix according to the second texture features of the first feature point set and the second feature point set; algorithm according to expectation maximization, solving and calculating the first characteristic difference matrix and the second characteristic difference matrix through a Gaussian mixture model and aBayesian law; after a posterior probability matrix of Bayesian rules of the first feature difference matrix and the second feature difference matrix based on a Gaussian mixture model is obtained, calculation is conducted according to the posterior probability matrix, and point set coordinates are obtained until the calculation result of the expectation maximization algorithm converges or reachesthe preset number of iterations; and obtaining a registered image according to the point set coordinates.

The invention relates to the field of computer vision image processing, and discloses a retinal blood vessel image segmentation method and device. The method comprises the following steps: in a data processing stage, firstly, acquiring a to-be-detected retinal blood vessel image; then, performing data enhancement and preprocessing on the retina fundus image; in the training stage, firstly, a network model based on Transform optimization is constructed, and then network model training is carried out by using a processed training image; in a test stage, inputting the retinal blood vessel image into a trained network model for image segmentation; and finally, weighted average is carried out on prediction results of a plurality of retinal blood vessel images output by the network model to obtain the classification probability of each pixel, a final segmentation result graph is obtained, and the segmentation precision of the retinal blood vessel images is improved by adopting the retinal blood vessel image segmentation method.

The invention discloses an ophthalmic imaging system with automatic retinal feature detection, which mainly comprises a user information module, a position measurement correction module, a retina scanning module, an imaging module, a data screening module, an information storage module and an abnormal information recording module. The system of the invention transmits patient case information through the Internet, protects patient privacy, and provides a communication platform for medical and nursing personnel of related specialties, thereby promoting the growth of related industries. And thesystem function is comprehensive, it can effectively check the anterior retinal segment, posterior retinal segment and iris of the user, and can collect and record the abnormal information of the user. At the same time, the user's identity can be recognized according to the biometrics on the user's retina.

The invention provides a retinal vessel segmentation method based on an improved U-Net network. Image enhancement is performed on a color eye fundus image, so that the contrast ratio between a blood vessel and a background in the image is improved, and a training data set is amplified. A U-Net encoder-decoder structure is used as a basic segmentation framework, a dense convolution block and a CDBR layer structure are designed to replace a traditional convolution block, learning of multi-scale feature information is achieved, and the feature extraction capacity of the model is improved. Meanwhile, an attention mechanism is introduced at a jump connection part of the model, so that the model is enabled to allocate weights again, the importance degree of a feature channel is adjusted, noise is suppressed, the problem of blood vessel information loss in an up-sampling process at a decoder end is solved, and a GAB-D2BUNet network model is constructed based on the above technologies. According to the method, an internationally disclosed retina fundus blood vessel data set DRIVE is adopted for training, and finally the optimal segmentation model is reserved to verify the segmentation performance of the model. The retina fundus blood vessel segmentation method achieves the task of accurately segmenting the retina fundus blood vessel, and has better segmentation performance.

The invention provides a retinal blood vessel branch angle calculation method and device and electronic equipment, and the method comprises the steps: obtaining a retinal fundus image, and carrying out the binarization processing of the retinal fundus image, and obtaining a black and white image; based on the black-and-white image, determining an optic disc center point, and based on the black-and-white image, determining two target branch blood vessels of the retina; and based on the optic disc center point and the two target branch blood vessels, obtaining the branch angle of the retinal blood vessels. According to the retinal blood vessel branch angle calculation method and device and the electronic equipment, the blood vessel branch angle of the human eye retina can be accurately calculated.

The invention belongs to the field of medical image segmentation, and provides a retinal vascular image segmentation method and system based on differential attention. The method comprises the following steps: acquiring a retinal vascular image; and obtaining a retinal fundus vascular image segmentation result based on the retinal vascular image and a differential attention-based multi-scale residual network, wherein the differential attention-based multi-scale residual network comprises a multi-scale input module, an encoder module, a differential amplification module and a decoder module; the multi-scale input module is used for extracting multi-scale information of the retinal vascular image; the encoder module is used for encoding the multi-scale information; the differential amplification module is used for extracting low-frequency information and high-frequency information of the encoded multi-scale information and then extracting features of the low-frequency information and the high-frequency information; and an attention mechanism is introduced into the decoder module, so that an attention degree of an area needing to be paid attention to is improved, irrelevant areas are inhibited, and the extracted low-frequency and high-frequency features are finally restored to original resolutions.

The invention discloses a deep learning-based fundus image classification device and method for diagnosing eye diseases. A deep learning-based fundus image classification device for diagnosing an eye disease according to one embodiment of the present invention comprises: a data preprocessing unit for normalizing image data before learning a model by preprocessing retinal fundus image data; and a retina image classification unit that classifies a retina image by training and testing the pre-processed retina fundus image data, and classifies the retina image using layered 10-fold cross validation.

The invention discloses a method for detecting exudate of a retinal fundus image. The method comprises the following steps: step 1, shooting a fundus image by using a mydriasis-free fundus camera; step 2, preprocessing the fundus image; step 3, enhancing the image; step 4, carrying out optic disk positioning on the image; step 5, positioning the exudate area; step 6, carrying out classification detection on the exudate areas; and step 7, generating a fundus image of the marked exudate. A color space model is used for preprocessing the fundus image, the image can be divided into color and grayscale information, the method is suitable for processing the gray scale image, the image enhancement effect is improved, the processed image is clearer, medical staff can judge the illness state conveniently, meanwhile, classification detection is conducted on the appearing area of suspicious penetrating fluid. After the detection is finished, the computer automatically prints a fundus image marked with the exudate.

The invention discloses a retinal blood vessel segmentation method based on a convolutional neural network, which includes: preprocessing the retinal fundus image; performing block extraction on training set images; constructing a convolutional neural network for blood vessel segmentation, and using the extracted image blocks to perform Training; in the prediction stage, multiple consecutive overlapping segments are extracted for each image, and the classification probability of each pixel is obtained by averaging multiple prediction results, and the final segmentation result map is obtained. The new convolutional neural network structure proposed by the present invention for retinal vessel segmentation is a symmetrical network based on the Encoder-Decoder structure, and two skip connections are added between the Encoder part and the Decoder part. The network can not only realize the end-to-end segmentation of retinal images, but also can obtain accurate segmentation results on a limited data set, and can effectively avoid the problem of gradient disappearance, which has certain advantages compared with existing algorithms.