Eye fundus image-based retina fixation point training method and device
Through the retinal gaze point training method based on fundus images, the patient's fundus images are collected and processed, and personalized training plans are formulated and optimized and adjusted, which solves the problem that personalized training plans cannot be formulated in the existing technology, which significantly improves the training effect.
Patent Information
- Application Number
- CN202410965850.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-18
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2044-07-18
AI Technical Summary
The existing retinal gaze training methods cannot formulate personalized training plans for patients based on fundus images, resulting in poor training results.
By formulating a personalized retinal gaze training plan based on fundus image acquisition, processing and analysis, and optimizing and adjusting it during the training process, using image acquisition module, image processing module, analysis and evaluation module, gaze training module and feedback optimization module to achieve Intelligent training.
The personalization and effect of retinal gaze point training has been improved, and the training effect and efficiency of patients have been improved through personalized plans and real-time optimization and adjustment.
Smart Images

Figure CN118902814A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of retinal training, and in particular to a retinal fixation point training method and device based on fundus images. Background Art
[0002] The retina is the inner layer of the eyeball wall, which is divided into the blind part and the visual part of the retina. The blind part includes the iris part of the retina and the ciliary body part of the retina, each attached to the inside of the iris and ciliary body, and is a component of the iris and ciliary body. The visual part of the retina is often referred to as the retina. It is a soft and transparent membrane that is tightly attached to the inner surface of the choroid and has the function of sensing light stimulation.
[0003] The Chinese patent with publication number CN118229706A discloses a single-stage segmentation method and device for retinal fundus images, which includes constructing a segmentation network model; obtaining retinal fundus image data, the retinal fundus image data including a plurality of retinal fundus image samples, each retinal fundus image sample including a retinal fundus image and a corresponding semantic annotation map; inputting the retinal fundus image into a convolutional feature extractor, and inputting the semantic annotation map corresponding to the retinal fundus image into an auxiliary task module, so as to train the segmentation network model and pre-position the position of the segmentation target during the training process; obtaining a retinal fundus image to be segmented, and segmenting the retinal fundus image to be segmented based on the trained segmentation network model to generate a target segmentation result map, thereby improving the segmentation efficiency and accuracy; however, the patent has the following defects:
[0004] The existing retinal fixation point training program cannot be customized for patients based on fundus images, resulting in poor retinal fixation point training results for patients. Summary of the invention
[0005] The purpose of the present invention is to provide a retinal fixation point training method and device based on fundus images, which can formulate a personalized retinal fixation point training plan for patients based on fundus images, improve the retinal fixation point training effect of patients, and solve the problems raised in the above-mentioned background technology.
[0006] To achieve the above object, the present invention provides the following technical solutions:
[0007] A retinal fixation point training method based on fundus images comprises the following steps:
[0008] S1: According to the retinal fixation point training requirements based on the fundus image, the fundus camera is adjusted to obtain the real-time fundus image of the patient based on the fundus camera;
[0009] S2: Acquire a real-time fundus image of the patient, perform filtering processing, brightness and contrast adjustment processing, and feature extraction processing on the real-time fundus image of the patient, and determine a fundus feature image of the patient;
[0010] S3: Analyze and evaluate the patient's fundus characteristic image based on the normal fundus standard image, and determine the patient's fundus image analysis and evaluation result;
[0011] S4: Develop a retinal fixation point training program based on fundus images to conduct intelligent retinal fixation point training for patients;
[0012] S5: Obtain patient feedback and suggestions based on retinal fixation training, and optimize and adjust the retinal fixation training program based on patient feedback and suggestions.
[0013] According to another aspect of the present invention, a retinal fixation point training device based on fundus images is provided, which is used to implement the retinal fixation point training method based on fundus images as described above, comprising:
[0014] An image acquisition module, used to acquire real-time images of the patient's fundus;
[0015] An image processing module is used to process the real-time fundus image of the patient and determine the fundus feature image of the patient;
[0016] An analysis and evaluation module is used to analyze and evaluate the patient's fundus feature image and determine the patient's fundus image analysis and evaluation results;
[0017] The fixation training module is used to formulate a retinal fixation point training program based on fundus images and conduct intelligent retinal fixation point training for patients;
[0018] The feedback optimization module is used to obtain patient feedback suggestions based on retinal fixation point training and optimize the retinal fixation point training program.
[0019] Preferably, the image acquisition module includes:
[0020] A shooting adjustment unit, used for performing shooting adjustment on the fundus camera;
[0021] According to the retinal fixation point training requirements based on fundus images, the fundus camera is slowly moved toward the patient, and the patient is made to look at the fixation light built into the fundus camera examination lens;
[0022] Rotate the focus adjustment knob to align the two cleavage lines, and move the fundus camera slightly up and down, front and back, left and right, so that the two distance indicator points are clearly outlined and centered;
[0023] An image acquisition unit, used for acquiring a real-time fundus image of the patient;
[0024] The patient's eyes are photographed using a fundus camera to obtain real-time fundus images of the patient.
[0025] Preferably, the image processing module includes:
[0026] An image filtering unit, used for filtering the real-time fundus image of the patient;
[0027] Obtain real-time images of the patient's fundus;
[0028] Based on the filter, the real-time fundus image of the patient is filtered;
[0029] Under the condition of retaining the detailed features of the patient's real-time fundus image, the noise of the patient's real-time fundus image is suppressed, and the noise in the patient's real-time fundus image is effectively removed;
[0030] An image adjustment unit, used for adjusting the brightness and contrast of the patient's real-time fundus image;
[0031] Obtain real-time images of the patient's fundus;
[0032] Use image segmentation technology to divide the patient's real-time fundus image into regions;
[0033] Analyze the average brightness of each area;
[0034] When the average brightness of a certain area is too low, increase the brightness of that area;
[0035] When the average brightness of a certain area is too high, reduce the brightness of the area;
[0036] The brightness of the entire patient's fundus real-time image is kept consistent through interpolation or reconstruction techniques;
[0037] Based on the adaptive adjustment method of global contrast and adjacent pixel relationship, the grayscale difference of pixels around each pixel in the patient's real-time fundus image is calculated;
[0038] Automatically adjust the contrast of the pixel according to the grayscale difference;
[0039] Determine the high-definition real-time image of the patient's fundus;
[0040] A feature extraction unit, used for extracting features from a real-time fundus image of a patient;
[0041] Obtain high-definition real-time images of the patient's fundus;
[0042] Extract features from high-definition real-time images of the patient's fundus;
[0043] Determine the patient's fundus feature image.
[0044] Preferably, the analysis and evaluation module includes:
[0045] A standard storage unit, used for storing a normal fundus standard image;
[0046] According to the retinal fixation point training requirements based on the fundus image, a normal fundus standard image is pre-set, and the pre-set normal fundus standard image is stored;
[0047] An indexing and retrieving unit, used for indexing and retrieving a normal fundus standard image;
[0048] According to the retinal fixation point training requirements based on the fundus image, the pre-set normal fundus standard image is indexed, and the indexed normal fundus standard image is retrieved;
[0049] An analysis and evaluation unit, used for analyzing and evaluating the patient's fundus feature images;
[0050] Acquire the patient's fundus feature image;
[0051] Acquire a normal fundus standard image;
[0052] Analyze and evaluate the patient's fundus feature images based on the normal fundus standard images;
[0053] Determine the patient's fundus image analysis and evaluation results.
[0054] Preferably, the patient's fundus feature image is analyzed and evaluated, and the following operations are performed:
[0055] When the patient's fundus feature image is within the range of the normal fundus standard image, the patient's fundus image analysis and evaluation result is that the patient's fundus image is normal;
[0056] When the patient's fundus feature image is not within the range of the normal fundus standard image, the patient's fundus image analysis and evaluation result is that the patient's fundus image is abnormal.
[0057] Preferably, the analysis and evaluation unit comprises:
[0058] The historical assessment data acquisition subunit is used to:
[0059] When the evaluation result shows that the patient's fundus image is abnormal, historical evaluation data is obtained;
[0060] The historical evaluation data includes: abnormal fundus images of different abnormal types, and diagnostic data corresponding to different types of abnormal fundus feature images;
[0061] The data set establishment subunit is used to read the abnormal fundus images corresponding to each abnormal type respectively and establish an abnormal fundus image data set;
[0062] A geometric identification subunit is used to perform geometric identification on the abnormal fundus image corresponding to each abnormal type, and obtain geometric features of the abnormal fundus image corresponding to each abnormal type;
[0063] Model building subunit for:
[0064] The geometric features of each type of abnormal fundus image are learned, and an initial abnormality recognition model based on the abnormal fundus image dataset is constructed;
[0065] A first limiting condition of the initial abnormality identification model based on the abnormality type, and at the same time, a second limiting condition of the initial abnormality identification model based on the diagnostic data;
[0066] The first limiting condition and the second limiting condition are configured in the abnormality recognition model, and a target abnormality recognition model is obtained according to the configuration result;
[0067] The abnormal region acquisition subunit is used to read the abnormal fundus image of the current patient when the fundus image of the patient is abnormal, overlap the abnormal fundus image of the current patient with the normal fundus standard image, and match the abnormal region of the abnormal fundus image according to the overlap result;
[0068] The model recognition subunit is used to input the abnormal area into the target abnormality recognition model for recognition, and output the target abnormality type corresponding to the abnormal area and the target diagnostic data matching the abnormal area based on the recognition result;
[0069] The abnormality assessment report subunit is used to generate an abnormality assessment report when the patient's fundus image is abnormal based on the target abnormality type and target diagnostic data.
[0070] Preferably, the gaze training module comprises:
[0071] A program formulation unit, used for formulating a retinal fixation point training program based on fundus images;
[0072] Obtain the patient's fundus image analysis and evaluation results;
[0073] Conduct in-depth mining and correlation analysis on the patient's fundus image analysis and evaluation results;
[0074] Combined with the patient's fundus feature images, a retinal fixation point training program based on fundus images is developed;
[0075] Intelligent training unit, used to conduct intelligent retinal fixation training for patients;
[0076] Obtain retinal fixation point training scheme based on fundus images;
[0077] The patients are given intelligent retinal fixation training according to the retinal fixation training program.
[0078] Preferably, the feedback optimization module includes:
[0079] A feedback suggestion unit, used for obtaining patient feedback suggestions based on retinal fixation point training;
[0080] When patients are undergoing intelligent retinal fixation point training, the training feedback and suggestions of patients for intelligent retinal fixation point training are collected in real time;
[0081] Identify patient feedback recommendations based on retinal fixation training;
[0082] An optimization and adjustment unit, used for optimizing and adjusting the retinal fixation point training program;
[0083] Obtain patient feedback suggestions based on retinal fixation training;
[0084] Optimize and adjust the retinal fixation training program based on patient feedback;
[0085] Determine the retinal fixation point training program that is most suitable for the patient and provide the possibility for individualized retinal fixation point training.
[0086] Preferably, the feedback optimization module includes:
[0087] A fundus real-time image acquisition unit, used to obtain the training dimensions of the retinal fixation point training program, and when performing retinal fixation point training, to acquire the patient's fundus real-time image under each training dimension;
[0088] A first calculation unit is used to read a normal fundus standard image and calculate a target similarity between the normal fundus standard image and the fundus real-time image;
[0089]
[0090] in, represents the target similarity; i represents the ordinal value of the training dimension; n represents the total number of training dimensions; M represents the area value of the normal fundus standard image; m ir represents the similarity area value between the real-time fundus image and the normal fundus standard image under the i-th training dimension; m iw represents the difference area value between the real-time fundus image and the normal fundus standard image under the i-th training dimension;
[0091] The second computing unit is configured to:
[0092] Determine the patient satisfaction value corresponding to each training dimension in the retinal fixation training program based on patient feedback suggestions;
[0093] According to the target similarity and the patient satisfaction value corresponding to each training dimension in the retinal fixation point training scheme, a comprehensive evaluation value when the patient is trained on the retinal fixation point based on the retinal fixation point scheme is calculated;
[0094]
[0095] Where F represents the comprehensive evaluation value of the retinal fixation point training for patients based on the retinal fixation point scheme; y i represents the patient satisfaction value corresponding to the i-th training dimension; Y i represents the patient satisfaction threshold corresponding to the i-th training dimension, and y i ≤Y i ; σ i represents the influence weight of the i-th training dimension on the comprehensive evaluation value; Φ represents the similarity threshold, and ψ represents the influence weight of target similarity on the comprehensive evaluation value;
[0096] Optimize and adjust the judgment unit to:
[0097] Obtaining a comprehensive evaluation standard value, and comparing the comprehensive evaluation standard value with the comprehensive evaluation value to determine whether the retinal fixation point training program needs to be optimized and adjusted;
[0098] If the comprehensive evaluation standard value is equal to or greater than the comprehensive evaluation value, it is determined that there is no need to optimize the retinal fixation point training program;
[0099] Otherwise, it is determined that the retinal fixation point training program needs to be optimized and adjusted.
[0100] Compared with the prior art, the present invention has the following beneficial effects:
[0101] The present invention adjusts the shooting of a fundus camera according to the demand for retinal fixation point training based on fundus images, obtains real-time fundus images of patients based on the fundus camera, processes the real-time fundus images of patients, determines fundus feature images of patients, analyzes and evaluates the fundus feature images of patients based on normal fundus standard images, determines fundus image analysis and evaluation results of patients, formulates a retinal fixation point training program based on fundus images, conducts intelligent retinal fixation point training for patients, and simultaneously obtains patient feedback and suggestions based on retinal fixation point training, optimizes and adjusts the retinal fixation point training program based on the patient feedback and suggestions, provides possibility for individualized retinal fixation point training, can formulate individualized retinal fixation point training programs for patients based on fundus images, and can improve the retinal fixation point training effect of patients.
[0102] By determining the historical evaluation data, the initial abnormality recognition model can be effectively constructed according to the diagnostic data corresponding to abnormal fundus images of different abnormal types and different types of abnormal fundus feature images. The initial abnormality recognition model can be effectively configured based on the first limiting condition and the second limiting condition to obtain the target abnormality recognition model, thereby ensuring the effectiveness and intelligence of the target abnormality recognition model, and then achieving accurate identification of abnormal areas, thereby effectively acquiring the abnormality evaluation report through the output target type and target diagnostic data, and the abnormality evaluation report can provide strong and reliable data support for generating retinal fixation point training programs.
[0103] By determining the training dimensions of the retinal fixation point training program, the real-time fundus image of the patient under each training dimension can be effectively collected, and then the target similarity can be calculated. By determining the patient satisfaction value corresponding to each training dimension and combining the target similarity, the comprehensive evaluation value of the retinal fixation point training can be effectively calculated, and then the judgment on whether the retinal fixation point training program needs to be optimized and adjusted can be effectively realized. The monitoring and judgment on whether the retinal fixation point training program needs to be optimized and adjusted can be effectively realized, so as to achieve real-time optimization processing when the retinal fixation point training program needs to be optimized and adjusted, thereby avoiding the cost waste of optimizing and adjusting the retinal fixation point training program at any time, and achieving intelligent regulation of optimizing and adjusting the retinal fixation point training program. BRIEF DESCRIPTION OF THE DRAWINGS
[0104] Figure 1 It is a flow chart of the retinal fixation point training method based on fundus images of the present invention;
[0105] Figure 2 This is a framework diagram of the retinal fixation point training device based on fundus images of the present invention. DETAILED DESCRIPTION
[0106] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0107] In order to solve the problem that the existing retinal fixation training program cannot be customized for patients based on fundus images, resulting in poor retinal fixation training results for patients, please refer to Figure 1-Figure 2 , this embodiment provides the following technical solutions:
[0108] A retinal fixation point training device based on fundus images comprises: an image acquisition module, an image processing module, an analysis and evaluation module, a fixation training module and a feedback optimization module.
[0109] It should be noted that through the interactive communication among the image acquisition module, image processing module, analysis and evaluation module, gaze training module and feedback optimization module, individualized retinal gaze point training is possible, and personalized retinal gaze point training plans can be formulated for patients based on fundus images, which can improve the retinal gaze point training effect of patients.
[0110] Among them, the image acquisition module is used to collect real-time images of the patient's fundus;
[0111] In this embodiment, as a preferred technical solution of the present invention, the image acquisition module includes:
[0112] A shooting adjustment unit, used for performing shooting adjustment on the fundus camera;
[0113] According to the retinal fixation point training requirements based on fundus images, the fundus camera is slowly moved toward the patient, and the patient is made to look at the fixation light built into the fundus camera examination lens;
[0114] Rotate the focus adjustment knob to align the two cleavage lines, and move the fundus camera slightly up and down, front and back, left and right, so that the two distance indicator points are clearly outlined and centered;
[0115] An image acquisition unit, used for acquiring a real-time fundus image of the patient;
[0116] The patient's eyes are photographed using a fundus camera to obtain real-time fundus images of the patient.
[0117] The image processing module is used to process the real-time fundus image of the patient and determine the fundus feature image of the patient;
[0118] In this embodiment, as a preferred technical solution of the present invention, the image processing module includes:
[0119] An image filtering unit, used for filtering the real-time fundus image of the patient;
[0120] Obtain real-time images of the patient's fundus;
[0121] Based on the filter, the real-time fundus image of the patient is filtered;
[0122] Under the condition of retaining the detailed features of the patient's real-time fundus image, the noise of the patient's real-time fundus image is suppressed, and the noise in the patient's real-time fundus image is effectively removed;
[0123] An image adjustment unit, used for adjusting the brightness and contrast of the patient's real-time fundus image;
[0124] Obtain real-time images of the patient's fundus;
[0125] Use image segmentation technology to divide the patient's real-time fundus image into regions;
[0126] Analyze the average brightness of each area;
[0127] When the average brightness of a certain area is too low, increase the brightness of that area;
[0128] When the average brightness of a certain area is too high, reduce the brightness of the area;
[0129] The brightness of the entire patient's fundus real-time image is kept consistent through interpolation or reconstruction techniques;
[0130] Based on the adaptive adjustment method of global contrast and adjacent pixel relationship, the grayscale difference of pixels around each pixel in the patient's real-time fundus image is calculated;
[0131] Automatically adjust the contrast of the pixel according to the grayscale difference;
[0132] Determine the high-definition real-time image of the patient's fundus;
[0133] A feature extraction unit, used for extracting features from a real-time fundus image of a patient;
[0134] Obtain high-definition real-time images of the patient's fundus;
[0135] Extract features from high-definition real-time images of the patient's fundus;
[0136] Determine the patient's fundus feature image.
[0137] The analysis and evaluation module is used to analyze and evaluate the patient's fundus feature image and determine the patient's fundus image analysis and evaluation results;
[0138] In this embodiment, as a preferred technical solution of the present invention, the analysis and evaluation module includes:
[0139] A standard storage unit, used for storing a normal fundus standard image;
[0140] According to the retinal fixation point training requirements based on the fundus image, a normal fundus standard image is pre-set, and the pre-set normal fundus standard image is stored;
[0141] An indexing and retrieving unit, used for indexing and retrieving a normal fundus standard image;
[0142] According to the retinal fixation point training requirements based on the fundus image, the pre-set normal fundus standard image is indexed, and the indexed normal fundus standard image is retrieved;
[0143] An analysis and evaluation unit, used for analyzing and evaluating the patient's fundus feature images;
[0144] Acquire the patient's fundus feature image;
[0145] Acquire a normal fundus standard image;
[0146] Analyze and evaluate the patient's fundus feature images based on the normal fundus standard images;
[0147] Determine the patient's fundus image analysis and evaluation results.
[0148] In this embodiment, as a preferred technical solution of the present invention, the patient's fundus feature image is analyzed and evaluated, and the following operations are performed:
[0149] When the patient's fundus feature image is within the range of the normal fundus standard image, the patient's fundus image analysis and evaluation result is that the patient's fundus image is normal;
[0150] When the patient's fundus feature image is not within the range of the normal fundus standard image, the patient's fundus image analysis and evaluation result is that the patient's fundus image is abnormal.
[0151] Among them, the fixation training module is used to formulate a retinal fixation point training program based on fundus images and conduct intelligent retinal fixation point training for patients;
[0152] In this embodiment, as a preferred technical solution of the present invention, the gaze training module includes:
[0153] A program formulation unit, used for formulating a retinal fixation point training program based on fundus images;
[0154] Obtain the patient's fundus image analysis and evaluation results;
[0155] Conduct in-depth mining and correlation analysis on the patient's fundus image analysis and evaluation results;
[0156] Combined with the patient's fundus feature images, a retinal fixation point training program based on fundus images is developed;
[0157] Intelligent training unit, used to conduct intelligent retinal fixation training for patients;
[0158] Obtain retinal fixation point training scheme based on fundus images;
[0159] The patients are given intelligent retinal fixation training according to the retinal fixation training program.
[0160] Among them, the feedback optimization module is used to obtain patient feedback suggestions based on retinal fixation point training and optimize the retinal fixation point training program.
[0161] In this embodiment, as a preferred technical solution of the present invention, the feedback optimization module includes:
[0162] A feedback suggestion unit, used for obtaining patient feedback suggestions based on retinal fixation point training;
[0163] When patients are undergoing intelligent retinal fixation point training, the training feedback and suggestions of patients for intelligent retinal fixation point training are collected in real time;
[0164] Identify patient feedback recommendations based on retinal fixation training;
[0165] An optimization and adjustment unit, used for optimizing and adjusting the retinal fixation point training program;
[0166] Obtain patient feedback suggestions based on retinal fixation training;
[0167] Optimize and adjust the retinal fixation training program based on patient feedback;
[0168] Determine the retinal fixation point training program that is most suitable for the patient and provide the possibility for individualized retinal fixation point training.
[0169] Therefore, according to the demand for retinal fixation point training based on fundus images, the fundus camera is adjusted to obtain real-time fundus images of patients based on the fundus camera, the real-time fundus images of patients are processed to determine the characteristic fundus images of patients, and the characteristic fundus images of patients are analyzed and evaluated based on normal fundus standard images to determine the analysis and evaluation results of the fundus images of patients, and a retinal fixation point training program based on fundus images is formulated to conduct intelligent retinal fixation point training for patients. At the same time, patient feedback and suggestions based on retinal fixation point training are obtained, and the retinal fixation point training program is optimized and adjusted based on the patient feedback and suggestions, which provides the possibility for individualized retinal fixation point training, and can formulate personalized retinal fixation point training programs for patients based on fundus images, which can improve the retinal fixation point training effect of patients.
[0170] In order to better demonstrate the retinal fixation point training process based on fundus images, this embodiment now provides a retinal fixation point training method based on fundus images, which is implemented based on the retinal fixation point training device based on fundus images as described above, and includes the following steps:
[0171] S1: According to the retinal fixation point training requirements based on the fundus image, the fundus camera is adjusted to obtain the real-time fundus image of the patient based on the fundus camera;
[0172] S2: Acquire a real-time fundus image of the patient, perform filtering processing, brightness and contrast adjustment processing, and feature extraction processing on the real-time fundus image of the patient, and determine a fundus feature image of the patient;
[0173] S3: Analyze and evaluate the patient's fundus characteristic image based on the normal fundus standard image, and determine the patient's fundus image analysis and evaluation result;
[0174] S4: Develop a retinal fixation point training program based on fundus images to conduct intelligent retinal fixation point training for patients;
[0175] S5: Obtain patient feedback and suggestions based on retinal fixation point training, optimize and adjust the retinal fixation point training program based on patient feedback and suggestions, provide the possibility for individualized retinal fixation point training, and develop personalized retinal fixation point training programs for patients based on fundus images, which can improve the effect of patients' retinal fixation point training.
[0176] This embodiment also provides a retinal fixation point training device based on fundus images, an analysis and evaluation unit, including:
[0177] The historical assessment data acquisition subunit is used to:
[0178] When the evaluation result shows that the patient's fundus image is abnormal, historical evaluation data is obtained;
[0179] The historical evaluation data includes: abnormal fundus images of different abnormal types, and diagnostic data corresponding to different types of abnormal fundus feature images;
[0180] The data set establishment subunit is used to read the abnormal fundus images corresponding to each abnormal type respectively and establish an abnormal fundus image data set;
[0181] A geometric identification subunit is used to perform geometric identification on the abnormal fundus image corresponding to each abnormal type, and obtain geometric features of the abnormal fundus image corresponding to each abnormal type;
[0182] Model building subunit for:
[0183] The geometric features of each type of abnormal fundus image are learned, and an initial abnormality recognition model based on the abnormal fundus image dataset is constructed;
[0184] A first limiting condition of the initial abnormality identification model based on the abnormality type, and at the same time, a second limiting condition of the initial abnormality identification model based on the diagnostic data;
[0185] The first limiting condition and the second limiting condition are configured in the abnormality recognition model, and a target abnormality recognition model is obtained according to the configuration result;
[0186] The abnormal region acquisition subunit is used to read the abnormal fundus image of the current patient when the fundus image of the patient is abnormal, overlap the abnormal fundus image of the current patient with the normal fundus standard image, and match the abnormal region of the abnormal fundus image according to the overlap result;
[0187] The model recognition subunit is used to input the abnormal area into the target abnormality recognition model for recognition, and output the target abnormality type corresponding to the abnormal area and the target diagnostic data matching the abnormal area based on the recognition result;
[0188] The abnormality assessment report subunit is used to generate an abnormality assessment report when the patient's fundus image is abnormal based on the target abnormality type and target diagnostic data.
[0189] In this embodiment, the historical evaluation data may be comprehensive data of abnormal fundus images corresponding to all patients determined to have abnormal fundus images and corresponding abnormal diagnostic data, wherein the diagnostic data may include diagnostic results based on the abnormal fundus images.
[0190] In this embodiment, the geometric features may be shape features distributed such as the contour of the abnormal fundus image corresponding to each abnormality type.
[0191] In this embodiment, the first limiting condition may be used as configuration information of the abnormality recognition model for the abnormality type, that is, the abnormality recognition model may accurately locate the abnormality type after recognizing the abnormal fundus image.
[0192] In this embodiment, the second limiting condition may be used as configuration information of the abnormality recognition model for the diagnostic data, that is, the abnormality recognition model may accurately locate the corresponding diagnostic result and other data after recognizing the abnormal fundus image.
[0193] In this embodiment, one abnormality type corresponds to one diagnostic data (ie, diagnostic result).
[0194] The working principle and beneficial effects of the above technical solution are: by determining the historical evaluation data, the initial abnormality recognition model can be effectively constructed according to the diagnostic data corresponding to abnormal fundus images of different abnormal types and different types of abnormal fundus feature images. The initial abnormality recognition model can be effectively configured based on the first limiting condition and the second limiting condition to obtain the target abnormality recognition model, thereby ensuring the effectiveness and intelligence of the target abnormality recognition model, and then achieving accurate identification of abnormal areas, thereby effectively obtaining the abnormality evaluation report through the output target type and target diagnostic data, and the abnormality evaluation report can provide strong and reliable data support for generating retinal fixation point training programs.
[0195] This embodiment also provides a retinal fixation point training device based on fundus images, a feedback optimization module, including:
[0196] A fundus real-time image acquisition unit, used to obtain the training dimensions of the retinal fixation point training program, and when performing retinal fixation point training, to acquire the patient's fundus real-time image under each training dimension;
[0197] A first calculation unit is used to read a normal fundus standard image and calculate a target similarity between the normal fundus standard image and the fundus real-time image;
[0198]
[0199] in, represents the target similarity; i represents the ordinal value of the training dimension; n represents the total number of training dimensions; M represents the area value of the normal fundus standard image; m ir represents the similarity area value between the real-time fundus image and the normal fundus standard image under the i-th training dimension; m iw represents the difference area value between the real-time fundus image and the normal fundus standard image under the i-th training dimension;
[0200] The second computing unit is configured to:
[0201] Determine the patient satisfaction value corresponding to each training dimension in the retinal fixation training program based on patient feedback suggestions;
[0202] According to the target similarity and the patient satisfaction value corresponding to each training dimension in the retinal fixation point training scheme, a comprehensive evaluation value when the patient is trained on the retinal fixation point based on the retinal fixation point scheme is calculated;
[0203]
[0204] Where F represents the comprehensive evaluation value of the retinal fixation point training for patients based on the retinal fixation point scheme; y i represents the patient satisfaction value corresponding to the i-th training dimension; Y i represents the patient satisfaction threshold corresponding to the i-th training dimension, and y i ≤Y i ; σ i represents the influence weight of the i-th training dimension on the comprehensive evaluation value; Φ represents the similarity threshold, and ψ represents the influence weight of target similarity on the comprehensive evaluation value;
[0205] Optimize and adjust the judgment unit to:
[0206] Obtaining a comprehensive evaluation standard value, and comparing the comprehensive evaluation standard value with the comprehensive evaluation value to determine whether the retinal fixation point training program needs to be optimized and adjusted;
[0207] If the comprehensive evaluation standard value is equal to or greater than the comprehensive evaluation value, it is determined that there is no need to optimize the retinal fixation point training program;
[0208] Otherwise, it is determined that the retinal fixation point training program needs to be optimized and adjusted.
[0209] In this embodiment, r and w are used to distinguish similar region values from different region values, respectively, wherein r represents similarity and w represents difference.
[0210] In this embodiment, the similarity threshold may be set in advance and used to measure the proportion of the target similarity to the similarity threshold, so as to better achieve the objectivity of obtaining the comprehensive evaluation value.
[0211] In this embodiment, the patient satisfaction threshold can be set in advance, and one training dimension corresponds to one patient satisfaction threshold, which is used to measure the proportion of the patient satisfaction value to the patient satisfaction threshold under the training dimension, so as to realize the patient satisfaction level under the training dimension.
[0212] In this embodiment, the comprehensive evaluation standard value may be set in advance and used as a standard for measuring whether the retinal fixation point training program needs to be optimized and adjusted.
[0213] In this embodiment, the training dimensions may include, but are not limited to: fixed gaze training, target tracking, scanning training, etc.
[0214] In this embodiment, the comprehensive evaluation value can be an evaluation degree of the current retinal fixation point training program obtained after comprehensive measurement of target similarity and patient satisfaction, so that effective judgment on optimizing and adjusting the retinal fixation point training program can be achieved based on the evaluation degree.
[0215] The working principle and beneficial effects of the above technical solution are: by determining the training dimensions of the retinal fixation point training program, the real-time fundus image collection of the patient under each training dimension can be effectively realized, and then the target similarity can be calculated. By determining the patient satisfaction value corresponding to each training dimension and combining the target similarity, the comprehensive evaluation value of the retinal fixation point training can be effectively calculated, and then the judgment on whether the retinal fixation point training program needs to be optimized and adjusted can be effectively realized, and the monitoring and judgment on whether the retinal fixation point training program needs to be optimized and adjusted can be effectively realized, so as to achieve real-time optimization processing when the retinal fixation point training program needs to be optimized and adjusted, thereby avoiding the cost waste of optimizing and adjusting the retinal fixation point training program at any time, and achieving intelligent regulation of optimizing and adjusting the retinal fixation point training program.
[0216] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.
[0217] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.
Claims
1. A retinal fixation point training method based on fundus images, characterized in that: The steps include: S1: According to the retinal fixation point training requirements based on the fundus image, the fundus camera is adjusted to obtain the real-time fundus image of the patient based on the fundus camera; S2: Acquire a real-time fundus image of the patient, perform filtering processing, brightness and contrast adjustment processing, and feature extraction processing on the real-time fundus image of the patient, and determine a fundus feature image of the patient; S3: Analyze and evaluate the patient's fundus characteristic image based on the normal fundus standard image, and determine the patient's fundus image analysis and evaluation result; S4: Develop a retinal fixation point training program based on fundus images to conduct intelligent retinal fixation point training for patients; S5: Obtain patient feedback and suggestions based on retinal fixation training, and optimize and adjust the retinal fixation training program based on patient feedback and suggestions.
2. A retinal fixation point training device based on fundus images, used to implement a retinal fixation point training method based on fundus images as claimed in claim 1, characterized in that: include: An image acquisition module, used to acquire real-time images of the patient's fundus; An image processing module is used to process the real-time fundus image of the patient and determine the fundus feature image of the patient; An analysis and evaluation module is used to analyze and evaluate the patient's fundus feature image and determine the patient's fundus image analysis and evaluation results; The fixation training module is used to formulate a retinal fixation point training program based on fundus images and conduct intelligent retinal fixation point training for patients; The feedback optimization module is used to obtain patient feedback suggestions based on retinal fixation point training and optimize the retinal fixation point training program.
3. The retinal fixation point training device based on fundus images according to claim 2, characterized in that: The image acquisition module comprises: A shooting adjustment unit, used for performing shooting adjustment on the fundus camera; According to the retinal fixation point training requirements based on fundus images, the fundus camera is slowly moved toward the patient, and the patient is made to look at the fixation light built into the fundus camera examination lens; Rotate the focus adjustment knob to align the two cleavage lines, and move the fundus camera slightly up and down, front and back, left and right, so that the two distance indicator points are clearly outlined and centered; An image acquisition unit, used for acquiring a real-time fundus image of the patient; The patient's eyes are photographed using a fundus camera to obtain real-time fundus images of the patient.
4. The retinal fixation point training device based on fundus images according to claim 3, characterized in that: The image processing module comprises: An image filtering unit, used for filtering the real-time fundus image of the patient; Obtain real-time images of the patient's fundus; Based on the filter, the real-time fundus image of the patient is filtered; Under the condition of retaining the detailed features of the patient's real-time fundus image, the noise of the patient's real-time fundus image is suppressed, and the noise in the patient's real-time fundus image is effectively removed; An image adjustment unit, used for adjusting the brightness and contrast of the patient's real-time fundus image; Obtain real-time images of the patient's fundus; Use image segmentation technology to divide the patient's real-time fundus image into regions; Analyze the average brightness of each area; When the average brightness of a certain area is too low, increase the brightness of that area; When the average brightness of a certain area is too high, reduce the brightness of the area; The brightness of the entire patient's fundus real-time image is kept consistent through interpolation or reconstruction techniques; Based on the adaptive adjustment method of global contrast and adjacent pixel relationship, the grayscale difference of pixels around each pixel in the patient's real-time fundus image is calculated; Automatically adjust the contrast of the pixel according to the grayscale difference; Determine the high-definition real-time image of the patient's fundus; A feature extraction unit, used for extracting features from a real-time fundus image of a patient; Obtain high-definition real-time images of the patient's fundus; Extract features from high-definition real-time images of the patient's fundus; Determine the patient's fundus feature image.
5. The retinal fixation point training device based on fundus images according to claim 4, characterized in that: The analysis and evaluation module includes: A standard storage unit, used for storing a normal fundus standard image; According to the retinal fixation point training requirements based on the fundus image, a normal fundus standard image is pre-set, and the pre-set normal fundus standard image is stored; An indexing and retrieving unit, used for indexing and retrieving a normal fundus standard image; According to the retinal fixation point training requirements based on the fundus image, the pre-set normal fundus standard image is indexed, and the indexed normal fundus standard image is retrieved; An analysis and evaluation unit, used for analyzing and evaluating the patient's fundus feature images; Acquire the patient's fundus feature image; Acquire a normal fundus standard image; Analyze and evaluate the patient's fundus feature images based on the normal fundus standard images; Determine the patient's fundus image analysis and evaluation results.
6. The retinal fixation point training device based on fundus images according to claim 5, characterized in that: Analyze and evaluate the patient's fundus feature images and perform the following operations: When the patient's fundus feature image is within the range of the normal fundus standard image, the patient's fundus image analysis and evaluation result is that the patient's fundus image is normal; When the patient's fundus feature image is not within the range of the normal fundus standard image, the patient's fundus image analysis and evaluation result is that the patient's fundus image is abnormal.
7. The retinal fixation point training device based on fundus images according to claim 5, characterized in that: Analytical evaluation unit, including: The historical assessment data acquisition subunit is used to: When the evaluation result shows that the patient's fundus image is abnormal, historical evaluation data is obtained; The historical evaluation data includes: abnormal fundus images of different abnormal types, and diagnostic data corresponding to different types of abnormal fundus feature images; The data set establishment subunit is used to read the abnormal fundus images corresponding to each abnormal type respectively and establish an abnormal fundus image data set; A geometric recognition subunit is used to perform geometric recognition on the abnormal fundus image corresponding to each abnormal type, and obtain geometric features of the abnormal fundus image corresponding to each abnormal type; Model building subunit for: The geometric features of each type of abnormal fundus image are learned, and an initial abnormality recognition model based on the abnormal fundus image dataset is constructed; A first limiting condition of the initial abnormality identification model based on the abnormality type, and at the same time, a second limiting condition of the initial abnormality identification model based on the diagnostic data; The first limiting condition and the second limiting condition are configured in the abnormality recognition model, and a target abnormality recognition model is obtained according to the configuration result; The abnormal region acquisition subunit is used to read the abnormal fundus image of the current patient when the fundus image of the patient is abnormal, overlap the abnormal fundus image of the current patient with the normal fundus standard image, and match the abnormal region of the abnormal fundus image according to the overlap result; The model recognition subunit is used to input the abnormal area into the target abnormality recognition model for recognition, and output the target abnormality type corresponding to the abnormal area and the target diagnostic data matching the abnormal area based on the recognition result; The abnormality assessment report subunit is used to generate an abnormality assessment report when the patient's fundus image is abnormal based on the target abnormality type and target diagnostic data.
8. The retinal fixation point training device based on fundus images according to claim 6, characterized in that: The gaze training module comprises: A program formulation unit, used for formulating a retinal fixation point training program based on fundus images; Obtain the patient's fundus image analysis and evaluation results; Conduct in-depth mining and correlation analysis on the patient's fundus image analysis and evaluation results; Combined with the patient's fundus feature images, a retinal fixation point training program based on fundus images is developed; Intelligent training unit, used to conduct intelligent retinal fixation training for patients; Obtain retinal fixation point training scheme based on fundus images; The patients are given intelligent retinal fixation training according to the retinal fixation training program.
9. The retinal fixation point training device based on fundus images according to claim 8, characterized in that: The feedback optimization module comprises: A feedback suggestion unit, used for obtaining patient feedback suggestions based on retinal fixation point training; When patients are undergoing intelligent retinal fixation point training, the training feedback and suggestions of patients for intelligent retinal fixation point training are collected in real time; Identify patient feedback recommendations based on retinal fixation training; An optimization and adjustment unit, used for optimizing and adjusting the retinal fixation point training program; Obtain patient feedback suggestions based on retinal fixation training; Optimize and adjust the retinal fixation training program based on patient feedback; Determine the retinal fixation point training program that is most suitable for the patient and provide the possibility for individualized retinal fixation point training.
10. The retinal fixation point training device based on fundus images according to claim 2, characterized in that: Feedback optimization module, including: A fundus real-time image acquisition unit, used to obtain the training dimensions of the retinal fixation point training program, and when performing retinal fixation point training, to acquire the patient's fundus real-time image under each training dimension; A first calculation unit is used to read a normal fundus standard image and calculate a target similarity between the normal fundus standard image and the fundus real-time image; in, represents the target similarity; i represents the ordinal value of the training dimension; n represents the total number of training dimensions; M represents the area value of the normal fundus standard image; m ir represents the similarity area value between the real-time fundus image and the normal fundus standard image under the i-th training dimension; m iw represents the difference area value between the real-time fundus image and the normal fundus standard image under the i-th training dimension; The second computing unit is configured to: Determine the patient satisfaction value corresponding to each training dimension in the retinal fixation training program based on patient feedback suggestions; According to the target similarity and the patient satisfaction value corresponding to each training dimension in the retinal fixation point training scheme, a comprehensive evaluation value when the patient is trained on the retinal fixation point based on the retinal fixation point scheme is calculated; Where F represents the comprehensive evaluation value of the retinal fixation point training for patients based on the retinal fixation point scheme; y i represents the patient satisfaction value corresponding to the i-th training dimension; Y i represents the patient satisfaction threshold corresponding to the i-th training dimension, and y i ≤Y i ; σ i represents the influence weight of the i-th training dimension on the comprehensive evaluation value; Φ represents the similarity threshold, and ψ represents the influence weight of target similarity on the comprehensive evaluation value; Optimize and adjust the judgment unit to: Obtaining a comprehensive evaluation standard value, and comparing the comprehensive evaluation standard value with the comprehensive evaluation value to determine whether the retinal fixation point training program needs to be optimized and adjusted; If the comprehensive evaluation standard value is equal to or greater than the comprehensive evaluation value, it is determined that there is no need to optimize the retinal fixation point training program; Otherwise, it is determined that the retinal fixation point training program needs to be optimized and adjusted.
Citation Information
Patent Citations
Image brightness self-adaptive adjusting method for vehicle-mounted image
CN115100083A
Integrated visual inspection and training device
CN115737388A
Ophthalmology full-process medical service system and method based on intelligent technology
CN117747077A
Cited By
Gastric content estimation method based on intelligent image recognition
CN119326446A
