Glaucoma multi-mode intelligent identification method and system based on transfer learning

Abstract

The invention relates to the technical field of computer vision, in particular to a multi-modal intelligent glaucoma recognition method and system based on transfer learning, and the method comprises the following steps: collecting fundus images of a glaucoma patient in multiple stages based on medical image data, analyzing the texture and shape changes in the images, and obtaining a multi-modal image; and simulating changes of the pathological features in a plurality of development stages, and generating a dynamic pathological feature sequence. According to the method, retina changes of a glaucoma patient in multiple disease development stages are simulated, visual feature information of pathological features in multiple development stages is captured, visual features of fundus images of the patient are extracted and compared, the disease development stages of the patient are evaluated, and the feature recognition process is optimized in combination with transfer learning. The method improves the generalization ability of the model and the diagnosis accuracy, achieves the accurate evaluation of the visual impairment through combining the measurement and prediction of the intraocular pressure data, and helps medical personnel to judge the retina state of a glaucoma patient.

Description

Technical Field

[0001] The present invention relates to the field of computer vision technology, and particularly to a glaucoma multi-modal intelligent recognition method and system based on transfer learning. Background Art

[0002] The field of computer vision technology focuses on enabling computers to obtain, process, analyze, and understand visual information captured from the real world through images and multi-dimensional data, and perform specific tasks and make decisions. By using image recognition, pattern recognition, and deep learning, it can achieve various problems such as image classification, object detection, scene reconstruction, event detection, image segmentation, robot navigation, and autonomous driving, and is applied to multiple fields such as medical diagnosis, industrial inspection, driverless vehicles, security monitoring, and agricultural automation.

[0003] Among them, the glaucoma recognition method aims to identify and analyze the eye images of patients through computer vision technology. The process includes the acquisition of fundus images and the automatic analysis of the retina and optic nerve. By analyzing the visual data from fundus photography and various medical images, key indicators of glaucoma are detected using feature recognition and image processing techniques, including the shape and size of the optic disc. By identifying glaucoma patients at an early stage, it provides a basis for timely treatment, reduces vision damage caused by the disease, helps doctors diagnose efficiently, and improves the accuracy and efficiency of diagnosis.

[0004] Traditional glaucoma recognition techniques rely on single-stage image processing and feature recognition techniques, lacking sufficient flexibility and accuracy in dealing with complex lesions, unable to effectively simulate the time-series changes of pathological features in the early lesion stage, resulting in inaccurate disease assessment, unable to capture the initial tiny changes in pathological development, leading to missed treatment opportunities. When analyzing the correlation between intraocular pressure and vision damage, it lacks the ability to comprehensively analyze multi-dimensional data and cannot provide a multi-angle perspective for disease monitoring, restricting the timeliness and effectiveness of treatment strategies. Lacking continuous monitoring of the dynamic development of lesions, disease management relies on intermittent clinical assessments rather than continuous data monitoring, resulting in a lag in the judgment of the patient's condition and affecting the treatment effect and the patient's quality of life. Summary of the Invention

[0005] In order to solve the technical problem of inaccurate disease assessment existing in the prior art, an embodiment of the present invention provides a glaucoma multi-modal intelligent recognition method and system based on transfer learning. The technical solution is as follows:

[0006] On the one hand, a glaucoma multi-modal intelligent recognition method based on transfer learning is provided. The method includes:

[0007] S1: Based on medical image data, collect fundus images of glaucoma patients at multiple stages, analyze the texture and shape changes in the images, simulate the changes of pathological features at multiple development stages, and generate a dynamic pathological feature sequence;

[0008] S2: Based on the dynamic pathological feature sequence, analyze the fundus image data of the target patient. By analyzing the pixel gradient changes, identify the edge information in the image, and label multiple key information regions in the fundus image to generate fundus image annotation information;

[0009] S3: Utilize the fundus image annotation information to extract the retinal features of the target patient by analyzing the pixel and texture information in multiple regions of the image, evaluate the disease development stage of the patient, and generate feature stage evaluation information;

[0010] S4: Based on the feature stage evaluation information, measure the intraocular pressure data of the patient in real time, combine with the disease development stage of the patient, calculate the predicted values of the intraocular pressure at multiple time points, and generate an intraocular pressure trend prediction result;

[0011] S5: Based on the intraocular pressure trend prediction result, analyze the correlation between visual impairment and intraocular pressure data. By calculating the contrast difference between the damaged area and the healthy area on the image, analyze the degree of damage to multiple partitions on the retina to generate visual impairment evaluation information.

[0012] As a further solution of the present invention, the dynamic pathological feature sequence includes retinal nerve fiber layer change information, optic disc morphological variation data, and retinal vascular structure information. The fundus image annotation information includes the annotation information of the optic disc edge, the recognition result of the macula area, and the retinal vascular annotation information. The feature stage evaluation information includes the disease development stage calibration information, the nerve fiber layer thickness value, and the optic disc shape information. The intraocular pressure trend prediction result includes the real-time measurement data set, the intraocular pressure fluctuation prediction data, and the intraocular pressure change trend information. The visual impairment evaluation information includes the damaged area pixel contrast information, the visual field damaged area information, and the visual function evaluation result.

[0013] As a further solution of the present invention, the steps of collecting fundus images of glaucoma patients at multiple stages based on medical image data, analyzing the texture and shape changes in the images, and simulating the changes of pathological features at multiple development stages to generate a dynamic pathological feature sequence are specifically as follows:

[0014] S101: Based on medical image data, extract the fundus image data of glaucoma patients at multiple stages through the hospital database, sort the images according to the development stage, and generate a fundus image extraction result;

[0015] S102: Based on the extraction result of the fundus image, analyze the texture of the retinal layer and the shape change of the optic disc in the image sequence, and generate a morphological comparison image;

[0016] S103: Based on the morphological comparison image, analyze and identify the pathological features at multiple stages, calculate the numerical parameters of morphological changes, including the changes in edge sharpness and color, and generate a dynamic pathological feature sequence.

[0017] As a further solution of the present invention, the steps of analyzing the fundus image data of the target patient based on the dynamic pathological feature sequence, identifying the edge information in the image by analyzing the pixel gradient change, and annotating multiple key information regions in the fundus image to generate the fundus image annotation information are specifically as follows:

[0018] S201: Based on the dynamic pathological feature sequence, collect the fundus image data of the target patient, optimize the image quality by adjusting the brightness and contrast, and generate an adjusted patient image;

[0019] S202: Based on the adjusted patient image, calculate the pixel gradient, identify the edge information in the image, and generate an edge-enhanced image;

[0020] S203: Based on the edge-enhanced image, identify multiple key regions in the fundus image, and perform information annotation on the fundus image data of the target patient, including the optic disc and macula, to generate the fundus image annotation information.

[0021] As a further solution of the present invention, the specific formula for calculating the pixel gradient is:

[0022]

[0023] where G represents the calculated edge intensity value, and the pixel points with high edge intensity are shown as more obvious boundary lines in the image, G x represents the pixel gradient value of the image in the horizontal direction, G y represents the pixel gradient value of the image in the vertical direction, K is the weight coefficient for adjusting the contribution of the sum of gradient squares, and W is the weight coefficient for adjusting the contribution of the gradient difference.

[0024] As a further solution of the present invention, the steps of using the fundus image annotation information, extracting the retinal features of the target patient by analyzing the pixel and texture information in multiple regions of the image, and evaluating the disease development stage of the patient to generate the feature stage evaluation information are specifically as follows:

[0025] S301: Based on the fundus image annotation information, extract the pixel data and texture information of the image, analyze the pixel intensity, color depth, and texture pattern of multiple regions, calculate the differences in visual information between multiple regions, and generate texture shape parameters;

[0026] S302: Based on the texture shape parameters, by comparing the visual features of the fundus image of the target patient with the image features of multiple stages, identify the similarity features, including the texture of the nerve fiber layer and the shape of the optic disc, and generate a feature comparison analysis result;

[0027] S303: Based on the feature comparison analysis result, by analyzing the correspondence between the pathological features in the fundus image of the target patient and the known pathological stages, evaluate the disease stage of the target patient and generate feature stage evaluation information.

[0028] As a further solution of the present invention, the steps of measuring the intraocular pressure data of the patient in real time based on the feature stage evaluation information, combining with the disease development stage of the patient, calculating the predicted values of the intraocular pressure at multiple time points, and generating an intraocular pressure trend prediction result are specifically as follows:

[0029] S401: Based on the feature stage evaluation information, collect and record the intraocular pressure measurement data of the patient, and record the measurement time point information to generate an intraocular pressure data record;

[0030] S402: Based on the intraocular pressure data record, calculate the change trend of the intraocular pressure of the target patient through time series analysis to generate change trend prediction information;

[0031] S403: Based on the change trend prediction information, considering the influence of the disease development stage on the change of the intraocular pressure, calculate the predicted values of the intraocular pressure of the target patient at multiple time points to generate an intraocular pressure trend prediction result.

[0032] As a further solution of the present invention, the specific formula for calculating the predicted values of the intraocular pressure of the target patient at multiple time points is:

[0033]

[0034] Among them, P(t) represents the intraocular pressure value at the predicted time point t, t represents the current time point, P(t - i) represents the intraocular pressure data at the historical time point t - i, i is the time index, w i is the weight coefficient of the intraocular pressure data at the target time point, and n is the total number of historical time points considered in the prediction model.

[0035] As a further solution of the present invention, based on the intraocular pressure trend prediction result, analyze the correlation between vision impairment and intraocular pressure data, and by calculating the contrast difference between the damaged area and the healthy area on the image, analyze the degree of damage to multiple partitions on the retina to generate vision impairment assessment information. The specific steps are as follows:

[0036] S501: Based on the intraocular pressure trend prediction result, by analyzing the correlation between intraocular pressure data and the disease development stage, evaluate the relationship between intraocular pressure changes and vision impairment, and combine the patient's intraocular pressure data to predict the patient's disease development stage and generate correlation calculation data;

[0037] S502: Based on the correlation calculation data, extract the pixel data of the damaged area and the healthy area through the patient's fundus image data, analyze the difference between the contrast of the damaged area and the contrast of the surrounding healthy area, and generate retinal contrast difference data;

[0038] S503: Use the retinal contrast difference data to analyze the degree of vision impairment in multiple areas on the retina through the contrast difference and generate vision impairment assessment information.

[0039] On the other hand, a glaucoma multi-modal intelligent recognition system based on transfer learning is provided. This system is applied to the glaucoma multi-modal intelligent recognition method based on transfer learning. The system includes:

[0040] The image data extraction module, based on medical image data, extracts the fundus images of glaucoma patients at multiple stages through a medical database, preprocesses the images, adjusts the image contrast and brightness, and generates a processed image data set;

[0041] The image texture analysis module, based on the processed image data set, analyzes the texture and shape of the images in the processed image data set, simulates the changes in pathological features at multiple development stages, identifies the visual features of the fundus image data of glaucoma patients at multiple stages, and generates a dynamic pathological feature sequence;

[0042] The edge information recognition module, based on the dynamic pathological feature sequence, analyzes the fundus image data of the target patient, extracts the edge information in the image and marks multiple key areas, and generates fundus image annotation information;

[0043] The visual feature extraction module, based on the fundus image annotation information, analyzes the pixel and texture information of multiple areas in the target image, extracts the visual feature information of the retina, evaluates the patient's disease stage, and generates feature stage evaluation information;

[0044] Based on the feature stage evaluation information, the intraocular pressure prediction module measures the intraocular pressure data of the patient, takes into account the development stage of the patient's condition, calculates the predicted values of the intraocular pressure at multiple time points, and generates the intraocular pressure trend prediction result;

[0045] The visual impairment analysis module uses the intraocular pressure trend prediction result to analyze the correlation between visual impairment and intraocular pressure data, evaluates the degree of visual impairment of the target patient by calculating the contrast difference between the visually impaired area and the healthy area, and generates the visual impairment evaluation information.

[0046] The beneficial effects brought by the technical solution provided by the embodiments of the present invention at least include:

[0047] By simulating the retinal changes of glaucoma patients at multiple stages of disease development, capturing the visual feature information of pathological features at multiple development stages, extracting and comparing the visual features of the patient's fundus images, evaluating the stage of the patient's disease development, optimizing the feature recognition process by combining transfer learning, improving the generalization ability of the model and the accuracy of diagnosis, combining the measurement and prediction of intraocular pressure data, achieving an accurate assessment of visual impairment, and helping medical staff to judge the retinal status of glaucoma patients. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0049] Figure 1 It is a schematic diagram of the working process of the present invention;

[0050] Figure 2 It is a detailed flowchart of S1 of the present invention;

[0051] Figure 3 It is a detailed flowchart of S2 of the present invention;

[0052] Figure 4 It is a detailed flowchart of S3 of the present invention;

[0053] Figure 5 It is a detailed flowchart of S4 of the present invention;

[0054] Figure 6 It is a detailed flowchart of S5 of the present invention;

[0055] Figure 7 It is a system flowchart of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0056] The technical solutions in the present invention will be described below with reference to the accompanying drawings.

[0057] In the embodiments of the present invention, words such as "exemplarily" and "for example" are used to represent examples, illustrations or explanations. Any embodiment or design solution described as an "example" in the present invention should not be construed as being more preferred or having more advantages than other embodiments or design solutions. Exactly speaking, the use of the word "example" is intended to present concepts in a specific manner. In addition, in the embodiments of the present invention, the meaning expressed by "and / or" can be both, or either of the two can be selected.

[0058] In the embodiments of the present invention, "image" and "picture" can sometimes be used interchangeably. It should be noted that when the difference is not emphasized, the meaning they express is the same. "(of)", "corresponding", and "corresponding" can sometimes be used interchangeably. It should be noted that when the difference is not emphasized, the meaning they express is the same.

[0059] In the embodiments of the present invention, sometimes subscripts such as W 1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0060] To make the technical problems to be solved, technical solutions and advantages of the present invention clearer, the following will be described in detail with reference to the accompanying drawings and specific embodiments.

[0061] The embodiments of the present invention provide a glaucoma multi-modal intelligent recognition method based on transfer learning. As Figure 1 shown in the flowchart of the glaucoma multi-modal intelligent recognition method based on transfer learning, the processing flow of this method can include the following steps:

[0062] S1: Based on medical image data, collect fundus images of glaucoma patients at multiple stages, analyze the texture and shape changes in the images, simulate the changes of pathological features at multiple development stages, and generate a dynamic pathological feature sequence;

[0063] S2: Based on the dynamic pathological feature sequence, analyze the fundus image data of the target patient, identify the edge information in the image by analyzing the pixel gradient change, and label multiple key information regions in the fundus image to generate fundus image annotation information;

[0064] S3: Utilize the fundus image annotation information, extract the retinal features of the target patient by analyzing the pixel and texture information in multiple regions of the image, evaluate the disease development stage of the patient, and generate feature stage evaluation information;

[0065] S4: Based on the feature stage evaluation information, measure the intraocular pressure data of the patient in real time. Combine with the disease development stage of the patient, calculate the predicted values of the intraocular pressure at multiple time points, and generate the intraocular pressure trend prediction result;

[0066] S5: Based on the intraocular pressure trend prediction result, analyze the correlation between visual impairment and intraocular pressure data. By calculating the contrast difference between the damaged area and the healthy area in the image, analyze the degree of damage to multiple partitions on the retina, and generate the visual impairment evaluation information.

[0067] The dynamic pathological feature sequence includes the change information of the retinal nerve fiber layer, the morphological variation data of the optic disc, and the retinal vascular structure information. The fundus image annotation information includes the annotation information of the optic disc edge, the recognition result of the macula area, and the retinal vascular annotation information. The feature stage evaluation information includes the disease development stage calibration information, the nerve fiber layer thickness value, and the optic disc shape information. The intraocular pressure trend prediction result includes the real-time measurement data set, the intraocular pressure fluctuation prediction data, and the intraocular pressure change trend information. The visual impairment evaluation information includes the damaged area pixel contrast information, the visual field damaged area information, and the visual function evaluation result.

[0068] Please refer to Figure 2 , based on the medical image data, collect the fundus images of glaucoma patients at multiple stages, analyze the texture and shape changes in the images, simulate the changes of pathological features at multiple development stages, and the steps to generate the dynamic pathological feature sequence are specifically as follows:

[0069] S101: Based on the medical image data, through the hospital database, extract the fundus image data of glaucoma patients at multiple stages, sort the images according to the development stage, and generate the fundus image extraction result;

[0070] In the sub-step of S101, based on the hospital database, collect and classify the fundus images of glaucoma patients. Access the diagnosis records of patients through the medical data management system, extract the fundus image data related to glaucoma. The data includes image files in JPEG or DICOM format. Use image processing software to perform preliminary cleaning and format standardization on the images, remove the noise in the images and adjust the contrast and brightness to make the image quality meet the analysis requirements. According to the pathological development stage mentioned in the patient's medical record, use Python programming and image processing libraries to sort the images according to the pathological stage. The sorting algorithm is based on the time stamp in the file metadata of the image and the doctor's diagnosis record to ensure that the images of each stage can be correctly classified. The classified images will be saved in the database folder for subsequent steps to call.

[0071] S102: Based on the fundus image extraction result, analyze the texture of the retinal layer and the shape change of the optic disc in the image sequence, and generate the morphological comparison image;

[0072] In sub-step S102, based on the sorted fundus image data, image analysis techniques are used to analyze the texture of the retinal layers and the shape changes of the optic disc. During the process, image segmentation techniques are adopted. Using a convolutional neural network model based on deep learning, key structures in the image, such as the boundary of the optic disc and the texture of the retinal layers, are identified and tracked. The network is constructed using the TensorFlow and Keras frameworks. By training the model to identify the key features of the retina and the optic disc, the historical fundus image dataset is used for model training, and the model parameters are optimized through cross-validation. After training, the model will be applied to the image data to extract the texture and shape features of each image, compare the differences between consecutive stage images, and generate a morphological comparison map. The target image will highlight the pathological change areas for subsequent medical evaluation and diagnosis.

[0073] S103: Based on the morphological comparison images, analyze and identify the pathological features of multiple stages, calculate the numerical parameters of morphological changes, including the changes in edge sharpness and color, and generate a dynamic pathological feature sequence;

[0074] In sub-step S103, based on the generated morphological comparison images, image feature extraction and machine learning classification algorithms are used. This includes using an edge detection algorithm to extract the edge sharpness information in the image and adopting color analysis techniques to calculate the color changes in the image. The target techniques are all implemented in the MATLAB environment, using its image processing toolbox. The target numerical features are input into a support vector machine classifier to analyze the change trends and classifications of the image features at each stage. The training of the classifier is based on the image dataset of known stages. The generated dynamic pathological feature sequence depicts the pathological changes from the early stage to the late stage, providing an accurate disease progression monitoring tool for clinical practice.

[0075] Please refer to Figure 3 , based on the dynamic pathological feature sequence, analyze the fundus image data of the target patient. By analyzing the pixel gradient changes, identify the edge information in the image, and label multiple key information areas in the fundus image. The specific steps for generating the fundus image annotation information are as follows:

[0076] S201: Based on the dynamic pathological feature sequence, collect the fundus image data of the target patient, optimize the image quality by adjusting the brightness and contrast, and generate the adjusted patient image;

[0077] In sub-step S201, fundus image data of the target patient is collected through an automated image processing system, connected to a medical image storage and communication system, and the latest fundus image of the target patient is automatically downloaded. The original image data collected often has problems such as uneven brightness or low contrast due to unsatisfactory shooting conditions. Using Adobe Photoshop software, the automatic color adjustment function is used to adjust the brightness and contrast of the image. By setting filters and adjustment layers, the visual effect of the image is enhanced, making details clearly visible. The non-linear editing curve is applied to carefully adjust the bright and dark parts of the image to ensure that important medical features in the image, including blood vessels and optic nerve discs, are clearly visible. The adjusted image is saved in a medical image format for subsequent medical analysis and diagnosis.

[0078] S202: Based on the adjusted patient image, calculate the pixel gradient, identify the edge information in the image, and generate an edge-enhanced image;

[0079] The specific formula for calculating the pixel gradient is:

[0080]

[0081] Among them, G represents the calculated edge intensity value. Pixel points with high edge intensity appear as more obvious boundary lines in the image. G x represents the pixel gradient value of the image in the horizontal direction, and G y represents the pixel gradient value of the image in the vertical direction. K is a weight coefficient for adjusting the contribution of the sum of squared gradients, and W is a weight coefficient for adjusting the contribution of the gradient difference.

[0082] Formula:

[0083]

[0084] Detailed explanation of the formula and the derivation process of the formula calculation:

[0085] The formula is used to calculate the edge intensity of each pixel point in the image and extract the edge information in the image;

[0086] Meaning of parameters and set values:

[0087] G x represents the gradient of the image in the horizontal direction. Assuming the horizontal gradient at the target point is 120, it reflects the color change rate of this point in the horizontal direction;

[0088] G y represents the gradient of the image in the vertical direction. Assuming the vertical gradient at the target point is 80, it indicates the color change rate of this point in the vertical direction;

[0089] K is an adjustment coefficient for the sum of squared gradients, assumed to be 0.5, which is used to balance the overall brightness and contrast of the image;

[0090] W is an adjustment coefficient for the gradient difference, assumed to be 0.3, which is used to enhance the detailed performance of the edges;

[0091] Substitute the parameters into the formula for calculation:

[0092]

[0093]

[0094] The result 72.2 indicates that at the given pixel point, the intensity of the edge is 72.2, indicating that the edge information at this point is obvious. The result is used to extract the edge information in the image and identify multiple key regions.

[0095] S203: Based on the edge-enhanced image, identify multiple key regions in the fundus image, and perform information annotation on the fundus image data of the target patient, including the optic disc and macula, to generate fundus image annotation information;

[0096] In the sub-step of S203, based on the edge-enhanced image, use image analysis software to identify and annotate key regions. Manually identify key regions such as the optic disc and macula in the fundus image by using the image annotation tool in Adobe Illustrator software. During the annotation process, doctors or technicians will accurately outline the boundaries of the target regions according to the edge-enhanced image, use different color markers for differentiation, and add annotations to each marker, such as "optic disc", "macular area", to ensure that the target important regions can be clearly identified in the subsequent medical reports. Use the template matching method in image processing technology to automatically identify and mark common structures, such as blood vessel directions. After the target is automatically marked, it will be confirmed and adjusted by doctors. The annotation information will be saved in the metadata of the image and stored in XML format for further data analysis and research.

[0097] Please refer to Figure 4 , using the fundus image annotation information, by analyzing the pixel and texture information of multiple regions in the image, extracting the retinal features of the target patient, and evaluating the disease development stage of the patient. The specific steps for generating the feature stage evaluation information are as follows:

[0098] S301: Based on the fundus image annotation information, extract the pixel data and texture information of the image, analyze the pixel intensity, color depth, and texture pattern of multiple regions, calculate the difference in visual information between multiple regions, and generate texture shape parameters;

[0099] In sub-step S301, image pixel data and texture information are extracted using the image processing software MATLAB. Advanced image processing functions are applied to the labeled fundus images to extract the pixel intensity and color depth of key regions such as the optic disc and macula. Techniques used include grayscale conversion and color space analysis. The target technique helps accurately determine the shades of different colors in the image. Texture analysis algorithms, such as the gray-level co-occurrence matrix, are applied to analyze the roughness and directionality of the image texture. The target texture features reflect the microscopic changes in the retinal health status. By calculating and comparing the texture differences between different regions, texture shape parameters are generated. The target parameters reveal the characteristics of pathological changes, such as the degree of nerve fiber layer damage or retinal thinning.

[0100] S302: Based on the texture shape parameters, by comparing the visual features of the target patient's fundus image with the image features at multiple stages, similarity features are identified, including the texture of the nerve fiber layer and the shape of the optic disc, and a feature comparison analysis result is generated.

[0101] In sub-step S302, based on the generated texture shape parameters, an image comparison tool is used for feature analysis. The texture shape parameter data is processed and standardized through data analysis platforms such as the Pandas and NumPy libraries in Python. Image comparison algorithms, such as the Scale-Invariant Feature Transform (SIFT) in feature matching technology, are used to compare the target patient's fundus image with the archived multi-stage case images. By identifying the feature points in the images and comparing their geometric and texture similarities, the texture changes of the optic disc and nerve fiber layer are evaluated. The generated feature comparison analysis result provides a similarity score between the patient's lesion stage and other cases, helping doctors judge the disease progression and changes.

[0102] S303: Based on the feature comparison analysis result, by analyzing the correspondence between the pathological features in the target patient's fundus image and the known pathological stages, the disease stage of the target patient is evaluated, and feature stage evaluation information is generated.

[0103] In sub-step S303, by using the feature comparison analysis result, the disease stage of the target patient is evaluated. During the process, classification algorithms in machine learning techniques, such as support vector machines, are used to match the result of the feature comparison analysis with the features of the known pathological stages in the database. The classification algorithm is based on feature learning and accurately assigns the patient's image to the corresponding pathological stage, obtaining a result including a stage score. The target score reflects the similarity between the target patient and each pathological stage, generating feature stage evaluation information. This information is crucial for formulating personalized treatment plans and monitoring the disease progression, providing a scientific basis for doctors to make judgments.

[0104] Please refer to Figure 5, based on the characteristic stage evaluation information, the steps of measuring the intraocular pressure data of the patient in real time, combining with the disease development stage of the patient, calculating the predicted values of the intraocular pressure at multiple time points, and generating the intraocular pressure trend prediction result are specifically as follows:

[0105] S401: Based on the characteristic stage evaluation information, collect and record the intraocular pressure measurement data of the patient, and record the measurement time point information to generate the intraocular pressure data record;

[0106] In the sub-step of S401, a non-contact tonometer dedicated to ophthalmology is used to measure the intraocular pressure of the patient to ensure accurate intraocular pressure values. The intraocular pressure is measured by the slight pulse reflection of gas on the cornea to reduce the discomfort of the patient. After each measurement, the intraocular pressure data and measurement time of the patient are automatically recorded through the connected medical information system. The data recording work is completed by the medical data management software, and all diagnostic information of the patient is updated and saved in real time. After collection, the generated intraocular pressure data record is stored in the form of a spreadsheet. Each data includes the date, time, and corresponding intraocular pressure value. This data recording method facilitates the subsequent analysis of the historical changes and potential trends of the intraocular pressure and provides a convenient monitoring tool for doctors.

[0107] S402: Based on the intraocular pressure data record, calculate the change trend of the intraocular pressure of the target patient through time series analysis to generate the change trend prediction information;

[0108] In the sub-step of S402, time series analysis technology is applied to analyze the change trend of the patient's intraocular pressure. The statistical software R is used for data processing. The intraocular pressure data is sorted into a time series format. The future change of the intraocular pressure is predicted through the autoregressive moving average model. Combining the characteristics of autoregression and moving average, the trend and periodic fluctuation of the time series data are simulated and predicted. By adjusting the parameters of the model, such as the number of autoregressive terms and moving average terms, the model can fit the actual data more accurately. The generated change trend prediction information indicates the possible maximum and minimum values of the intraocular pressure at future time points, providing a scientific basis for doctors to formulate treatment plans.

[0109] S403: Based on the change trend prediction information, considering the influence of the disease development stage on the change of the intraocular pressure, calculate the predicted values of the intraocular pressure of the target patient at multiple time points to generate the intraocular pressure trend prediction result;

[0110] The specific formula for calculating the predicted values of the intraocular pressure of the target patient at multiple time points is:

[0111]

[0112] Among them, P(t) represents the intraocular pressure value at the predicted time point t, t represents the current time point, P(t - i) represents the intraocular pressure data at the time point t - i in history, i is the time index, wi is the weight coefficient of the intraocular pressure data at the target time point, and n is the total number of historical time points considered in the prediction model.

[0113] Formula:

[0114]

[0115] Detailed explanation of the formula and the derivation process of the formula calculation:

[0116] The formula is used to calculate the predicted intraocular pressure at the target time point, and the predicted value is used to combine the fundus image analysis results to evaluate the visual acuity status of the target patient;

[0117] Meaning and setting values of parameters:

[0118] i: Time index, used to iterate over past data points, assuming n = 4;

[0119] P(t - i): Intraocular pressure data at the historical time point t - i, assumed to be 110 mmHg, 120 mmHg, 130 mmHg, and 125 mmHg respectively;

[0120] w i : Weight coefficient, assumed to be 0.1, 0.2, 0.3, and 0.4 respectively;

[0121] Substitute the parameters into the formula for calculation:

[0122]

[0123] P(t) = 124;

[0124] The result 124 indicates that the predicted value of the intraocular pressure at the target time point is 124 mmHg. The calculation process is used to evaluate the changing trend of the patient's intraocular pressure and calculate the predicted values of the intraocular pressure at multiple time points.

[0125] Please refer to Figure 6 , based on the intraocular pressure trend prediction results, analyze the correlation between visual impairment and intraocular pressure data. By calculating the contrast difference between the damaged area and the healthy area on the image, the steps for analyzing the degree of damage to multiple partitions on the retina and generating visual impairment assessment information are as follows:

[0126] S501: Based on the intraocular pressure trend prediction results, by analyzing the correlation between intraocular pressure data and the disease development stage, evaluate the relationship between intraocular pressure changes and visual impairment, and combine the patient's intraocular pressure data to predict the disease development stage of the patient and generate correlation calculation data;

[0127] In sub-step S501, a statistical analysis method is used to analyze the relationship between intraocular pressure changes and vision impairment. A Pearson correlation coefficient calculation tool is employed to evaluate the correlation between intraocular pressure data and the disease development stage. The historical intraocular pressure data and vision examination records of patients are sorted and analyzed through the statistical software SPSS. The generated dataset includes intraocular pressure values, vision levels, and examination dates. Using a multiple regression analysis model, with intraocular pressure as the independent variable and the degree of vision impairment as the dependent variable, a regression analysis is conducted to evaluate the quantitative relationship between intraocular pressure changes and vision impact. Through this method, the correlation between intraocular pressure changes and vision decline can be quantified, providing a quantitative basis for subsequent treatment, and generating correlation calculation data, recording the data processing and result interpretation at each step.

[0128] S502: Based on the correlation calculation data, through the patient's fundus image data, extract the pixel data of the damaged area and the healthy area, analyze the difference in contrast between the damaged area and the surrounding healthy area, and generate retinal contrast difference data;

[0129] In sub-step S502, by analyzing the pixel data of the damaged area and the healthy area, the retinal contrast difference is analyzed using the contrast calculation formula. The formula is Calculate the contrast difference of each area to generate retinal contrast difference data;

[0130] In the formula, C represents the average contrast difference, N represents the total number of pixel points analyzed, I D (i) represents the brightness value of the i-th pixel point in the damaged area, I H (i) represents the brightness value of the i-th pixel point in the healthy area, and i is the index of the pixel point;

[0131] Detailed explanation of the formula and the formula calculation derivation process:

[0132] Assume that in the target image area, the brightness value of the damaged area is I D =[120, 125, 130, 135, 140, 145, 150, 155, 160, 165], and the brightness value of the healthy area is I H =[80, 85, 90, 95, 100, 105, 110, 115, 120, 125]. Calculate the contrast difference C:

[0133]

[0134] The result 40 indicates that the average contrast difference between the damaged area and the healthy area is 40, and the result is used to help doctors judge the degree of vision impairment.

[0135] S503: Use the retinal contrast difference data, analyze the visual impairment degree of multiple regions on the retina through the contrast difference, and generate visual impairment assessment information;

[0136] In the sub-step of S503, use the retinal contrast difference data to evaluate the visual impairment degree of multiple regions, apply image processing technology and visual analysis algorithms to analyze the contrast difference on the retina, focus on the visual impairment of the damaged retinal area relative to the healthy area, use the image processing toolbox in MATLAB to perform edge detection and contrast enhancement to improve the visualization effect of the damaged area, calculate the contrast value of each region through quantitative analysis techniques such as regional contrast analysis method, and compare the target data with the clinical visual impairment criteria to generate visual impairment assessment information, providing a basis for diagnosing different stages of retinal diseases and formulating personalized treatment plans to ensure the accuracy and practicality of the analysis results.

[0137] Please refer to Figure 7 , a glaucoma multi-modal intelligent recognition system based on transfer learning. The glaucoma multi-modal intelligent recognition system based on transfer learning is used to execute the above-mentioned glaucoma multi-modal intelligent recognition method based on transfer learning. The system includes:

[0138] The image data extraction module extracts fundus images of glaucoma patients at multiple stages based on medical image data through a medical database, preprocesses the images, adjusts the image contrast and brightness, and generates a processed image data set;

[0139] The image texture analysis module analyzes the texture and shape of the images in the processed image data set based on the processed image data set, simulates the changes of pathological features at multiple development stages, identifies the visual features of the fundus image data of glaucoma patients at multiple stages, and generates a dynamic pathological feature sequence;

[0140] The edge information recognition module analyzes the fundus image data of the target patient based on the dynamic pathological feature sequence, extracts the edge information in the image and labels multiple key regions, and generates fundus image annotation information;

[0141] The visual feature extraction module analyzes the pixel and texture information of multiple regions in the target image based on the fundus image annotation information, extracts the visual feature information of the retina, evaluates the disease stage of the patient, and generates feature stage assessment information;

[0142] The intraocular pressure prediction module measures the intraocular pressure data of the patient based on the feature stage assessment information, considers the development stage of the patient's condition, calculates the predicted values of the intraocular pressure at multiple time points, and generates an intraocular pressure trend prediction result;

[0143] The visual impairment analysis module uses the prediction results of the intraocular pressure trend to analyze the correlation between visual impairment and intraocular pressure data. By calculating the contrast difference between the visually impaired area and the healthy area, it evaluates the degree of visual impairment of the target patient and generates visual impairment assessment information.

[0144] The above embodiments can be implemented in whole or in part by software, hardware (such as circuits), firmware, or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions described in the embodiments of the present invention are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (such as infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that contains one or more collections of available media. The available media can be magnetic media (such as floppy disks, hard disks, magnetic tapes), optical media (such as DVDs), or semiconductor media. The semiconductor media can be a solid-state drive.

[0145] It should be understood that the term "and / or" in this article is merely a description of the association relationship between associated objects, indicating that three relationships can exist. For example, A and / or B can represent: A exists alone, A and B exist simultaneously, and B exists alone. Here, A and B can be singular or plural. In addition, the character " / " in this article generally represents an "or" relationship between the associated objects before and after, but it may also represent an "and / or" relationship, which can be specifically understood with reference to the context before and after.

[0146] In the present invention, "at least one" means one or more, and "a plurality" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of single items or plural items. For example, at least one of a, b, or c can represent: a, b, c, a - b, a - c, b - c, or a - b - c, where a, b, and c can be single or multiple.

[0147] It should be understood that in various embodiments of the present invention, the sequence numbers of the above processes do not imply the order of execution. The order of execution of each process should be determined by its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present invention.

[0148] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.

[0149] Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working processes of the devices, apparatuses, and units described above can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0150] In several embodiments provided by the present invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling, direct coupling, or communication connection to each other can be through some interfaces. The indirect coupling or communication connection of the devices or units can be in an electrical, mechanical, or other form.

[0151] The units described as separate components may or may not be physically separated. The components displayed as units may or may not be physical units, that is, they can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0152] In addition, the functional units in various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit.

[0153] When the above-mentioned functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The foregoing storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs.

[0154] As described above, the above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or substitutions, which should all be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. A multimodal intelligent recognition method for glaucoma based on transfer learning, characterized in that: The method comprises: Based on medical image data, fundus images of glaucoma patients at multiple stages are collected, texture and shape changes in the images are analyzed, changes in pathological features at multiple development stages are simulated, and dynamic pathological feature sequences are generated; Based on the dynamic pathological feature sequence, the fundus image data of the target patient is analyzed, edge information in the image is identified by analyzing pixel gradient changes, and multiple key information areas in the fundus image are annotated to generate fundus image annotation information; Utilizing the fundus image annotation information, by analyzing pixel and texture information of multiple regions in the image, extracting retinal features of the target patient, assessing the stage of disease progression of the patient, and generating feature stage assessment information; Based on the characteristic stage evaluation information, the intraocular pressure data of the patient is measured in real time, and the predicted values ​​of the intraocular pressure at multiple time points are calculated in combination with the patient's disease progression stage to generate an intraocular pressure trend prediction result; Based on the intraocular pressure trend prediction results, the correlation between visual impairment and intraocular pressure data is analyzed, and the degree of damage to multiple partitions on the retina is analyzed by calculating the contrast difference between the damaged area and the healthy area in the image to generate visual impairment assessment information.

2. The multimodal intelligent recognition method for glaucoma based on transfer learning according to claim 1, characterized in that: The dynamic pathological feature sequence includes retinal nerve fiber layer change information, optic disc morphological variation data, and retinal vascular structure information; the fundus image annotation information includes annotation information of the optic disc edge, recognition results of the macular area, and retinal vascular annotation information; the characteristic stage evaluation information includes disease progression stage calibration information, nerve fiber layer thickness values, and optic disc shape information; the intraocular pressure trend prediction results include real-time measurement data sets, intraocular pressure fluctuation prediction data, and intraocular pressure change trend information; the vision impairment assessment information includes pixel contrast information of the damaged area, visual field damage area information, and vision function assessment results.

3. The multimodal intelligent recognition method for glaucoma based on transfer learning according to claim 1, characterized in that: Based on medical image data, fundus images of glaucoma patients at multiple stages are collected, texture and shape changes in the images are analyzed, and changes in pathological features at multiple development stages are simulated. The steps to generate dynamic pathological feature sequences are as follows: Based on medical image data, fundus image data of glaucoma patients at multiple stages are extracted through hospital databases, and images are sorted according to the development stage to generate fundus image extraction results; Based on the fundus image extraction result, analyzing the texture of the retinal layer and the shape change of the optic disc in the image sequence to generate a morphological contrast image; Based on the morphological contrast images, the pathological features of multiple stages are analyzed and identified, and the numerical parameters of the morphological changes, including the changes in edge sharpness and color, are calculated to generate a dynamic pathological feature sequence.

4. The multimodal intelligent recognition method for glaucoma based on transfer learning according to claim 1, characterized in that: Based on the dynamic pathological feature sequence, the fundus image data of the target patient is analyzed, the edge information in the image is identified by analyzing the pixel gradient change, and multiple key information areas in the fundus image are annotated. The steps of generating fundus image annotation information are specifically as follows: Based on the dynamic pathological feature sequence, fundus image data of the target patient is collected, image quality is optimized by adjusting brightness and contrast, and an adjusted patient image is generated; Based on the adjusted patient image, calculating pixel gradients, identifying edge information in the image, and generating an edge-enhanced image; Based on the edge-enhanced image, multiple key areas in the fundus image are identified, and information annotation is performed on the fundus image data of the target patient, including the optic disc and the macula, to generate fundus image annotation information.

5. The multimodal intelligent recognition method for glaucoma based on transfer learning according to claim 4 is characterized in that: The specific formula for calculating pixel gradient is: Among them, G represents the calculated edge strength value. Pixels with high edge strength appear as more obvious boundary lines in the image. x Represents the pixel gradient value of the image in the horizontal direction, G y It represents the pixel gradient value of the image in the vertical direction, K is the weight coefficient for adjusting the contribution of the gradient sum of squares, and W is the weight coefficient for adjusting the contribution of the gradient difference.

6. The multimodal intelligent recognition method for glaucoma based on transfer learning according to claim 1, characterized in that: The steps of extracting the retinal features of the target patient and evaluating the stage of disease progression of the patient by analyzing the pixel and texture information of multiple regions in the image using the fundus image annotation information are as follows: Based on the fundus image annotation information, extract pixel data and texture information of the image, analyze pixel intensity, color depth and texture pattern of multiple regions, calculate the difference of visual information between multiple regions, and generate texture shape parameters; Based on the texture shape parameters, by comparing the visual features of the fundus image of the target patient with the image features of multiple stages, similar features are identified, including the texture of the nerve fiber layer and the shape of the optic disc, and feature comparison analysis results are generated; Based on the feature comparison and analysis results, the target patient's disease stage is evaluated by analyzing the correspondence between the pathological features in the target patient's fundus image and the known pathological stages, and feature stage evaluation information is generated.

7. The multimodal intelligent recognition method for glaucoma based on transfer learning according to claim 1, characterized in that: Based on the characteristic stage evaluation information, the intraocular pressure data of the patient is measured in real time, and the predicted values ​​of the intraocular pressure at multiple time points are calculated in combination with the patient's disease progression stage. The steps of generating the intraocular pressure trend prediction result are specifically as follows: Based on the characteristic stage evaluation information, collecting and recording the patient's intraocular pressure measurement data, and recording the measurement time point information, to generate an intraocular pressure data record; Based on the intraocular pressure data record, calculating the intraocular pressure change trend of the target patient through time series analysis, and generating change trend prediction information; Based on the change trend prediction information, considering the impact of the disease progression stage on the change in intraocular pressure, the intraocular pressure prediction values ​​of the target patient at multiple time points are calculated to generate an intraocular pressure trend prediction result.

8. The multimodal intelligent identification method for glaucoma based on transfer learning according to claim 7, characterized in that: The specific formula for calculating the predicted intraocular pressure values ​​of the target patient at multiple time points is: Where P(t) represents the intraocular pressure value at the predicted time point t, t represents the current time point, P(ti) represents the intraocular pressure data at the historical time point ti, i is the time index, and w i is the weight coefficient for the IOP data at the target time point, and n is the total number of historical time points considered in the prediction model.

9. The multimodal intelligent recognition method for glaucoma based on transfer learning according to claim 1, characterized in that: Based on the intraocular pressure trend prediction result, the correlation between visual impairment and intraocular pressure data is analyzed, and the degree of damage to multiple partitions on the retina is analyzed by calculating the contrast difference between the damaged area and the healthy area on the image. The steps of generating visual impairment assessment information are specifically as follows: Based on the intraocular pressure trend prediction result, by analyzing the correlation between the intraocular pressure data and the stage of disease progression, the relationship between intraocular pressure changes and visual impairment is evaluated, and the stage of disease progression of the patient is predicted in combination with the patient's intraocular pressure data, generating correlation calculation data; Based on the correlation calculation data, pixel data of the damaged area and the healthy area are extracted through the patient's fundus image data, and the difference between the contrast of the damaged area and the contrast of the surrounding healthy area is analyzed to generate retinal contrast difference data; The retinal contrast difference data is used to analyze the degree of vision impairment in multiple areas on the retina through contrast differences, and vision impairment assessment information is generated.

10. A multimodal intelligent recognition system for glaucoma based on transfer learning, characterized in that: According to the multimodal intelligent recognition method for glaucoma based on transfer learning according to any one of claims 1 to 9, the system comprises: The image data extraction module extracts fundus images of glaucoma patients at multiple stages based on medical image data through a medical database, preprocesses the images, adjusts the image contrast and brightness, and generates a processed image data set; The image texture analysis module performs texture and shape analysis on the images in the processed image data set based on the processed image data set, simulates the changes of pathological features in multiple development stages, identifies the visual features of fundus image data of glaucoma patients in multiple stages, and generates a dynamic pathological feature sequence; The edge information recognition module analyzes the fundus image data of the target patient based on the dynamic pathological feature sequence, extracts the edge information in the image, annotates multiple key areas, and generates fundus image annotation information; The visual feature extraction module analyzes the pixel and texture information of multiple areas in the target image based on the fundus image annotation information, extracts the visual feature information of the retina, evaluates the patient's disease stage, and generates feature stage evaluation information; The intraocular pressure prediction module measures the intraocular pressure data of the patient based on the characteristic stage evaluation information, calculates the predicted values ​​of the intraocular pressure at multiple time points considering the development stage of the patient's condition, and generates an intraocular pressure trend prediction result; The vision impairment analysis module uses the intraocular pressure trend prediction results to analyze the correlation between vision impairment and intraocular pressure data, evaluates the degree of vision impairment of the target patient by calculating the contrast difference between the vision impairment area and the healthy area, and generates vision impairment assessment information.

Images