Ophthalmic disease analysis method based on deep learning

Abstract

The invention discloses an ophthalmic disease analysis method based on deep learning, and belongs to the technical field of intelligent analysis. In order to solve the problems that data diversity and standardization are insufficient and diagnosis accuracy is limited, through diversified and standardized data collection and processing, robustness and accuracy of a model are enhanced, a complex ophthalmic disease diagnosis scene can be coped with, cloud deployment improves diagnosis efficiency and convenience, and the diagnosis efficiency is improved. The real-time feedback mechanism provides data support for continuous improvement of the model, so that the model can more accurately adapt to clinical requirements, feature extraction and classification are automatically performed on the ophthalmology images through the deep learning model, the automation degree of diagnosis is greatly improved, the deep learning model has strong feature learning ability, and the diagnosis accuracy is improved. According to the method, fine features of the ophthalmic diseases can be learned, so that more accurate diagnosis is realized, and the deep learning model has stronger generalization ability through a large amount of training data and an optimization algorithm and can adapt to different types of ophthalmic disease images.

Description

Technical Field

[0001] This invention relates to the field of intelligent analysis technology, and in particular to a deep learning-based method for analyzing ophthalmic diseases. Background Technology

[0002] There are many types of eye diseases, involving various conditions of the eyeball and related tissues. Common eye diseases include myopia, dry eye syndrome, conjunctivitis, keratitis, and glaucoma. These diseases can be caused by a variety of factors, such as prolonged use of the eyes, genetic factors, increased intraocular pressure, or infection. Symptoms of eye diseases are diverse and may include decreased vision, eye pain, red eyes, tearing, and increased discharge. If these symptoms occur, you should seek medical attention promptly for diagnosis and treatment by a professional doctor.

[0003] When diagnosing eye diseases, the main method is to compare and analyze photos of the patient's eyes with photos of their past eye conditions to determine the type of eye disease.

[0004] However, the following problems still exist in actual operation:

[0005] Traditional methods for diagnosing ophthalmic diseases often rely on limited data sources and inadequate data processing methods. This results in poor performance of trained models when dealing with complex clinical scenarios and image data. In addition, existing methods for diagnosing ophthalmic diseases may be affected by various factors such as diagnostic tools and physician experience, leading to certain errors in diagnostic results. Summary of the Invention

[0006] The purpose of this invention is to provide a deep learning-based method for analyzing ophthalmic diseases, in order to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a deep learning-based method for analyzing ophthalmic diseases, comprising the following steps:

[0008] Data acquisition and processing involves collecting data from electronic medical records, public datasets, and hospital records, as well as collecting ophthalmic disease image data, including fundus images and OCT images. The ophthalmic disease image data is cleaned and enhanced, and preprocessed by cropping, scaling, and normalization.

[0009] Data labeling and segmentation: Identify the types of ophthalmic diseases in ophthalmic disease image data; perform data structuring processing on the target ophthalmic disease data according to the ophthalmic disease type; label each image with a corresponding ophthalmic disease label; the labeling information of the ophthalmic disease label includes the type, severity, and location information of the disease; and segment the labeled ophthalmic disease image data.

[0010] The learning model is built by constructing a deep learning model containing a convolutional neural network. The convolutional neural network is used to extract feature information from ophthalmic disease image data. The preprocessed ophthalmic disease image data is input into the deep learning model, and the key feature information of the ophthalmic disease image is extracted through the convolutional neural network.

[0011] The training process involves learning the model, constructing a loss function to guide the training of the deep learning model, using a labeled and partitioned training set and key feature information to train the deep learning model, optimizing the model parameters during training, monitoring the model performance through the validation set, adjusting the model parameters and hyperparameters based on the performance on the validation set, and saving the model parameters when the model performs well on the validation set.

[0012] The learning model is evaluated by using a test set to measure its performance after training and comparing it with the diagnostic results of human experts. Based on the evaluation results, the model is iterated and optimized to obtain a successfully trained eye disease analysis model.

[0013] Application analysis and optimization involves deploying the trained eye disease analysis model to the cloud and local servers. Doctors can use the eye disease analysis model to assist in the diagnosis and analysis of patients through web applications and mobile applications. Feedback from doctors after using the model is collected to continuously improve and optimize the model.

[0014] Furthermore, the data acquisition and processing specifically includes the following steps:

[0015] Electronic medical records and hospital record collection: Extracting diagnostic records and treatment history of ophthalmic diseases from the hospital's electronic medical record system, while collecting patient data directly related to ophthalmic diseases, including age, gender, and family medical history;

[0016] Public dataset acquisition: Find and obtain publicly available image datasets of ophthalmic diseases;

[0017] Image data collection for ophthalmic diseases involves acquiring image data of ophthalmic diseases from image databases of hospitals and clinics, and recording the image acquisition time and equipment model information.

[0018] Furthermore, the data acquisition and processing specifically includes the following steps:

[0019] Data cleaning removes duplicate, blurry, damaged, and unrecognizable images. Image enhancement techniques are applied to expand the dataset. Based on the characteristics of ophthalmic disease images, image regions containing key information are cropped. Unnecessary background and interference information in the images are removed. An appropriate scaling ratio is selected according to the model's input requirements to scale the images to a uniform size. The pixel values ​​of the images are normalized to the range of [0, 1] or [-1, 1] to eliminate differences in brightness and contrast between different images.

[0020] Furthermore, the data labeling and partitioning specifically includes the following steps:

[0021] Disease type identification involves reviewing ophthalmic disease image data based on diagnostic results, identifying the type of ophthalmic disease in each image, and labeling each image with a corresponding ophthalmic disease tag. The tag information includes the type, severity, and location information of the disease. The severity is quantitatively assessed based on the disease progression stage and lesion extent, and the location information is accurate to the specific location of the lesion in the fundus image and OCT image.

[0022] Furthermore, the data labeling and partitioning specifically includes the following steps:

[0023] The preprocessed data is divided into training set, validation set and test set;

[0024] Of the data, 80% is used as the training set to train the deep learning model, 10% is used as the validation set to verify the model's performance and tune hyperparameters, and 10% is used as the test set to ultimately test the model's performance.

[0025] Furthermore, the construction of the learning model specifically includes the following steps:

[0026] Convolutional neural networks were chosen as the base network. Based on the characteristics and analysis requirements of ophthalmic disease images, the specific number of layers and structure of the network were designed. Based on the type of ophthalmic disease labels, the structure of the output layer and the binary cross-entropy loss function were designed.

[0027] Furthermore, the construction of the learning model specifically includes the following steps:

[0028] Based on the size and number of channels of the preprocessed ophthalmic disease image data, the input layer size of the model is set, and based on the annotation information of the ophthalmic disease labels, the output layer structure of the model is set.

[0029] Furthermore, the training of the learning model specifically includes the following steps:

[0030] Before training begins, the model's weights and bias parameters are initialized, the labeled and partitioned training set is input into the model, and the model's output is calculated through forward propagation.

[0031] Calculate the loss value between the model output and the true label using the defined loss function;

[0032] The gradient of the loss value with respect to the model parameters is calculated using the backpropagation algorithm, and the model parameters are updated using the SGD optimizer.

[0033] During training, the model's performance is monitored using a validation set, which includes accuracy, recall, and F1 score. Model parameters and hyperparameters are then adjusted based on the performance on the validation set.

[0034] When the model's performance on the validation set reaches a preset threshold or the performance improvement is not significant over multiple consecutive epochs, save the parameters and structure of the current model.

[0035] Furthermore, the evaluation of the learning model specifically includes the following steps:

[0036] The test set is input into the trained model to obtain the model's prediction results;

[0037] The model's prediction results are evaluated using predetermined evaluation metrics, and the specific values ​​of each evaluation metric are calculated to quantify the model's performance.

[0038] Obtain the diagnostic results of human experts on the test set data, compare the model's prediction results with the diagnostic results of human experts, and iterate and optimize the model based on the comparison results.

[0039] Repeat the process of training, evaluating, and optimizing the model until its performance meets the expected requirements.

[0040] Furthermore, the application analysis and optimization specifically includes the following steps:

[0041] Cloud deployment involves deploying the trained eye disease analysis model to a cloud server, configuring the server environment, writing API interfaces, and enabling web and mobile applications to call the cloud model for diagnostic analysis via network requests based on the API interfaces.

[0042] As needed, deploy eye disease analysis models on doctors' local servers and computers;

[0043] A feedback mechanism is set up so that doctors can provide feedback after using the eye disease analysis model. The feedback includes the accuracy of the diagnostic results, suggestions for improving the user experience, and problems encountered by the model in practical applications.

[0044] Collect and analyze feedback data provided by doctors, and continuously improve and optimize the eye disease analysis model based on the feedback data.

[0045] Compared with the prior art, the beneficial effects of the present invention are:

[0046] 1. The diversity and standardization in the data acquisition and processing of this invention help train a more robust model that can cope with various complex clinical scenarios and image data. Ophthalmic disease diagnosis records, treatment history, and patient data (such as age, gender, and family medical history) extracted from electronic medical records and hospital records can be combined with image data to provide the model with richer contextual information. This correlation analysis helps the model to understand the characteristics and influencing factors of ophthalmic diseases more comprehensively, thereby improving the accuracy and reliability of diagnosis.

[0047] 2. This invention can quantitatively evaluate the model's performance through pre-determined evaluation indicators, providing clear performance metrics. It also enables developers to clearly understand the model's performance on ophthalmic disease diagnosis tasks, thereby allowing for targeted optimization. By comparing the model's prediction results with those of human experts, it provides real-world feedback to the model, revealing not only where the model falls short of human experts but also providing directions for improvement.

[0048] 3. This invention deploys the trained eye disease analysis model to a cloud server, allowing doctors to access the model anytime, anywhere via network requests for diagnostic analysis. This improves diagnostic efficiency and convenience, enabling doctors to obtain diagnostic results more quickly and provide patients with more timely medical services. Doctors can personalize the model according to their actual work environment and needs, ensuring data security and privacy. It also enhances the system's flexibility and scalability. Real-time feedback provides valuable data support for continuous model improvement and optimization, allowing the model to adapt to clinical needs, improve diagnostic accuracy and reliability, and continuously improve and optimize the eye disease analysis model based on collected and analyzed feedback data from doctors. This makes the model improvement more precise and efficient, quickly solving problems encountered in practical applications and improving model performance and user experience. Attached Figure Description

[0049] Figure 1 This is a flowchart illustrating the deep learning-based ophthalmic disease analysis method of the present invention. Detailed Implementation

[0050] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0051] To address the technical issues that traditional ophthalmic disease diagnostic methods often rely on limited data sources and inadequate data processing, resulting in poor performance of trained models when dealing with complex clinical scenarios and image data, and the potential for errors in diagnostic results due to factors such as diagnostic tools and physician experience, please refer to [link to relevant documentation]. Figure 1 The present invention provides the following technical solutions:

[0052] A deep learning-based method for analyzing ophthalmic diseases includes the following steps:

[0053] Data acquisition and processing involves collecting data from electronic medical records, public datasets, and hospital records, as well as collecting ophthalmic disease image data, including fundus images and OCT images. The ophthalmic disease image data is cleaned and enhanced, and preprocessed by cropping, scaling, and normalization.

[0054] Data labeling and segmentation: Identify the types of ophthalmic diseases in ophthalmic disease image data; perform data structuring processing on the target ophthalmic disease data according to the ophthalmic disease type; label each image with a corresponding ophthalmic disease label; the labeling information of the ophthalmic disease label includes the type, severity, and location information of the disease; and segment the labeled ophthalmic disease image data.

[0055] The learning model is built by constructing a deep learning model containing a convolutional neural network. The convolutional neural network is used to extract feature information from ophthalmic disease image data. The preprocessed ophthalmic disease image data is input into the deep learning model, and the key feature information of the ophthalmic disease image is extracted through the convolutional neural network.

[0056] The training process involves learning the model, constructing a loss function to guide the training of the deep learning model, using a labeled and partitioned training set and key feature information to train the deep learning model, optimizing the model parameters during training, monitoring the model performance through the validation set, adjusting the model parameters and hyperparameters based on the performance on the validation set, and saving the model parameters when the model performs well on the validation set.

[0057] The learning model is evaluated by using a test set to measure its performance after training and comparing it with the diagnostic results of human experts. Based on the evaluation results, the model is iterated and optimized to obtain a successfully trained eye disease analysis model.

[0058] Application analysis and optimization involves deploying the trained eye disease analysis model to the cloud and local servers. Doctors can use the eye disease analysis model to assist in the diagnosis and analysis of patients through web applications and mobile applications. Feedback from doctors after using the model is collected to continuously improve and optimize the model.

[0059] In the above embodiments, deep learning models can automatically learn and extract key features from ophthalmic disease images, thereby more accurately identifying and analyzing ophthalmic diseases. This method is more objective and reliable than traditional manual diagnosis, reducing human error. Deep learning models can process large amounts of image data in a short time, providing doctors with fast and efficient diagnostic support. When there are many patients or complex image data, deep learning models can significantly reduce the workload of doctors and improve diagnostic efficiency.

[0060] In the above embodiments, by performing fine-grained analysis on images of ophthalmic diseases, deep learning models can provide detailed diagnostic results on disease type, severity, and location information. This helps doctors develop personalized treatment plans based on the specific conditions of patients, improving treatment outcomes. Deploying the trained ophthalmic disease analysis model to the cloud and local servers enables doctors to conduct remote diagnoses through web and mobile applications, helping to solve the problem of uneven distribution of medical resources and allowing more patients to enjoy professional ophthalmic medical services.

[0061] In the above embodiments, by collecting feedback from doctors after using the model, the deep learning model can be continuously improved and optimized. This iterative optimization process enables the model to continuously adapt to new clinical scenarios and data distributions, maintaining the stability and advancement of its diagnostic performance.

[0062] Data acquisition and processing specifically includes the following steps:

[0063] Electronic medical records and hospital record collection: Extracting diagnostic records and treatment history of ophthalmic diseases from the hospital's electronic medical record system, while collecting patient data directly related to ophthalmic diseases, including age, gender, and family medical history;

[0064] Public dataset acquisition: Find and obtain publicly available image datasets of ophthalmic diseases;

[0065] Image data collection for ophthalmic diseases involves acquiring image data of ophthalmic diseases from image databases of hospitals and clinics, and recording the image acquisition time and equipment model information.

[0066] Data cleaning removes duplicate, blurry, damaged, and unrecognizable images. Image enhancement techniques are applied to expand the dataset. Based on the characteristics of ophthalmic disease images, image regions containing key information are cropped. Unnecessary background and interference information in the images are removed. An appropriate scaling ratio is selected according to the model's input requirements to scale the images to a uniform size. The pixel values ​​of the images are normalized to the range of [0, 1] or [-1, 1] to eliminate differences in brightness and contrast between different images.

[0067] In the above embodiments, the data cleaning step removes duplicate, blurry, damaged, and unrecognizable images, ensuring the quality and accuracy of the training data. Image enhancement technology can expand the dataset, increase the model's generalization ability, crop out image regions containing key information, and remove unnecessary background and interference information, enabling the model to focus more on learning disease features and improving the model's diagnostic accuracy. Image scaling and normalization eliminate the differences in brightness and contrast between different images, making it easier for the model to extract useful feature information from images, thereby optimizing the model's performance.

[0068] In the above embodiments, the diversity and standardization of data acquisition and processing help train a more robust model that can cope with various complex clinical scenarios and image data. Ophthalmic disease diagnosis records, treatment history, and patient data (such as age, gender, and family medical history) extracted from electronic medical records and hospital records can be combined with image data to provide the model with richer contextual information. This correlation analysis helps the model to understand the characteristics and influencing factors of ophthalmic diseases more comprehensively, thereby improving the accuracy and reliability of diagnosis.

[0069] In the above embodiments, finding and obtaining publicly available ophthalmic disease image datasets can further expand the training data and improve the generalization ability of the model. At the same time, public datasets are usually subject to strict quality control and annotation, which helps to improve the training effect of the model.

[0070] Data labeling and partitioning also includes the following steps:

[0071] Disease type identification involves reviewing ophthalmic disease image data based on diagnostic results, identifying the type of ophthalmic disease in each image, and labeling each image with a corresponding ophthalmic disease tag. The tag information includes the type, severity, and location information of the disease. The severity is quantitatively assessed based on the disease progression stage and lesion extent, and the location information is accurate to the specific location of the lesion in the fundus image and OCT image.

[0072] The preprocessed data is divided into training set, validation set and test set;

[0073] Of the data, 80% is used as the training set to train the deep learning model, 10% is used as the validation set to verify the model's performance and tune hyperparameters, and 10% is used as the test set to ultimately test the model's performance.

[0074] In the above embodiments, by identifying and labeling corresponding ophthalmic disease tags through disease type identification, including disease type, severity, and precise location information, the deep learning model can learn more comprehensive and detailed disease features, thereby improving the accuracy of diagnosis. The validation set is used to monitor the performance of the model during the training process, and hyperparameters such as learning rate and batch size are adjusted based on the performance of the validation set to ensure that the model can continuously optimize its performance during training. By dividing the data into training set, validation set, and test set, it is possible to effectively prevent the model from overfitting on the training set and ignoring its generalization ability to new data.

[0075] In the above embodiments, accurate location information and quantitative assessment of severity help doctors better understand the distribution and progression of the disease, thereby developing more precise treatment plans. This is of great significance for improving the efficiency of diagnosis and treatment of ophthalmic diseases and the quality of life of patients. By evaluating the model through a test set and iterating and optimizing based on the evaluation results, the performance of the model can be continuously improved.

[0076] The learning model building process also includes the following steps:

[0077] Convolutional neural networks were chosen as the base network. Based on the characteristics and analysis requirements of ophthalmic disease images, the specific number of layers and structure of the network were designed. Based on the type of ophthalmic disease labels, the structure of the output layer and the binary cross-entropy loss function were designed.

[0078] Based on the size and number of channels of the preprocessed ophthalmic disease image data, the input layer size of the model is set, and based on the annotation information of the ophthalmic disease labels, the output layer structure of the model is set.

[0079] In the above embodiments, a convolutional neural network is selected as the base network, and the specific number of layers and structure of the network are designed according to the characteristics and analysis needs of ophthalmic disease images. This customized network structure can better capture the specific features of ophthalmic disease images, thereby improving the diagnostic accuracy of the model. According to the type of ophthalmic disease label, the structure of the output layer and the binary cross-entropy loss function are designed so that the model can accurately predict the type, severity and location information of the disease, which not only improves the predictive ability of the model, but also makes it more in line with actual clinical needs.

[0080] In the above embodiments, the input layer size of the model is set according to the size and number of channels of the preprocessed ophthalmic disease image data, ensuring that the model can handle ophthalmic disease images of various sizes and types. This highly adaptable model design can make full use of existing ophthalmic disease image data and improve data utilization. Through reasonable network layer number, structure design, and loss function selection, the model can converge quickly during training and achieve high prediction accuracy, while making the model more stable during training and reducing the risk of overfitting and underfitting.

[0081] In the above embodiments, convolutional neural networks possess powerful feature extraction capabilities. Visualization and other techniques can be used to demonstrate the regions and features the model focuses on when identifying ophthalmic diseases, thereby improving the model's interpretability. This is crucial for doctors to understand and trust the model's diagnostic results. Customized network structure design makes the model easy to extend and upgrade. When new ophthalmic disease image data or analysis needs arise, the network structure can be adjusted or new network layers added to adapt to the new requirements, thus maintaining the model's advanced nature and practicality.

[0082] Learning model training also includes the following steps:

[0083] Before training begins, the model's weights and bias parameters are initialized, the labeled and partitioned training set is input into the model, and the model's output is calculated through forward propagation.

[0084] Calculate the loss value between the model output and the true label using the defined loss function;

[0085] The gradient of the loss value with respect to the model parameters is calculated using the backpropagation algorithm, and the model parameters are updated using the SGD optimizer.

[0086] During training, the model's performance is monitored using a validation set, which includes accuracy, recall, and F1 score. Model parameters and hyperparameters are then adjusted based on the performance on the validation set.

[0087] When the model's performance on the validation set reaches a preset threshold or the performance improvement is not significant over multiple consecutive epochs, save the parameters and structure of the current model.

[0088] In the above embodiments, by inputting the labeled and partitioned training set into the model, and combining forward propagation, loss function calculation and backpropagation algorithms, the model can gradually learn the features in ophthalmic disease images, thereby continuously optimizing its performance. This iterative training method enables the model to gradually adapt to the data distribution and improve the accuracy and robustness of diagnosis.

[0089] In the above embodiments, during training, a validation set is used to monitor model performance, including metrics such as accuracy, recall, and F1 score. These metrics comprehensively evaluate model performance, helping developers identify and adjust model issues promptly. Furthermore, adjusting model parameters and hyperparameters based on validation set performance can further optimize model performance, making it more aligned with actual clinical needs.

[0090] In the above embodiments, by using a validation set to monitor model performance during training and stopping training when performance improvement is not significant, overfitting can be effectively prevented. Overfitting is a common problem in deep learning model training, causing the model to perform well on the training set but degrade on the test set. Through reasonable monitoring and adjustment mechanisms, overfitting can be avoided, improving the model's generalization ability.

[0091] The evaluation of learning models also includes the following steps:

[0092] The test set is input into the trained model to obtain the model's prediction results;

[0093] The model's prediction results are evaluated using predetermined evaluation metrics, and the specific values ​​of each evaluation metric are calculated to quantify the model's performance.

[0094] Obtain the diagnostic results of human experts on the test set data, compare the model's prediction results with the diagnostic results of human experts, and iterate and optimize the model based on the comparison results.

[0095] Repeat the process of training, evaluating, and optimizing the model until its performance meets the expected requirements.

[0096] In the above embodiments, the model's performance can be quantitatively evaluated using pre-defined evaluation metrics. This quantitative evaluation not only provides clear performance metrics but also enables developers to clearly understand the model's performance on ophthalmic disease diagnosis tasks, allowing for targeted optimization. Comparing the model's predictions with the diagnoses of human experts provides real-world feedback to the model. This comparison not only reveals where the model falls short of human experts but also provides directions for improvement. By continuously comparing the model's predictions with human expert diagnoses and iterating and optimizing accordingly, the model's diagnostic accuracy and reliability can be significantly improved.

[0097] Application analysis and optimization also include the following steps:

[0098] Cloud deployment involves deploying the trained eye disease analysis model to a cloud server, configuring the server environment, and writing API interfaces. Web applications and mobile applications can then use these API interfaces to call the cloud model for diagnostic analysis via network requests.

[0099] As needed, deploy eye disease analysis models on doctors' local servers and computers;

[0100] A feedback mechanism is set up so that doctors can provide feedback after using the eye disease analysis model. The feedback includes the accuracy of the diagnostic results, suggestions for improving the user experience, and problems encountered by the model in practical applications.

[0101] Collect and analyze feedback data provided by doctors, and continuously improve and optimize the eye disease analysis model based on the feedback data.

[0102] In the above embodiments, by deploying the trained eye disease analysis model to a cloud server, doctors can access the cloud model for diagnostic analysis anytime, anywhere via network requests. This improves the efficiency and convenience of diagnosis, allowing doctors to obtain diagnostic results more quickly and provide patients with more timely medical services. Doctors can personalize the model according to their actual work environment and needs, ensuring data security and privacy. The system's flexibility and scalability are also improved. Real-time feedback provides valuable data support for continuous model improvement and optimization, enabling the model to continuously adapt to clinical needs, improve diagnostic accuracy and reliability, and collect and analyze feedback data provided by doctors. Based on this data, the eye disease analysis model can be continuously improved and optimized, making the model improvement more precise and efficient. This allows for rapid resolution of problems encountered in practical applications, improving model performance and user experience.

[0103] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for analyzing ophthalmic diseases based on deep learning, characterized in that: The following steps are involved: Data collection and processing: collect data from electronic medical records, public data sets, and hospital records, collect ophthalmic disease image data, including fundus images, OCT images, etc., clean and enhance the ophthalmic disease image data, and perform cropping, scaling, and normalization preprocessing on the images; Data annotation and division, identifying the types of ophthalmic diseases in ophthalmic disease image data, performing data structuring processing on the target ophthalmic disease data according to the types of ophthalmic diseases, annotating each image with a corresponding ophthalmic disease label, wherein the annotation information of the ophthalmic disease label includes the type, severity, and location information of the disease, and performing data division on the annotated ophthalmic disease image data; Learning model construction: build a deep learning model including a convolutional neural network, where the convolutional neural network is used to extract feature information from ophthalmic disease image data. The preprocessed ophthalmic disease image data is input into the deep learning model, and the key feature information of the ophthalmic disease image is extracted through the convolutional neural network. Learning model training, constructing a loss function to guide the training of deep learning models, using the labeled and divided training set and key feature information to train the deep learning model, optimizing the model parameters during the training process, and monitoring the performance of the model through the validation set, adjusting the model parameters and hyperparameters based on the performance on the validation set, and saving the model parameters when the validation set performs well; Learning model evaluation: After training, the model performance is evaluated using the test set and compared with the diagnosis results of human experts. The model is iterated and optimized based on the evaluation results to obtain a successfully trained eye disease analysis model. Application analysis optimization: deploy the trained eye disease analysis model to the cloud and local servers. Doctors use the eye disease analysis model through web applications and mobile applications to assist in diagnosis and analysis of patients. Collect feedback from doctors after using the model and continuously improve and optimize the model.

2. The method for analyzing ophthalmic diseases based on deep learning according to claim 1, characterized in that: The data collection and processing further includes the following steps: Electronic medical records and hospital records collection: extracting the diagnosis records and treatment history of ophthalmic diseases from the hospital's electronic medical record system, and collecting patient data directly related to ophthalmic diseases, including age, gender, and family medical history; Public dataset acquisition: find and obtain publicly available ophthalmic disease image datasets; Ophthalmic disease image data collection: obtain ophthalmic disease image data from the image database of hospitals and clinics, and record the image acquisition time and equipment model information.

3. The method for analyzing ophthalmic diseases based on deep learning according to claim 2, characterized in that: The data collection and processing further includes the following steps: Data cleaning: remove duplicate, blurred, damaged, and unrecognizable images, apply image enhancement technology to expand the data set, crop the image area containing key information according to the characteristics of ophthalmic disease images, remove unnecessary background and interference information in the image, select the appropriate scaling ratio according to the input requirements of the model, scale the image to a uniform size, normalize the pixel value of the image to the range of [0, 1] or [-1, 1], and eliminate the brightness and contrast differences between different images.

4. The method for analyzing ophthalmic diseases based on deep learning according to claim 1, characterized in that: The data annotation division specifically further includes the following steps: Disease type identification: review the ophthalmic disease image data based on the diagnosis results, identify the type of ophthalmic disease in each image, and label each image with a corresponding ophthalmic disease label. The label information includes the type, severity and location information of the disease, wherein the severity is quantitatively evaluated based on the stage of disease progression and the range of the lesion, and the location information is accurate to the specific location of the lesion in the fundus image and OCT image.

5. The method for analyzing ophthalmic diseases based on deep learning according to claim 4, characterized in that: The data annotation division specifically further includes the following steps: Divide the preprocessed data into training set, validation set and test set; Among them, 80% of the data is used as a training set, which is used to train the deep learning model; 10% of the data is used as a validation set, which is used to verify the model performance and adjust the hyperparameters; and 10% of the data is used as a test set, which is used to finally test the model performance.

6. The method for analyzing ophthalmic diseases based on deep learning according to claim 1, characterized in that: The learning model construction specifically includes the following steps: A convolutional neural network was selected as the basic network. According to the characteristics of ophthalmic disease images and analysis requirements, the specific number of layers and structure of the network were designed. According to the type of ophthalmic disease labels, the structure of the output layer and the binary cross entropy loss function were designed.

7. The method for analyzing ophthalmic diseases based on deep learning according to claim 6, characterized in that: The learning model construction specifically includes the following steps: The input layer size of the model is set according to the size and number of channels of the preprocessed ophthalmic disease image data, and the output layer structure of the model is set according to the annotation information of the ophthalmic disease labels.

8. The method for analyzing ophthalmic diseases based on deep learning according to claim 1, characterized in that: The learning model training specifically includes the following steps: Before starting training, initialize the model's weights and bias parameters, input the labeled and divided training set into the model, and calculate the model's output through forward propagation; Use the defined loss function to calculate the loss between the model output and the true label; The gradient of the loss value to the model parameters is calculated through the back-propagation algorithm, and the model parameters are updated using the SGD optimizer; During the training process, the validation set is used to monitor the performance of the model, including the accuracy, recall, and F1 score indicators, and the model parameters and hyperparameters are adjusted according to the performance on the validation set; When the performance of the model on the validation set reaches the preset threshold or the performance improvement for multiple consecutive epochs is not obvious, save the parameters and structure of the current model.

9. The method for analyzing ophthalmic diseases based on deep learning according to claim 1, characterized in that: The learning model evaluation specifically includes the following steps: Input the test set into the trained model to obtain the model's prediction results; Use pre-determined evaluation indicators to evaluate the prediction results of the model and calculate the specific values ​​of each evaluation indicator to quantify the performance of the model; Obtain the diagnosis results of human experts on the data in the test set, compare the prediction results of the model with the diagnosis results of human experts, and iterate and optimize the model based on the comparison results with the diagnosis results of human experts; The process of model training, evaluation, and optimization is repeated until the performance of the model meets the expected requirements.

10. The method for analyzing ophthalmic diseases based on deep learning according to claim 1, characterized in that: The application analysis optimization specifically further includes the following steps: Cloud deployment: deploy the trained eye disease analysis model to the cloud server, configure the server environment, write the API interface, and the web application and mobile application call the cloud model through the network request based on the API interface for diagnosis and analysis; Deploy eye disease analysis models on doctors’ local servers and computers as needed; Setting up a feedback mechanism through which doctors can provide feedback after using the eye disease analysis model. The feedback includes the accuracy of the diagnosis results, suggestions for improving the user experience, and problems encountered in the actual application of the model. Collect and analyze feedback data provided by doctors, and continuously improve and optimize the eye disease analysis model based on the feedback data provided by doctors.

Images