An intelligent diagnosis system for fundus image multi-disease probability visualization

Abstract

The application discloses an intelligent diagnosis system for fundus image multi-disease probability visualization, and particularly relates to the technical field of intelligent diagnosis, which comprises the following steps: constructing a patient fundus database for multi-disease joint diagnosis; generating a sample data set for multi-disease joint diagnosis model training based on the patient fundus database; training the sample data set through an image recognition model; receiving training samples generated by the multi-disease data module; performing feature extraction and multi-disease probability prediction on the training samples based on the image recognition model; visualizing the multi-disease probability vector; collecting risk weight information and disease degradation information of patient fundus images; comprehensively analyzing the risk weight information and the disease degradation information; constructing a display evaluation model; obtaining a generation result of the display evaluation model; comparing the generation result with a preset threshold; and screening visualized patient fundus images. The application helps doctors to quickly identify key cases and improve efficiency.

Description

Technical Field

[0001] This invention relates to the field of intelligent diagnostic technology, and more specifically, to an intelligent diagnostic system for visualizing the probability of multiple diseases in fundus images. Background Technology

[0002] In traditional ophthalmic image uploading, manually uploading patient images one by one is extremely tedious, especially when dealing with a large number of binocular color fundus images. Uploading each image manually is necessary, which becomes excessively cumbersome for multiple patients, requiring medical staff to expend a significant amount of time and energy. Furthermore, initial fundus images often have various problems, such as image tilt due to different shooting angles, black borders in the image, and inconsistencies in brightness and color temperature. These issues can affect the accuracy of subsequent diagnoses. Therefore, existing technologies use automatic cropping, rotation, and black border removal to process and beautify images, resulting in binocular color fundus images of uniform size, brightness, and color temperature. However, if the batch-processed images are displayed in a disorganized manner, it can make it difficult for medical staff to quickly obtain key information. Summary of the Invention

[0003] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide an intelligent diagnostic system for visualizing the probability of multiple diseases in fundus images, thereby addressing the problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: An intelligent diagnostic system for visualizing the probability of multiple diseases in fundus images includes a multi-disease data module, an image recognition module, a selective display module, and a threshold comparison module, with the modules interconnected. The multi-disease data module is used to construct a patient fundus database for the joint diagnosis of multiple diseases. Based on the patient fundus database, a sample dataset is generated for training the multi-disease joint diagnosis model, and the sample dataset is trained by an image recognition model. The image recognition module is used to receive training samples generated by the multi-disease data module, extract features and predict the probability of multiple diseases based on the image recognition model, and visualize the multi-disease probability vector. The selective display module is used to collect risk weight information and disease deterioration information from patients' fundus images, and to comprehensively analyze the risk weight information and disease deterioration information to construct a display evaluation model; The threshold comparison module is used to obtain the generated results of the evaluation model and compare the generated results with preset thresholds to filter the visualized patient fundus images.

[0005] In a preferred embodiment, fundus image data from multiple sources are collected, including hospital clinical data, public datasets, and historical follow-up data. The data covers a variety of ophthalmic disease types and normal samples. Each fundus image is labeled with multiple tags, the tags including at least two or more disease types, and each sample corresponds to a multi-disease tag vector. Data from different sources are standardized, including image format standardization, resolution standardization, and tag annotation.

[0006] In a preferred embodiment, the image recognition model is a DenseNet model, including an improved classification layer and a Dropout layer. The image recognition model adopts the DenseNet121 model with a DenseNet-BC structure. The image recognition model includes multiple Dense blocks, each Dense block consisting of multiple densely connected layers connected in series. Each densely connected layer sequentially includes a batch normalization layer, an activation layer, and a convolutional layer. In the Dense block, the outputs of the first n-1 densely connected layers are concatenated in parallel along the channel dimension and used as the input of the nth densely connected layer. The classification layer of the DenseNet model is improved to receive features from the output of the upper convolutional layer and image quality data of the fundus image.

[0007] In a preferred embodiment, the risk weight information is represented by a risk weight coefficient, and the logic for obtaining the risk weight coefficient is as follows: based on the recognition of the target patient's fundus image by the image recognition module, the multi-disease probability vector output by the image recognition module is obtained, and the multi-disease probability vector output by the image recognition module is marked as: ,in, , This represents the predicted probability value for the nth type of fundus disease. And satisfy Or the probability values ​​are independent of each other, where n is the preset number of categories of fundus diseases; Obtain a pre-defined risk score for a fundus disease and mark the pre-defined risk score for that fundus disease as follows: Where i represents the preset type of fundus disease, i = 1, 2, 3, ..., n, and i is a positive integer; The risk weighting coefficient is determined through weighted calculation. The formula for calculating the risk weighting coefficient is as follows: ;in, This is the risk weighting coefficient.

[0008] In a preferred embodiment, the disease deterioration information is represented by a disease deterioration coefficient. The logic for obtaining the disease deterioration coefficient is as follows: based on the fundus image data of the target patient, the lesion area, number of lesions, lesion density, and degree of damage to key structures of multiple diseases in the target patient are determined, and the lesion area, number of lesions, lesion density, and degree of damage to key structures of multiple diseases in the target patient are respectively marked as: , , , ; The target patient is matched with similar patients, and a set of similar patients with the most similar characteristics is selected. Based on the similar patient set, the lesion area, number of lesions, lesion density, and degree of damage to key structures of each similar patient are extracted. Combined with the corresponding time series information, the rate of change of lesion area, number of lesions, lesion density, and degree of damage to key structures of each similar patient in each time period is calculated. By statistically analyzing the rate of change of similar patients, the average rate of deterioration of lesion area, average rate of deterioration of lesion number, average rate of deterioration of lesion density, and average rate of deterioration of degree of damage to key structures of similar patients are obtained. The average rate of deterioration is used as the reference rate of deterioration of multiple diseases of the target patient. The disease deterioration coefficient is calculated using the following formula: in, This represents the disease deterioration coefficient. These are the rate of deterioration of average lesion area, average lesion number, average lesion density, and average critical structural damage degree among similar patients. The weights are respectively the lesion area, number of lesions, lesion density, and degree of damage to key structures of the target patient's multiple diseases.

[0009] In a preferred embodiment, risk weight information and disease deterioration information are comprehensively analyzed. A demonstration assessment model is constructed through a weighted calculation of the risk weight coefficient and the disease deterioration coefficient. The calculation formula for the demonstration assessment model is as follows: ;in, To display the evaluation coefficients, These are the risk weighting coefficient and the proportional coefficient of the disease deterioration coefficient, respectively. All are greater than 0.

[0010] In a preferred embodiment, the generated result is represented by a display evaluation coefficient. By setting a display evaluation coefficient threshold, the display evaluation coefficient of the target patient is compared with the display evaluation coefficient threshold. If the display evaluation coefficient is greater than the display evaluation coefficient threshold, the fundus image of the target patient is added to the visualization list; if the display evaluation coefficient is less than the display evaluation coefficient threshold, the fundus image of the target patient is temporarily hidden or processed with low priority.

[0011] The technical effects and advantages of this invention are as follows: This invention constructs a patient fundus database for joint diagnosis of multiple diseases and trains an image recognition model using a multi-label learning approach. This enables a single fundus image to simultaneously output the probability results of multiple diseases, breaking through the traditional single-disease diagnosis model and improving diagnostic efficiency and information utilization. Furthermore, through the synergistic effect of the selective display module and the threshold comparison module, this invention transforms complex multi-disease probability information and evaluation results into intuitive visual output, facilitating doctors to quickly identify key cases and improve clinical decision-making efficiency. Attached Figure Description

[0012] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings; Figure 1 This is a schematic diagram of the structure of the present invention. Detailed Implementation

[0013] 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.

[0014] Example 1 Figure 1 This is a schematic diagram of the structure of an intelligent diagnostic system for visualizing the probability of multiple fundus diseases according to the present invention. It includes a multi-disease data module, an image recognition module, a selective display module, and a threshold comparison module, and the modules are interconnected. The multi-disease data module is used to construct a patient fundus database for the joint diagnosis of multiple diseases. Based on the patient fundus database, a sample dataset is generated for training the multi-disease joint diagnosis model, and the sample dataset is trained by an image recognition model. The image recognition module is used to receive training samples generated by the multi-disease data module, extract features and predict the probability of multiple diseases based on the image recognition model, and visualize the multi-disease probability vector. The selective display module is used to collect risk weight information and disease deterioration information from patients' fundus images, and to comprehensively analyze the risk weight information and disease deterioration information to construct a display evaluation model; The threshold comparison module is used to obtain the generated results of the evaluation model and compare the generated results with preset thresholds to filter the visualized patient fundus images.

[0015] In the multi-disease data module, the acquisition logic of the patient fundus database is as follows: collect fundus image data from multiple sources, including hospital clinical data, public datasets, and historical follow-up data, which cover a variety of ophthalmic disease types and normal samples; perform multi-label annotation on each fundus image, with the labels including at least two or more disease types, and each sample corresponds to a multi-disease label vector to represent the existence status or probability distribution of each disease; and standardize the data from different sources, including image format standardization, resolution standardization, and label annotation.

[0016] Furthermore, the standardized fundus image data is subjected to quality screening to remove low-quality samples with blurriness, occlusion, or abnormal exposure; the consistency of multi-label annotation results is verified, and target labels are generated based on the multi-label result fusion strategy; the patient fundus database is stored in a structured manner and divided into training set, validation set, and test set.

[0017] The image recognition model is a DenseNet model, specifically a DenseNet121 model with a DenseNet-BC structure, including improved classification and dropout layers. This enables the model to effectively learn lesion features of fundus images with a limited number of training samples and accurately output multi-disease probability vectors. The image recognition model comprises multiple Dense blocks, each consisting of multiple densely connected layers connected in series. Each densely connected layer sequentially includes a batch normalization (BN) layer, a ReLU activation layer, and a convolutional (Cov) layer. In each Dense block, the outputs of the first n-1 densely connected layers are concatenated in parallel along the channel dimension and used as the input to the nth densely connected layer to form dense connections, enhancing feature reuse and mitigating the gradient vanishing problem.

[0018] Furthermore, to reduce the computational cost within each densely connected layer, a 1×1 convolutional layer is introduced to form a bottleneck structure, referred to as DenseNet-B. Dense blocks are connected by transition layers to control the number of feature channels and reduce the feature map size. These transition layers include batch normalization layers, ReLU layers, convolutional layers, and pooling layers. When the transition layers use a compression factor less than 1, the Dense block structure combined with the bottleneck layer forms the DenseNet-BC structure.

[0019] The classification layer of the DenseNet model has been improved to receive features from the output of the upper convolutional layers as well as image quality data of fundus images (output by the EyeQ model). By fusing the image quality data with the convolutional features, the model can more effectively distinguish lesions from image defects in the image, thereby improving the accuracy of multi-disease prediction.

[0020] To address the overfitting problem, a Dropout layer is introduced into the DenseNet model. The Dropout layer breaks the complex co-adaptation relationship between neurons by randomly dropping neuron connections, enabling the network to learn more robust feature representations during training. This improves the generalization ability of deep neural networks and reduces overfitting caused by high model complexity during training.

[0021] It's worth noting that the Dropout layer breaks the complex co-fitting relationships between neurons by randomly dropping them, thus forcing the network to learn with less overfitting. Introducing the Dropout layer into the DenseNet121 model allows different sub-networks to be trained in each batch, effectively improving the generalization ability of deep neural networks and significantly reducing overfitting during training. It also reduces overfitting caused by high model complexity and shortens training time.

[0022] In the visualization module, the visualization process of the multi-disease probability vector includes: associating the multi-disease prediction probability value corresponding to each fundus image output by the image recognition module with its corresponding disease label to form structured probability vector information; sorting different diseases according to the predicted probability of each disease, marking or highlighting high-probability disease categories to intuitively reflect disease risk, and displaying low-probability disease categories in gray or transparent to reduce information interference; supporting batch display of multiple fundus images, arranged uniformly according to patient, left and right eyes, or disease category, allowing quick browsing of key disease information in each image in batch display, improving the viewing efficiency of medical staff, and providing interactive functions, such as displaying specific probability values ​​by hovering the mouse, or switching between different disease views by clicking.

[0023] In the selective display module, the risk weight information is represented by a risk weight coefficient. The logic for obtaining the risk weight coefficient is as follows: based on the recognition of the target patient's fundus image by the image recognition module, the multi-disease probability vector output by the image recognition module is obtained, and the multi-disease probability vector output by the image recognition module is marked as follows: ,in, , This represents the predicted probability value for the nth type of fundus disease. And satisfy Or the probability values ​​are independent of each other, where n is the preset number of categories of fundus diseases; Obtain a pre-defined risk score for a fundus disease and mark the pre-defined risk score for that fundus disease as follows: Where i represents the preset type of fundus disease, i = 1, 2, 3, ..., n, and i is a positive integer; It should be noted that the preset risk score for fundus diseases is based on a clinical medical knowledge base to preset the severity of different fundus diseases. The severity includes the degree of visual impairment, the risk level of blindness, and the irreversibility of disease progression. Combined with historical case data, the clinical outcomes of different fundus diseases at different stages are statistically analyzed to obtain the risk benchmark value corresponding to each disease. The risk benchmark value is dynamically adjusted according to the patient's individual characteristics, including age, history of underlying diseases, previous fundus disease history, and lifestyle factors. Expert experience rules are introduced to increase the risk score weight of some high-risk diseases. The above factors are weighted and integrated to obtain the preset risk score for fundus diseases. The risk score is then normalized to meet the preset numerical range.

[0024] The risk weighting coefficient is determined through weighted calculation. The formula for calculating the risk weighting coefficient is as follows: ;in, This is the risk weighting coefficient.

[0025] It should be noted that by fusing the multi-disease prediction probabilities output by the image recognition module with the risk score constructed based on clinical medical knowledge, the bias caused by relying solely on model probabilities is avoided. This allows the assessment results to take into account both the probability of disease occurrence and the actual degree of harm, thereby improving the accuracy of the diagnostic results. Furthermore, by constructing risk weight coefficients and performing normalization processing, the risks of different types of fundus diseases can be compared under the same dimension, which facilitates the sorting, screening, and grading of multiple diseases. The larger the risk weight coefficient, the higher the disease risk present in the current patient's fundus image. Therefore, it is more necessary to visualize and display the fundus image.

[0026] The disease deterioration information is represented by a disease deterioration coefficient. The logic for obtaining the disease deterioration coefficient is as follows: Based on the fundus image data of the target patient, the lesion area, number of lesions, lesion density, and degree of damage to key structures of multiple diseases in the target patient are determined. The lesion area, number of lesions, lesion density, and degree of damage to key structures of multiple diseases in the target patient are respectively marked as: , , , ; It should be noted that the lesion area calculation is performed by counting pixels in the lesion mask for each disease, converting the number of pixels into the actual area; the number of lesions is calculated by performing connected component analysis or labeling on the segmented lesions to count the number of independent lesions for each disease; the lesion density is calculated by comparing the number of lesions with the area of ​​the lesion-covered region, which reflects the concentration of lesion distribution; and the degree of damage to key structures is achieved by automatically detecting and extracting features from key structures in the fundus (such as the macula of the retina, the optic nerve head, and blood vessel distribution), using image segmentation or deep learning models to label key structural regions, generating corresponding masks, analyzing the lesion masks for each disease, and calculating the coverage ratio of lesions in the key structural regions.

[0027] The target patient is matched with similar patients, and a set of similar patients with the most similar characteristics is selected. Based on the similar patient set, the lesion area, number of lesions, lesion density, and degree of damage to key structures of each similar patient are extracted. Combined with the corresponding time series information, the rate of change of lesion area, number of lesions, lesion density, and degree of damage to key structures of each similar patient in each time period is calculated. By statistically analyzing the rate of change of similar patients, the average rate of deterioration of lesion area, average rate of deterioration of lesion number, average rate of deterioration of lesion density, and average rate of deterioration of degree of damage to key structures of similar patients are obtained. The average rate of deterioration is used as the reference rate of deterioration of multiple diseases of the target patient. The disease deterioration coefficient is calculated using the following formula: in, This represents the disease deterioration coefficient. These are the rate of deterioration of average lesion area, average lesion number, average lesion density, and average critical structural damage degree among similar patients. The weights are respectively the lesion area, number of lesions, lesion density, and degree of damage to key structures of the target patient's multiple diseases.

[0028] It should be noted that the disease deterioration coefficient takes into account the lesion area, number of lesions, lesion density, and degree of damage to key structures, avoiding the assessment bias that may be caused by a single indicator. It can reflect the spatial distribution, cumulative number, and comprehensive damage to key visual structures of the disease, which is closer to the real clinical risk. Furthermore, it uses a set of similar patients to calculate the average deterioration rate, comparing the lesion progression of the target patient with the trend of changes in similar patients in history, thus achieving personalized assessment. The higher the disease deterioration coefficient, the faster the disease deteriorates in the target patient.

[0029] As a preferred technical solution of this application, risk weight information and disease deterioration information are comprehensively analyzed. A demonstration assessment model is constructed through a weighted calculation of the risk weight coefficient and the disease deterioration coefficient. The calculation formula for the demonstration assessment model is as follows: ;in, To display the evaluation coefficients, These are the risk weighting coefficient and the proportional coefficient of the disease deterioration coefficient, respectively. All are greater than 0.

[0030] It should be noted that the aforementioned assessment model achieves a quantitative assessment of the multiple retinal disease status of patients by comprehensively analyzing the risk weight information and disease deterioration information of the target patients. Among them, the risk weight coefficient reflects the potential harm of various retinal diseases to visual function and the risk of blindness, while the disease deterioration coefficient reflects the actual progression rate of multiple diseases and the damage to key structures. By weighted calculation, the two types of information are integrated to obtain a comprehensive assessment coefficient, which can simultaneously take into account the potential risks and actual progression status of diseases, providing clinicians with quantifiable, comparable, and visualized multi-disease risk assessment results.

[0031] In the threshold comparison module, the generated result is represented by a display evaluation coefficient. By setting a display evaluation coefficient threshold, the display evaluation coefficient of the target patient is compared with the display evaluation coefficient threshold. If the display evaluation coefficient is greater than the display evaluation coefficient threshold, the target patient's fundus image is added to the visualization list for doctors to display and analyze in the visualization interface. If the display evaluation coefficient is less than the display evaluation coefficient threshold, the target patient's fundus image is temporarily hidden or processed with low priority to reduce the doctor's workload.

[0032] It should be noted that the threshold comparison module can dynamically adjust the preset threshold based on doctor feedback and patient population characteristics to adapt to different risk levels and clinical application scenarios, enabling personalized and refined patient image screening. It also supports multi-level risk control and personalized threshold settings to suit different clinical scenarios.

[0033] This invention constructs a patient fundus database for joint diagnosis of multiple diseases and trains an image recognition model using a multi-label learning approach. This enables a single fundus image to simultaneously output the probability results of multiple diseases, breaking through the traditional single-disease diagnosis model and improving diagnostic efficiency and information utilization. Furthermore, through the synergistic effect of the selective display module and the threshold comparison module, this invention transforms complex multi-disease probability information and evaluation results into intuitive visual output, facilitating doctors to quickly identify key cases and improve clinical decision-making efficiency.

[0034] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0035] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as 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, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. 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 or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0036] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0037] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0038] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0039] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0040] If the aforementioned functions are implemented as 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 this application, in essence, or the part that contributes to the prior art, or a portion of the 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 to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0041] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. An intelligent diagnostic system for visualizing the probability of multiple diseases in fundus images, characterized in that, It includes a multi-disease data module, an image recognition module, a selective display module, and a threshold comparison module, with the modules interconnected. The multi-disease data module is used to construct a patient fundus database for the joint diagnosis of multiple diseases. Based on the patient fundus database, a sample dataset is generated for training the multi-disease joint diagnosis model, and the sample dataset is trained by an image recognition model. The image recognition module is used to receive training samples generated by the multi-disease data module, extract features and predict the probability of multiple diseases based on the image recognition model, and visualize the multi-disease probability vector. The selective display module is used to collect risk weight information and disease deterioration information from patients' fundus images, and to comprehensively analyze the risk weight information and disease deterioration information to construct a display evaluation model; The threshold comparison module is used to obtain the generated results of the evaluation model and compare the generated results with preset thresholds to filter the visualized patient fundus images.

2. The intelligent diagnostic system for visualizing the probability of multiple diseases in fundus images according to claim 1, characterized in that: Fundus image data from multiple sources were collected, including hospital clinical data, public datasets, and historical follow-up data. The data covered a variety of ophthalmic disease types and normal samples. Each fundus image was labeled with multiple tags, with the tags including at least two or more disease types. Each sample corresponded to a multi-disease tag vector. Data from different sources were standardized, including image format standardization, resolution standardization, and tag annotation.

3. The intelligent diagnostic system for visualizing the probability of multiple diseases in fundus images according to claim 2, characterized in that: The image recognition model is a DenseNet model, including an improved classification layer and a Dropout layer. The image recognition model adopts the DenseNet121 model with the DenseNet-BC structure. The image recognition model includes multiple Dense blocks, each of which consists of multiple densely connected layers connected in series. Each densely connected layer includes a batch normalization layer, an activation layer, and a convolutional layer in sequence. In the Dense block, the outputs of the first n-1 densely connected layers are concatenated in parallel along the channel dimension and used as the input of the nth densely connected layer. The classification layer of the DenseNet model is improved to receive features from the output of the upper convolutional layer and image quality data of the fundus image.

4. The intelligent diagnostic system for visualizing the probability of multiple diseases in fundus images according to claim 3, characterized in that: The risk weight information is represented by risk weight coefficients. The logic for obtaining the risk weight coefficients is as follows: based on the recognition of the target patient's fundus image by the image recognition module, the multi-disease probability vector output by the image recognition module is obtained, and the multi-disease probability vector output by the image recognition module is marked as follows: ,in, , This represents the predicted probability value for the nth type of fundus disease. And satisfy Or the probability values ​​are independent of each other, where n is the preset number of categories of fundus diseases; Obtain a pre-defined risk score for a fundus disease and mark the pre-defined risk score for that fundus disease as follows: Where i represents the preset type of fundus disease, i = 1, 2, 3, ..., n, and i is a positive integer; The risk weighting coefficient is determined through weighted calculation. The formula for calculating the risk weighting coefficient is as follows: ;in, This is the risk weighting coefficient.

5. The intelligent diagnostic system for visualizing the probability of multiple diseases in fundus images according to claim 4, characterized in that: The disease deterioration information is represented by a disease deterioration coefficient. The logic for obtaining the disease deterioration coefficient is as follows: Based on the fundus image data of the target patient, the lesion area, number of lesions, lesion density, and degree of damage to key structures of multiple diseases in the target patient are determined. The lesion area, number of lesions, lesion density, and degree of damage to key structures of multiple diseases in the target patient are respectively marked as: , , , ; The target patient is matched with similar patients, and a set of similar patients with the most similar characteristics is selected. Based on the similar patient set, the lesion area, number of lesions, lesion density, and degree of damage to key structures of each similar patient are extracted. Combined with the corresponding time series information, the rate of change of lesion area, number of lesions, lesion density, and degree of damage to key structures of each similar patient in each time period is calculated. By statistically analyzing the rate of change of similar patients, the average rate of deterioration of lesion area, average rate of deterioration of lesion number, average rate of deterioration of lesion density, and average rate of deterioration of degree of damage to key structures of similar patients are obtained. The average rate of deterioration is used as the reference rate of deterioration of multiple diseases of the target patient. The disease deterioration coefficient is calculated using the following formula: ;in, ； This represents the disease deterioration coefficient. These are the rate of deterioration of average lesion area, average lesion number, average lesion density, and average critical structural damage degree among similar patients. The weights are respectively the lesion area, number of lesions, lesion density, and degree of damage to key structures of the target patient's multiple diseases.

6. The intelligent diagnostic system for visualizing the probability of multiple diseases in fundus images according to claim 5, characterized in that: By comprehensively analyzing risk weight information and disease deterioration information, and through weighted calculation of risk weight coefficient and disease deterioration coefficient, a demonstration assessment model is constructed. The calculation formula for the demonstration assessment model is as follows: ;in, To display the evaluation coefficients, These are the risk weighting coefficient and the proportional coefficient of the disease deterioration coefficient, respectively. All are greater than 0.

7. The intelligent diagnostic system for visualizing the probability of multiple diseases in fundus images according to claim 6, characterized in that: The generated result is represented by a display evaluation coefficient. By setting a display evaluation coefficient threshold, the display evaluation coefficient of the target patient is compared with the display evaluation coefficient threshold. If the display evaluation coefficient is greater than the display evaluation coefficient threshold, the fundus image of the target patient is added to the visualization list. If the display evaluation coefficient is less than the display evaluation coefficient threshold, the fundus image of the target patient is temporarily hidden or processed with low priority.

Images