Bidirectional prediction method, control device, storage medium and system
Through a two-way prediction method based on fundus image and physical examination indicators, the correlation between fundus image characteristics and physical examination indicators is determined, and the in-depth prediction of retinopathy is achieved, which solves the problem of narrow disease prediction range in the prior art, and improves the prediction accuracy and reliability.
Patent Information
- Application Number
- CN202411448988.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-17
- Publication Date
- 2025-01-14
AI Technical Summary
In the prior art, disease prediction based on fundus images and fundus blood vessel characteristics can only be predicted in large-scale areas, and it has not involved in the detailed physical examination indicators of the body, especially in different retinal lesions, retinal blood vessel indicators and total physical examination indicators. The relationship between them has not been effectively discussed.
Using a two-way prediction method based on fundus images and physical examination indicators, the correlation between fundus images and multiple physical examination indicators is determined by obtaining fundus images and data from multiple physical examination indicators, and using preset vascular segmentation model and decision tree method, the Retinopathy is predicted and a two-way prediction result is constructed for detection.
In-depth and accurate prediction of diseases is achieved, data bias is avoided, and the reliability and accuracy of intelligent disease prediction is improved, providing rich information and evidence for disease prediction.
Smart Images
Figure CN119314678A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of medical technology, and in particular to a bidirectional prediction method based on fundus images and physical examination indicators, a control device, a storage medium and a system. Background Art
[0002] Automatic extraction of retinal vessels based on deep learning systems makes effective quantitative analysis of retinal vascular abnormalities a reality. Derivative indicators of retinal vessels include the Central Retinal Arteriolar Equivalent (CRAE), Central Retinal Venular Equivalent (CRVE), Arteriole-to-Venule Ratio (AVR) that describe vessel diameter, and fractal dimension, curvature tortuosity, and branching angle that describe vascular morphology. Changes in retinal vessel diameter reflect a series of subclinical pathophysiological responses to hyperglycemia, hypertension, inflammation, hypoxia, and endothelial dysfunction. Previous studies have provided the relationship between retinal vascular characteristics and indicators related to coronary artery disease.
[0003] However, in the prior art, disease prediction based on fundus images and fundus vascular characteristics is only a broad-scale prediction of the disease, without in-depth involvement of detailed physical examination indicators. Summary of the invention
[0004] The purpose of the embodiment of the present invention is to provide a bidirectional prediction method based on fundus images and physical examination indicators, which can accurately predict diseases in depth.
[0005] In order to achieve the above-mentioned purpose, an embodiment of the present invention provides a bidirectional prediction method based on fundus images and physical examination indicators, and the bidirectional prediction method based on fundus images and physical examination indicators includes: for multiple target detectors, obtaining fundus images and data of multiple physical examination indicators of each target detector; based on a bidirectional prediction feedback mechanism, determining the correlation between the characteristics of the fundus image and the multiple physical examination indicators, and predicting retinal lesions through the multiple physical examination indicators to obtain a bidirectional prediction result; and based on the bidirectional prediction result, detecting the person to be detected.
[0006] Optionally, the characteristics of the fundus image include diabetic retinopathy DR grade and vascular morphology characteristics, and determining the correlation between the characteristics of the fundus image and the multiple physical examination indicators includes: annotating the fundus image to obtain the DR grade of the fundus image; determining the correlation between different DR grades and the multiple physical examination indicators; extracting the vascular morphology characteristics of the fundus image through a preset vascular segmentation model; determining the correlation between the vascular morphology characteristics and the multiple physical examination indicators; and determining the correlation between the characteristics of the fundus image and the multiple physical examination indicators through the determined correlations between the different DR grades and the multiple physical examination indicators and the determined correlations between the vascular morphology characteristics and the multiple physical examination indicators, so as to determine the first main physical examination indicator among the multiple physical examination indicators.
[0007] Optionally, the preset vascular segmentation model uses the Inception-V3 model as the core architecture, and adaptively adjusts the Inception-V3 model to construct a U-net U-network vascular segmentation model, embeds a spatial pyramid pooling module in the vascular segmentation model, and adds a regional attention mechanism to the vascular segmentation model when constructing the vascular segmentation model.
[0008] Optionally, when training the blood vessel segmentation model, a negative sample loss calculation optimization strategy is added.
[0009] Optionally, the vascular morphological characteristics include an arteriovenous ratio and an arteriovenous width, and determining the correlation between the vascular morphological characteristics and the multiple physical examination indicators includes: determining the correlation between the arteriovenous ratio and the multiple physical examination indicators; and determining the correlation between the arteriovenous width and the multiple physical examination indicators.
[0010] Optionally, the prediction of retinal lesions through the multiple physical examination indicators includes: using a decision tree method to classify the retinal lesions through the multiple physical examination indicators, and determining the importance of each physical examination indicator among the multiple physical examination indicators; and when the importance of the physical examination indicator is greater than a preset threshold, determining a second main physical examination indicator among the multiple physical examination indicators.
[0011] Optionally, the detection of the person to be detected based on the bidirectional prediction result includes: constructing a training set, a verification set and a test set of a preset detection model through the bidirectional prediction result; and completing the training, verification and testing of the preset detection model through the training set, the verification set and the test set to obtain a trained detection model for detecting the person to be detected, and the preset detection model is built using a ConvNext-tiny network.
[0012] An embodiment of the present invention also provides a control device, which includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-mentioned bidirectional prediction method based on fundus images and physical examination indicators.
[0013] An embodiment of the present invention further provides a machine-readable storage medium, on which instructions are stored, and the instructions enable a machine to execute the above-mentioned bidirectional prediction method based on fundus images and physical examination indicators.
[0014] An embodiment of the present invention further provides a bidirectional prediction system, which includes a fundus imaging device and the above-mentioned control device.
[0015] Through the above technical solution, the embodiment of the present invention determines the correlation between the characteristics of the fundus image and multiple physical examination indicators based on the bidirectional prediction feedback mechanism, and predicts retinal lesions through multiple physical examination indicators to obtain a bidirectional prediction result; and based on the bidirectional prediction result, the person to be tested is tested. The embodiment of the present invention performs intelligent prediction of systemic diseases based on the bidirectional prediction feedback mechanism, which can avoid bias caused by data and other issues, improves the reliability and accuracy of intelligent prediction of diseases based on fundus images and systemic physical examination indicators, and provides rich information and evidence for intelligent prediction of diseases.
[0016] Other features and advantages of the embodiments of the present invention will be described in detail in the subsequent detailed description. BRIEF DESCRIPTION OF THE DRAWINGS
[0017] The accompanying drawings are used to provide a further understanding of the embodiments of the present invention and constitute a part of the specification. Together with the following specific implementations, they are used to explain the embodiments of the present invention, but do not constitute a limitation on the embodiments of the present invention. In the accompanying drawings:
[0018] Figure 1 It is a flowchart of a bidirectional prediction method based on fundus images and physical examination indicators provided by an embodiment of the present invention;
[0019] Figure 2 is a flowchart of an example bidirectional prediction method;
[0020] Figure 3 is a schematic diagram of an example of extracting a blood vessel region of a fundus image;
[0021] Figure 4 is a schematic diagram of an example of adding a regional attention mechanism to a blood vessel segmentation model;
[0022] Figure 5 is a schematic diagram of the example negative sample loss calculation optimization strategy;
[0023] Figure 6 is a schematic diagram of an example of determining a first major physical examination indicator;
[0024] Figure 7 is a schematic diagram of an example fundus image; and
[0025] Figure 8 is a schematic diagram of example detection results. DETAILED DESCRIPTION
[0026] The specific implementation of the embodiment of the present invention is described in detail below in conjunction with the accompanying drawings. It should be understood that the specific implementation described here is only used to illustrate and explain the embodiment of the present invention, and is not used to limit the embodiment of the present invention.
[0027] As mentioned above, changes in retinal vessel diameter reflect a series of subclinical pathophysiological responses to hyperglycemia, hypertension, inflammation, hypoxia, and endothelial dysfunction. Previous studies have shown the relationship between retinal vascular characteristics and coronary artery disease-related indicators. However, in the prior art, disease prediction based on fundus images and fundus vascular characteristics is only a large-scale prediction of the disease, without in-depth involvement of detailed physical examination indicators. In particular, the relationship between retinal vascular indicators and various physical examination indicators such as diabetic complications at different degrees of retinopathy severity.
[0028] Taking diabetes as an example, the onset and progression of diabetic retinopathy (DR) and diabetic nephropathy (DN) are closely related, and the severity of DR is related to renal function. In addition, DR is also correlated with cardiovascular disease (CVD), peripheral arterial disease (PAD), and stroke. Considering the possible coexistence and parallel development of multiple diabetic complications, as well as systemic metabolic disorders in diabetic patients, the embodiments of the present invention are intended to study the correlation between retinal vascular characteristics and systemic physical examination indicators in diabetic patients at different stages of diabetic retinopathy.
[0029] Figure 1 is a flowchart of a bidirectional prediction method based on fundus images and physical examination indicators provided by an embodiment of the present invention. Figure 1 , the bidirectional prediction method based on fundus images and physical examination indicators may include the following steps:
[0030] Step S110: For multiple target detectors, obtain fundus images and data of multiple physical examination indicators of each target detector.
[0031] The multiple physical examination indicators preferably in the embodiment of the present invention may include renal function indicators, blood routine indicators, liver function indicators, metabolic control indicators, carotid artery IMT, new diabetes risk markers, physical examination indicators and other related indicators.
[0032] To illustrate by way of example, a number of physical examination indicators can also be referred to as physical indicators, where renal function indicators include, for example, urine albumin, urine creatinine, blood uric acid, etc., routine blood test indicators include, for example, white blood cell count, platelets, hemoglobin, etc., liver function indicators include, for example, albumin, alkaline phosphatase, etc., metabolic control indicators include, for example, triglycerides, fasting plasma glucose, etc., physical examination indicators include, for example, body mass index, waist-to-hip ratio, etc., and other related indicators include, for example, carotid artery IMT, new risk markers for diabetes, C-peptide, high-sensitivity C-reactive protein, etc.
[0033] Step S120: Based on the bidirectional prediction feedback mechanism, the correlation between the characteristics of the fundus image and the multiple physical examination indicators is determined, and the retinal lesions are predicted by the multiple physical examination indicators to obtain a bidirectional prediction result.
[0034] Step S120 may include steps S121-S122: Step S121, determining the correlation between the characteristics of the fundus image and the multiple physical examination indicators; Step S122: predicting retinal lesions using the multiple physical examination indicators.
[0035] The preferred features of the fundus image in the embodiment of the present invention may include diabetic retinopathy DR grades and vascular morphology features.
[0036] Preferably, step S121 may include: annotating the fundus image to obtain the DR grade of the fundus image; determining the correlation between different DR grades and the multiple physical examination indicators; extracting the vascular morphological features of the fundus image through a preset vascular segmentation model; determining the correlation between the vascular morphological features and the multiple physical examination indicators; and determining the correlation between the characteristics of the fundus image and the multiple physical examination indicators through the determined correlations between the different DR grades and the multiple physical examination indicators and the determined correlations between the vascular morphological features and the multiple physical examination indicators, so as to determine the first main physical examination indicator among the multiple physical examination indicators.
[0037] Please refer to Figure 2 , for example, experts label the acquired fundus images of diabetic patients to obtain the DR grading of the fundus images. For example, the labeled categories are no diabetic retinopathy (DR0), mild non-proliferative diabetic retinopathy (DR1), moderate non-proliferative diabetic retinopathy (DR2), severe non-proliferative diabetic retinopathy (DR3), and proliferative diabetic retinopathy (DR4). Based on the DR grading, the correlation between different DR grading and multiple physical examination indicators in the acquired fundus images is calculated or statistically analyzed.
[0038] Next, the vascular morphological features of the fundus image are extracted through a preset vascular segmentation model; and the correlation between the vascular morphological features and multiple physical examination indicators is determined.
[0039] Among them, the correlation (r value) calculation method is used to quantify the strength and direction of the linear relationship between two variables. In medical data analysis, commonly used correlation analysis methods include Pearson correlation coefficient and Spearman's rank correlation coefficient. The value of r can range from -1 to 1, close to 1 indicates a strong positive correlation, close to -1 indicates a strong negative correlation, and close to 0 indicates no correlation.
[0040] The preset blood vessel segmentation model preferably in the embodiment of the present invention uses the Inception-V3 model as the core architecture, and adaptively adjusts the Inception-V3 model to construct a U-net U-net network blood vessel segmentation model. Preferably, a spatial pyramid pooling module is embedded in the blood vessel segmentation model. Preferably, when constructing the blood vessel segmentation model, a regional attention mechanism is added to the blood vessel segmentation model.
[0041] Following the above example, an artificial intelligence model is constructed to perform vascular segmentation of fundus images. The embodiment of the present invention is based on the Inception-V3 model as the core architecture of the preset vascular segmentation model, and adaptively adjusts it to construct a U-shaped structure similar to U-Net. The preset vascular segmentation model not only integrates the powerful feature extraction capability of Inception-V3, but also draws on the excellent performance of U-Net in the field of image segmentation, realizing deep mining and efficient utilization of features. In order to further enhance the model's ability to recognize and segment complex vascular structures, the embodiment of the present invention embeds a spatial pyramid pooling (ASPP) module in the vascular segmentation model. Through multi-scale feature fusion, the model's ability to capture vascular features of different scales in fundus images is effectively improved, thereby significantly improving the accuracy and robustness of vascular segmentation. The vascular segmentation model provided by the embodiment of the present invention can implement a dual-output optimization strategy, which not only improves the flexibility of the segmentation task, but also allows the model to perform detailed analysis of vascular structures at different levels.
[0042] Following the above example, when constructing a vascular segmentation model, the embodiment of the present invention adds a (key) region attention mechanism to the vascular segmentation model. The embodiment of the present invention preferably analyzes the whole body examination index based on the arteriovenous ratio and the arteriovenous width. The arteriovenous ratio and the arteriovenous width can be based on Figure 3 Therefore, in the process of blood vessel segmentation, we can focus on this circular area. Figure 3, using the arteriovenous ratio and arteriovenous width for analysis, these features are obtained through the circular area in the figure, where the inner ring of the circular ring is, for example, 0.5DD, and the outer ring is, for example, 1.0DD (DD: disc diameter). Further, in the training process of the vascular segmentation model, the features of the above-mentioned key focus areas can be added. For example, the features of the vascular key focus areas are extracted and fused with the features extracted from the original image, and then the fused features are used for model training, such as Figure 4 shown.
[0043] Further preferably, when training the blood vessel segmentation model, a negative sample loss calculation optimization strategy may be added.
[0044] Following the above example, when constructing the training samples of the vascular segmentation model and training the vascular segmentation model, since the number difference between positive samples (or positive samples) and negative samples (or negative samples) is relatively large, in order to reduce the decrease in model accuracy caused by too many negative samples, the embodiment of the present invention adds a negative sample loss calculation optimization strategy during the training of the vascular segmentation model. Among them, constructing the loss function may include: 1) performing an expansion operation on the vascular skeleton, and the expansion area may be approximately 4-5 times the width of the vascular. In the area after the expansion operation, the area that does not include the blood vessel is the negative sample area, and the blood vessel area is the positive sample area. Loss calculation is performed on the above-mentioned positive sample area and negative sample area, and then the network parameters are optimized.
[0045] The preferred vascular morphological features of the embodiment of the present invention may include the arteriovenous ratio and the arteriovenous width. Preferably, determining the significant difference between the vascular morphological features and the multiple physical examination indicators may include: determining the correlation between the arteriovenous ratio and the multiple physical examination indicators; and determining the correlation between the arteriovenous width and the multiple physical examination indicators.
[0046] The method for calculating the correlation between the vascular morphological characteristics and the multiple physical examination indicators is as described above and will not be repeated here.
[0047] Please refer to Figure 2 , determine the first major physical examination indicator among multiple physical examination indicators according to the correlation. Figure 3 As shown, through the above content, N physical examination indicators are determined as the first main physical examination indicators among the 37 physical examination indicators.
[0048] Preferably, step S122 may include: using a decision tree method to classify the retinal lesions through the multiple physical examination indicators, and determining the importance of each physical examination indicator in the multiple physical examination indicators; and when the importance of the physical examination indicator is greater than a preset threshold, determining the second main physical examination indicator in the multiple physical examination indicators.
[0049] Please refer to Figure 2 , using an example to illustrate, the decision tree model is trained by supervised learning. Among them, in each sample, multiple physical examination indicators and the classification results corresponding to the sample are trained to obtain a decision tree model, and the decision tree model can classify retinal lesions. It should be noted that the classification of retinal lesions here is the classification of DR or not, which is different from the DR grading method in step S121. Here, the decision tree model is used for (prediction) classification, and step S121 performs DR grading through expert experience. Preferably, through the decision tree model, the importance of each physical examination indicator in the multiple physical examination indicators can also be determined, as described in Table 1. When the importance of the physical examination indicator is greater than a preset threshold (for example, 2%), the second main physical examination indicator among the multiple physical examination indicators is determined, as shown in Table 2.
[0050] Table 1 Importance of each physical examination indicator
[0051]
[0052]
[0053] Table 2 Second major physical examination indicators
[0054]
[0055] Please refer to Figure 2 , continuing with the above example, based on the bidirectional prediction feedback mechanism, according to step S121, determine the correlation between the characteristics of the fundus image (diabetic retinopathy grade, vascular characteristics, etc.) and multiple physical examination indicators to determine the first main physical examination indicator; step S121, classify and predict diabetic retinopathy based on multiple physical examination indicators, and grade the importance coefficients of the whole body physical examination indicators to determine the second main physical examination indicator; based on the above steps S121-S122, the intelligent bidirectional prediction results of the fundus image and the whole body physical examination indicators of diabetic patients can be obtained, as shown in Table 3.
[0056] Table 3 Bidirectional prediction results
[0057] Serial number Bidirectional prediction results 1 Glycated hemoglobin 2 platelets 3 ALT 4 Serum creatinine 5 Serum uric acid 6 Urine microalbumin 7 Fasting plasma glucose 8 Alkaline phosphatase 9 Urine albumin / creatinine ratio 10 Triglycerides 11 High-sensitivity C-reactive protein 12 C-Peptide 13 albumin 14 Fasting insulin
[0058] Step S130: Based on the bidirectional prediction result, the person to be detected is detected.
[0059] Preferably, step S130 may include: constructing a training set, a validation set and a test set of a preset detection model through the bidirectional prediction results; and completing the training, validation and testing of the preset detection model through the training set, validation set and test set to obtain a trained detection model for detecting the person to be detected. The preset detection model is constructed using a ConvNext-tiny network.
[0060] Take an example to illustrate that the reference range of the physical examination index to be analyzed is obtained through the bidirectional prediction results. For example, for a physical examination index: those less than the reference value are regarded as the first category, those within the reference range are regarded as the second category, and those greater than the reference value are regarded as the third category. Then, the training set, validation set, and test set are divided in an 8:1:1 manner. The ConvNext-tiny network is used to build an artificial intelligence algorithm model and test it on the test set. The trained detection model is used to detect the person to be detected, such as Figure 4 and 5 Please refer to Figure 5 ,In the prediction results of the test, the activation area can be viewed and analyzed through the ,class activation map CAM, and the impact of the physical examination index on the ,fundus image and the impact area can be inferred.
[0061] Accordingly, the embodiment of the present invention determines the correlation between the characteristics of the fundus image and multiple physical examination indicators based on a two-way prediction feedback mechanism, and predicts retinal lesions through multiple physical examination indicators to obtain a two-way prediction result; and based on the two-way prediction result, the person to be tested is tested. Compared with predicting systemic diseases only through fundus images, the embodiment of the present invention performs intelligent prediction of systemic diseases based on a two-way prediction feedback mechanism, avoids bias caused by data and other issues, ensures the reliability and accuracy of intelligent prediction of diseases based on fundus images and systemic physical examination indicators, and provides rich information and evidence for intelligent prediction of diseases.
[0062] An embodiment of the present invention also provides a control device, which includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-mentioned bidirectional prediction method based on fundus images and physical examination indicators.
[0063] An embodiment of the present invention further provides a machine-readable storage medium, on which instructions are stored, and the instructions enable a machine to execute the above-mentioned bidirectional prediction method based on fundus images and physical examination indicators.
[0064] An embodiment of the present invention further provides a bidirectional prediction system, which includes a fundus imaging device and the above-mentioned control device.
[0065] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.
[0066] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0067] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.
[0068] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.
[0069] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.
[0070] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.
[0071] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.
[0072] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.
[0073] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included within the scope of the claims of the present application.
Claims
1. A bidirectional prediction method based on fundus images and physical examination indicators, characterized in that: The bidirectional prediction method based on fundus images and physical examination indicators includes: For multiple target detectors, fundus images and data of multiple physical examination indicators of each target detector are obtained; Based on a bidirectional prediction feedback mechanism, determining the correlation between the features of the fundus image and the multiple physical examination indicators, and predicting retinal lesions using the multiple physical examination indicators to obtain a bidirectional prediction result; and Based on the bidirectional prediction result, the person to be detected is detected.
2. The bidirectional prediction method based on fundus images and physical examination indicators according to claim 1 is characterized in that: The features of the fundus image include diabetic retinopathy DR grade and vascular morphology features, and determining the correlation between the features of the fundus image and the multiple physical examination indicators includes: Annotating the fundus image to obtain a DR grade of the fundus image; Determine the correlation between different DR grades and the multiple physical examination indicators; Extracting the vascular morphological features of the fundus image by using a preset vascular segmentation model; Determining the correlation between the vascular morphological characteristics and the multiple physical examination indicators; and The correlation between the characteristics of the fundus image and the multiple physical examination indicators is determined by determining the correlation between the different DR grades and the multiple physical examination indicators and the correlation between the determined vascular morphology characteristics and the multiple physical examination indicators, so as to determine the first main physical examination indicator among the multiple physical examination indicators.
3. The bidirectional prediction method based on fundus images and physical examination indicators according to claim 2 is characterized in that: The preset blood vessel segmentation model uses the Inception-V3 model as the core architecture, and adaptively adjusts the Inception-V3 model to construct a U-net U-net network blood vessel segmentation model. A spatial pyramid pooling module is embedded in the blood vessel segmentation model. When constructing a blood vessel segmentation model, a regional attention mechanism is added to the blood vessel segmentation model.
4. The bidirectional prediction method based on fundus images and physical examination indicators according to claim 3 is characterized in that: When training the blood vessel segmentation model, a negative sample loss calculation optimization strategy is added.
5. The bidirectional prediction method based on fundus images and physical examination indicators according to claim 2, characterized in that: The vascular morphological characteristics include an arteriovenous ratio and an arteriovenous width, and determining the correlation between the vascular morphological characteristics and the multiple physical examination indicators includes: Determining the correlation between the arteriovenous ratio and the plurality of physical examination indicators; and Determine the correlation between the arteriovenous width and the multiple physical examination indicators.
6. The bidirectional prediction method based on fundus images and physical examination indicators according to claim 1, characterized in that: The prediction of retinopathy by using the multiple physical examination indicators includes: Using a decision tree method, classifying the retinopathy according to the multiple physical examination indicators, and determining the importance of each of the multiple physical examination indicators; and When the importance of the physical examination indicator is greater than a preset threshold, a second main physical examination indicator among the multiple physical examination indicators is determined.
7. The bidirectional prediction method based on fundus images and physical examination indicators according to claim 1, characterized in that: The detecting the person to be detected based on the bidirectional prediction result includes: Constructing a training set, a validation set, and a test set of a preset detection model through the bidirectional prediction results; and The training, verification and testing of the preset detection model are completed through the training set, verification set and test set to obtain a trained detection model for detecting the person to be detected. The preset detection model is built using the ConvNext-tiny network.
8. A control device, characterized in that: The control device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the bidirectional prediction method based on fundus images and physical examination indicators according to any one of claims 1 to 7.
9. A machine-readable storage medium, characterized in that: The machine-readable storage medium stores instructions, which enable the machine to execute the bidirectional prediction method based on fundus images and physical examination indicators according to any one of claims 1-7.
10. A bidirectional prediction system, characterized in that: The bidirectional prediction system comprises a fundus imaging device and the control device according to claim 8.