Premature infant retinopathy onset risk prediction method and system

Abstract

The invention relates to the technical field of premature infant disease prediction, in particular to a premature infant retinopathy onset risk prediction method and system, and the method comprises the following steps: S1, clinical data collection: collecting the clinical data of a premature infant; s2, retina image data acquisition: acquiring retina image data of a premature infant, and acquiring microstructure and blood flow state information of retina blood vessels; s3, data preprocessing: preprocessing the acquired retina image data and clinical data; s4, correlation analysis of the retina image and clinical data: revealing potential correlation between the retina image and the health condition of the premature infant, and predicting the occurrence probability of retinopathy; s5, risk assessment and intervention suggestion: dynamically assessing the retinopathy risk of the premature infant, and providing personalized intervention suggestions; according to the method, high-value features in multi-source data are effectively captured, and the prediction accuracy of retinopathy is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of premature infant disease prediction, and particularly to a method and system for predicting the onset risk of retinopathy of prematurity. Background Art

[0002] Retinopathy of prematurity (ROP) is a disease that seriously affects the vision of premature infants. It mainly occurs when the retinal blood vessels of premature infants are not fully developed. Due to factors such as oxygen therapy and low birth weight, abnormal neovascular growth and retinal lesions occur. If this kind of lesion is not intervened in time, it may lead to permanent vision loss or even blindness. Therefore, early screening and precise intervention for retinopathy of prematurity have important clinical significance. However, due to the complex etiology and diverse risk factors of ROP, a single data collection and analysis method is difficult to meet the needs of precise prediction and effective intervention.

[0003] Currently, the screening and diagnosis of ROP mainly rely on manual analysis of retinal images or risk assessment methods based on simple statistical models. However, these methods have significant limitations: First, manual analysis of retinal images relies highly on the experience of doctors and is inefficient, and it is easy to miss the subtle features of early lesions. Second, the risk assessment based on simple statistical models fails to fully integrate retinal images and multi-dimensional clinical data, resulting in weak comprehensive analysis ability of risk factors. In addition, existing technologies are difficult to dynamically evaluate the lesion risk by combining real-time data, nor can they provide precise intervention suggestions for individuals, seriously restricting the effect of early warning and precise treatment of ROP.

[0004] In order to solve the above problems existing in the prior art, the present invention provides a method and system for predicting the onset risk of retinopathy of prematurity, improving the early screening ability and intervention efficiency of retinopathy of prematurity, providing a scientific decision-making basis for clinical practice, and effectively reducing the incidence and severity of retinal lesions. Summary of the Invention

[0005] The present invention provides a method and system for predicting the onset risk of retinopathy of prematurity.

[0006] A method for predicting the onset risk of retinopathy of prematurity includes the following steps: S1, Clinical data collection: Collect the clinical data of premature infants, including birth weight, gestational age, blood oxygen level, respiratory rate, and heart rate; S2, Retinal image data collection: Collect the retinal image data of premature infants to obtain the microscopic structure and blood flow state information of retinal blood vessels; S3, Data preprocessing: Preprocess the collected retinal image data and clinical data, including image denoising and data normalization processing; S4, Association Analysis of Retinal Images and Clinical Data: Perform an association analysis on the preprocessed retinal image data and clinical data to reveal the potential correlation between retinal images and the health status of premature infants and predict the probability of the occurrence of retinopathy; S5, Risk Assessment and Intervention Recommendations: Based on the results of the association analysis of retinal images and clinical data and combined with the clinically monitored data in real time, dynamically assess the risk of retinopathy in premature infants and provide personalized intervention recommendations.

[0007] Optionally, the clinical data collection in S1 includes: S11, Birth Weight Data Collection: Collect the birth weight data of premature infants through the electronic medical record system; S12, Gestational Age Data Collection: Obtain the gestational age data of premature infants through the records of obstetricians; S13, Blood Oxygen Level Data Collection: Collect the blood oxygen level data of premature infants through a continuous monitoring device (finger clip oximeter); S14, Respiratory Rate Data Collection: Record the respiratory rate data of premature infants through a respiratory monitoring device or a manual counting method; S15, Heart Rate Data Collection: Collect the heart rate data of premature infants through an electrocardiogram (ECG) monitoring instrument.

[0008] Optionally, the retinal image data collection in S2 includes: S21, Retinal Image Collection: Use an ophthalmic OCT (optical coherence tomography) device to obtain the retinal image data of premature infants; S22, Microstructure of Retinal Blood Vessels Collection: Based on the obtained retinal image data, obtain the microstructure of retinal blood vessels, including vessel diameter, branches, and curvature, through the imaging system of the ophthalmic device; S23, Retinal Blood Flow Status Monitoring: Use a laser Doppler flowmeter to monitor the retinal blood flow status in real time, including blood flow velocity and blood flow volume.

[0009] Optionally, the data preprocessing in S3 includes: S31, Image Denoising Processing: Use a median filtering algorithm to perform denoising processing on the retinal image data; S32, Normalization Processing: Use a minimum-maximum normalization algorithm to perform normalization processing on the collected clinical data and the retinal image data after image denoising processing.

[0010] Optionally, the association analysis of retinal images and clinical data in S4 includes: S41, Retinal Image Feature Extraction: Use a Convolutional Neural Network (CNN) model to extract retinal image features from the preprocessed retinal image data, including blood vessel diameter, branch density, curvature change, and texture features of the lesion area; S42, Clinical Data Feature Dimensionality Reduction: Use Principal Component Analysis (PCA) to perform dimensionality reduction on the preprocessed clinical data, mapping the multi-dimensional clinical data to a low-dimensional feature space; S43, Image and Clinical Data Association Analysis: Based on the extracted retinal image features and the dimensionality-reduced clinical data, perform association analysis through a Graph Neural Network (GNN) model to reveal the potential association between the retinal image and the health status of premature infants, and predict the occurrence probability of retinopathy.

[0011] Optionally, the retinal image feature extraction in S41 includes: S411, Convolutional Feature Extraction: Perform multi-layer convolutional operations on the preprocessed retinal image data to extract a high-dimensional feature vector containing blood vessel diameter, branch density, curvature change, and texture features; S412, Blood Vessel Diameter Calculation: Based on the results of convolutional feature extraction, detect the blood vessel area through a segmentation algorithm and calculate the diameter of each blood vessel to obtain the statistical quantity of the blood vessel diameter; S413, Branch Density Calculation: Use the binary image after blood vessel segmentation to calculate the ratio between the number of blood vessel branches and the total area, and define the branch density; S414, Curvature Change Calculation: Based on the results of convolutional feature extraction, calculate the local curvature of the blood vessels; S415, Lesion Area Texture Feature Analysis and Classification: Classify the output high-dimensional feature vector through a fully connected layer to identify the texture characteristics (bleeding points, microvascular dilation) of the lesion area.

[0012] Optionally, the clinical data feature dimensionality reduction in S42 includes: S421, Data Standardization: Perform standardization on the preprocessed multi-dimensional clinical data; S422, Covariance Matrix Calculation: Calculate the covariance matrix of the standardized clinical data; S423, Eigenvalue Decomposition: Perform eigenvalue decomposition on the covariance matrix to obtain eigenvalues and eigenvectors; S424, Dimensionality Reduction Mapping: Select the eigenvectors corresponding to the largest eigenvalues to form a dimensionality reduction transformation matrix , and project the standardized clinical data into a low-dimensional space.

[0013] Optionally, the image and clinical data association analysis in S43 includes: S431, Data graph construction: Based on the extracted retinal image features and the dimension-reduced clinical data, construct a data graph , where is the node set, is the edge set; S432, Graph convolution operation: Perform message passing and feature update on the data graph; S433, Output classification layer: After the graph convolution operation, finally obtain the high-dimensional feature representation of each node , and input it into the fully connected layer and the Softmax activation function to output the occurrence probability of retinopathy of prematurity.

[0014] Optionally, the risk assessment and intervention suggestions in S5 include: S51, Dynamic risk score calculation: Based on the results of the correlation analysis of retinal image data and clinical data and the clinically monitored data in real time, dynamically update the risk score of retinopathy of prematurity ; S52, Risk level classification: Compare the risk score with the preset upper risk threshold , lower risk threshold . When , it indicates that the risk level is low risk. When , it indicates that the risk level is medium risk. When , it indicates that the risk level is high risk; S53, Generation of personalized intervention suggestions: According to the classified risk levels, generate personalized intervention suggestions, specifically including: When the risk level is low risk, it is recommended to perform regular monitoring and maintain the intervention; When the risk level is medium risk, it is recommended to increase the monitoring frequency and adjust the treatment plan (such as optimizing oxygen therapy parameters); When the risk level is high risk, it is recommended to take intensive intervention measures, including emergency medical treatment or laser treatment.

[0015] A system for predicting the onset risk of retinopathy of prematurity, used to implement the above-mentioned method for predicting the onset risk of retinopathy of prematurity, includes the following modules: Clinical data acquisition module: Acquire the clinical data of premature infants, including birth weight, gestational age, blood oxygen level, respiratory rate, heart rate; Retinal image data acquisition module: Acquire the retinal image data of premature infants to obtain the microstructural and blood flow state information of retinal blood vessels; Data preprocessing module: Preprocess the acquired retinal image data and clinical data, including image denoising and data normalization processing; Data Association Analysis Module: Based on the preprocessed retinal image data and clinical data, through association analysis methods, it reveals the potential correlation between retinal images and the health status of premature infants, and predicts the probability of the occurrence of retinopathy. Risk Assessment and Intervention Module: Combining the results of data association analysis and real-time monitored clinical data, it dynamically assesses the risk of retinopathy in premature infants and provides personalized intervention suggestions.

[0016] Advantages of the present invention: In the present invention, through the clinical data acquisition module and the retinal image data acquisition module, comprehensive health information of premature infants is obtained, including clinical indicators such as birth weight, gestational age, blood oxygen level, as well as retinal image data, vascular microstructure, and blood flow state information. Combining image denoising and data normalization ensures the high quality and consistency of the data, effectively removes noise, retains the edge features and detailed information of retinal blood vessels, provides a reliable basis for the fusion analysis of multi-source data, and significantly improves the accuracy and efficiency of subsequent prediction models.

[0017] In the present invention, through retinal image feature extraction, clinical data feature dimensionality reduction, and association analysis of images and clinical data, deep fusion of retinal images and clinical data is achieved, accurately revealing the potential correlation between retinopathy and health status. Using convolutional neural networks, principal component analysis, and graph neural networks, high-value features in multi-source data are effectively captured, significantly improving the prediction accuracy of retinopathy, and providing strong technical support for the early screening of retinopathy in premature infants.

[0018] In the present invention, through dynamic risk score calculation combined with real-time monitored clinical data, the risk of retinopathy in premature infants is dynamically assessed, and the risk levels are divided based on scientifically set thresholds to ensure the timeliness and reliability of the assessment results. According to the divided risk levels, personalized intervention suggestions are generated, including measures such as regular monitoring, optimized treatment plans, and emergency interventions, effectively reducing the incidence of lesions and optimizing the treatment effect. Description of the Drawings

[0019] To more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only those of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0020] Figure 1 It is a schematic flowchart of the prediction method for the embodiments of the present invention; Figure 2 It is a schematic diagram of the system function modules for the embodiments of the present invention. Detailed implementation manners

[0021] The present invention will be described in detail below in conjunction with the accompanying drawings and specific embodiments. At the same time, it should be noted here that in order to make the embodiments more detailed, the following embodiments are the best and preferred embodiments. For some well-known technologies, those skilled in the art can also adopt other alternative methods for implementation; moreover, the accompanying drawings are only for more specifically describing the embodiments, and are not intended to specifically limit the present invention.

[0022] It should be pointed out that in the specification, when referring to "one embodiment", "embodiment", "exemplary embodiment", "some embodiments", etc., it indicates that the described embodiment may include specific features, structures or characteristics, but not necessarily every embodiment includes such specific features, structures or characteristics. In addition, when combining an embodiment to describe a specific feature, structure or characteristic, implementing such a feature, structure or characteristic in combination with other embodiments (whether explicitly described or not) should be within the knowledge scope of those skilled in the relevant art.

[0023] Generally, terms can be understood at least in part from their use in context. For example, at least in part depending on the context, the term "one or more" used herein can be used to describe any feature, structure or characteristic in a singular sense, or can be used to describe a combination of features, structures or characteristics in a plural sense. Additionally, the term "based on" can be understood as not necessarily intended to convey a set of exclusive factors, but rather, at least in part depending on the context, can alternatively allow for the existence of other factors that are not necessarily explicitly described.

[0024] As Figure 1 shown, a method for predicting the risk of retinopathy of prematurity includes the following steps: S1, Clinical data collection: Collect the clinical data of premature infants, including birth weight, gestational age, blood oxygen level, respiratory rate, and heart rate; S2, Retinal image data collection: Collect the retinal image data of premature infants to obtain the microstructural and blood flow state information of retinal blood vessels; S3, Data preprocessing: Preprocess the collected retinal image data and clinical data, including image denoising and data normalization; S4, Correlation analysis of retinal images and clinical data: Conduct correlation analysis on the preprocessed retinal image data and clinical data to reveal the potential correlation between retinal images and the health status of premature infants, and predict the occurrence probability of retinopathy; S5, Risk assessment and intervention suggestions: Based on the results of the correlation analysis of retinal images and clinical data, combined with the clinically monitored data in real time, dynamically evaluate the risk of retinopathy in premature infants and provide personalized intervention suggestions; Through the above content, the probability of retinopathy can be accurately predicted, and the risk can be dynamically assessed based on real-time data, providing personalized intervention recommendations for premature infants. This can improve the early warning ability of retinopathy of prematurity and timely detect potential risks, thereby providing a scientific basis for clinical practice, effectively reducing the incidence of the disease, and optimizing the treatment effect.

[0025] Clinical data collection in S1 includes: S11, Birth weight data collection: Birth weight data of preterm infants were collected through the electronic medical record system and obtained through delivery records or weighing equipment at birth; S12, gestational age data collection: gestational age data of premature infants were obtained through obstetrician records; S13, blood oxygen level data collection: blood oxygen level data of premature infants were collected through continuous monitoring equipment (finger clip oximeter); S14, respiratory rate data collection: respiratory rate data of premature infants were recorded by respiratory monitoring equipment or manual counting method; S15, heart rate data collection: Collect heart rate data of premature infants through electrocardiogram (ECG) monitoring equipment; Through the above content, the diversity and accuracy of data collection are ensured, which not only comprehensively covers the main risk factors of retinopathy of prematurity, but also provides high-quality basic data support for subsequent data analysis and risk assessment, significantly improving the reliability and practicality of the data.

[0026] Retinal image data collection in S2 includes: S21, retinal image acquisition: using ophthalmic OCT (optical coherence tomography) equipment to obtain retinal image data of premature infants; S22, retinal vascular microstructure acquisition: Based on the acquired retinal image data, the microstructure of retinal blood vessels, including vessel diameter, branching, and curvature, is acquired through the imaging system of the ophthalmic equipment; S23, retinal blood flow status monitoring: using laser Doppler blood flow meter to monitor retinal blood flow status in real time, including blood flow velocity and blood flow; Through the above content, the image data, vascular microstructure and blood flow status information of the premature infant's retina can be obtained comprehensively and accurately, which not only ensures the acquisition of high-resolution images, but also can quantitatively analyze the health status and blood flow of retinal blood vessels. By timely capturing tiny vascular changes and blood flow abnormalities, potential retinal disease risks can be detected early, thereby providing a scientific basis for personalized intervention and treatment plans.

[0027] Data preprocessing in S3 includes: S31, image denoising: the retinal image data is denoised using a median filtering algorithm, expressed as: ; wherein, is the pixel value of the denoised image, is the radius of the filtering window, is the pixel position within the window, is the condition for the neighborhood range; S32, normalization processing: The minimum-maximum normalization algorithm is used to normalize the collected clinical data and the retinal image data after image denoising, expressed as: ; wherein, is the original data, and are the minimum and maximum values in the dataset respectively, is the data value after normalization; Through the above content, while effectively removing noise, the edge features and detailed information of the retinal blood vessels are retained. At the same time, the minimum-maximum normalization is used to standardize the clinical data and image data, eliminating the differences in different data dimensions and measurement units, improving the accuracy and consistency of multi-source data fusion analysis, and significantly enhancing the performance of the retinal disease risk prediction model, providing a reliable data basis for the accurate prediction and personalized intervention of the disease risk in premature infants.

[0028] The correlation analysis of the retinal image and clinical data in S4 includes: S41, retinal image feature extraction: The convolutional neural network (CNN) model is used to extract retinal image features from the preprocessed retinal image data, including blood vessel diameter, branch density, curvature change, and texture features of the lesion area; S42, clinical data feature dimensionality reduction: The principal component analysis (PCA) is used to reduce the dimensionality of the preprocessed clinical data, mapping the multi-dimensional clinical data to a low-dimensional feature space; S43, image and clinical data correlation analysis: Based on the extracted retinal image features and the dimensionality-reduced clinical data, the graph neural network (GNN) model is used for correlation analysis to reveal the potential correlation between the retinal image and the health status of premature infants, and predict the occurrence probability of retinal diseases; Through the above content, the complexity problem of image and clinical data fusion analysis is effectively solved. It can not only automatically extract and screen high-value feature information, but also reveal the deep correlation between the retinal image and the health status, and accurately predict the occurrence probability of retinal diseases, significantly improving the efficiency of data processing and the accuracy of prediction, providing strong technical support for the early screening and personalized intervention of premature infant retinal diseases.

[0029] The retinal image feature extraction in S41 includes: S411, Convolutional feature extraction: Perform multi-layer convolutional operations on the preprocessed retinal image data to extract a high-dimensional feature vector containing blood vessel diameter, branch density, curvature change, and texture features, expressed as: ; Among them, is the eigenvalue extracted by the th convolutional kernel at the image position , is the weight matrix of the th convolutional kernel, is the pixel value at the corresponding position of the input image within the convolutional window, , respectively represent the row and column offsets of the convolutional kernel in the image, is the bias of the th convolutional kernel, is the ReLU activation function; S412, Blood vessel diameter calculation: Based on the results of convolutional feature extraction, detect the blood vessel area through a segmentation algorithm and calculate the diameter of each blood vessel to obtain the statistical quantity of the blood vessel diameter, expressed as: ; Among them, is the average blood vessel diameter, is the diameter of the th sampling point in the blood vessel area, is the number of sampling points; S413, Branch density calculation: Use the binary image after blood vessel segmentation to calculate the ratio between the number of blood vessel branches and the total area, and define the branch density, expressed as: ; Among them, is the branch density, is the number of detected blood vessel branches, is the total area in the retinal image; S414, Curvature change calculation: Based on the results of convolutional feature extraction, calculate the local curvature of the blood vessel to characterize the lesion-related features, expressed as: ; Among them, is the blood vessel curvature, , are the first-order derivatives of the blood vessel curve in the coordinate system, , are the second-order derivatives of the blood vessel curve in the coordinate system; S415, Texture Feature Analysis and Classification of Lesion Regions: Classify the output high-dimensional feature vectors through a fully connected layer to identify the texture characteristics (bleeding points, microvascular dilation) of the lesion regions, specifically including: Linear transformation of the fully connected layer: ; Among them, is the class score vector, is the flattened convolutional feature, is the weight matrix, is the bias vector; Softmax activation function: ; Among them, is the predicted probability of the th class, is the score of the th class, is the score of the th class, is the total number of classes; Class determination: ; Among them, is the predicted class label, means to find the index that makes the largest, that is, among all possible classes, select the class with the highest probability as the prediction result; Through the above content, key features in retinal images are automatically extracted, including blood vessel diameter, branch density, curvature change, and texture features of lesion regions, avoiding the limitations of traditional manual feature extraction. It not only improves the accuracy and robustness of feature extraction but also ensures the efficiency and accuracy of classification and risk prediction. Through multi-level feature analysis and classification, it provides a scientific basis for the early diagnosis and precise treatment of retinal diseases.

[0030] The dimensionality reduction of clinical data features in S42 includes: S421, Data standardization: Standardize the preprocessed multi-dimensional clinical data, expressed as: ; Among them, is the th feature value of the th sample, is the mean of the th feature, is the standard deviation of the th feature, is the standardized feature value; S422, Covariance Matrix Calculation: Calculate the covariance matrix of the clinically standardized data, expressed as: ; where is the covariance matrix, is the number of samples, is the th standardized eigenvector of the sample, is the sample mean vector; S423, Eigenvalue Decomposition: Perform eigenvalue decomposition on the covariance matrix to obtain eigenvalues and eigenvectors, expressed as: ; where is the result of the linear transformation of the covariance matrix on the eigenvector , is the th eigenvalue, representing the importance of the feature, is the corresponding eigenvector, representing the direction after dimensionality reduction; S424, Dimensionality Reduction Mapping: Select the eigenvectors corresponding to the largest eigenvalues to form the dimensionality reduction transformation matrix , and project the standardized clinical data into the low-dimensional space, expressed as: ; where is the eigenvector after dimensionality reduction of the th sample, is the dimensionality reduction matrix composed of eigenvectors, is the standardized eigenvector of the th sample; Through the above, not only is the data dimension effectively reduced, but also the key information most relevant to the risk of retinopathy is retained. At the same time, redundant features and noise are eliminated, significantly improving the data processing efficiency, reducing the model complexity, and avoiding the problem of information redundancy caused by the correlation of multi-dimensional features. Through the expression of the dimensionality-reduced feature space, it can provide more concise and efficient data support for the accurate prediction of retinopathy, ensuring that the model has higher robustness and practicality while guaranteeing the prediction accuracy.

[0031] The image and clinical data association analysis in S43 includes: S431, Data Graph Construction: Based on the extracted retinal image features and the clinically data after dimensionality reduction, construct the data graph , where is the node set, each node represents a sample, and the node features are composed of the combination of retinal image features and the clinically data after dimensionality reduction. is the edge set. The edges between nodes are established based on the similarity of samples and are calculated by cosine similarity, expressed as: ; where is the similarity weight of nodes and . , are the feature vectors of nodes and respectively; S432, graph convolution operation: Message passing and feature update are performed on the data graph. The feature update of each layer is expressed as: ; where is the feature representation of node at the -th layer, is the feature representation of node at the -th layer, is the set of neighbor nodes of node , is the edge weight between nodes and , is the normalization factor used to standardize the contribution of neighbor nodes to node , is the weight matrix of the -th layer, is the bias vector of the -th layer, is the ReLU activation function; S433, output classification layer: After the graph convolution operation, the high-dimensional feature representation of each node is finally obtained and input into the fully connected layer and the Softmax activation function. The occurrence probability of retinopathy is output, expressed as: ; where is the probability that node belongs to retinopathy, is the classification score of node , calculated by the fully connected layer, is the classification score of node , is the total number of classes in the classification task; ; where is the weight matrix of the fully connected layer, is the bias vector; Through the above, the deep - level correlation analysis of image features and clinical data is realized. It can not only capture potential patterns that are difficult to identify in a single data source, but also dynamically learn the interaction relationships of multi - source data through the message - passing mechanism of graph convolution, thereby more accurately predicting the occurrence probability of retinopathy, significantly improving the depth of data fusion and the accuracy of prediction, and providing strong technical support for the early warning and personalized intervention of retinopathy of prematurity.

[0032] The risk assessment and intervention suggestions in S5 include: S51, dynamic risk score calculation: Based on the results of the correlation analysis of retinal image data and clinical data and the clinically monitored data in real - time, dynamically update the risk score of retinopathy of prematurity , expressed as: ; Among them, is the dynamic risk score, is the lesion probability obtained from the correlation analysis of retinal image data and clinical data, is the risk factor assessment function of real - time clinical data, outputting the real - time impact on the lesion risk, 、 are the weight coefficients; ; Among them, is the clinically monitored data in real - time, is the weight of each clinical data item for risk assessment, is the th clinical data, is the total number of clinical data; S52, risk level classification: Compare the risk score with the preset upper risk threshold and the lower risk threshold . When , it indicates that the risk level is low - risk. When , it indicates that the risk level is medium - risk. When , it indicates that the risk level is high - risk; The upper risk threshold and the lower risk threshold are set based on historical data, specifically including: Collect historical risk score data: Collect a large number of evaluated risk scores of premature infants from historical data ; Calculate distribution statistical parameters: Calculate the mean and the standard deviation of the historical risk score data, expressed as: ; ; Among them, is the th historical risk score, is the mean of historical data, is the standard deviation of historical data, is the number of historical samples; Determine the risk threshold: Set the threshold using statistical methods 、 , expressed as: ; ; Among them, 、 are adjustable coefficients; S53, Generate personalized intervention suggestions: Generate personalized intervention suggestions according to the divided risk levels, specifically including: When the risk level is low risk, it is recommended to monitor regularly and maintain the intervention; When the risk level is medium risk, it is recommended to increase the monitoring frequency and adjust the treatment plan (such as optimizing oxygen therapy parameters); When the risk level is high risk, it is recommended to take intensive intervention measures, including emergency medical treatment or laser treatment; Through the above content, dynamically calculate the risk score of retinopathy of prematurity, and divide the risk level according to the set threshold, which can not only reflect the dynamic changes of the condition, but also provide personalized intervention suggestions, so as to optimize the clinical treatment plan. The threshold set based on historical data ensures the scientificity and reliability of the risk level division, and further improves the accuracy of risk assessment and the operability of intervention suggestions.

[0033] Such as Figure 2 shown, a system for predicting the onset risk of retinopathy of prematurity, used to implement the above-mentioned method for predicting the onset risk of retinopathy of prematurity, includes the following modules: Clinical data collection module: Collect the clinical data of premature infants, including birth weight, gestational age, blood oxygen level, respiratory rate, heart rate; Retinal image data collection module: Collect the retinal image data of premature infants, and obtain the microscopic structure and blood flow state information of the retinal blood vessels; Data preprocessing module: Preprocess the collected retinal image data and clinical data, including image denoising and data normalization processing; Data Association Analysis Module: Based on the preprocessed retinal image data and clinical data, through association analysis methods, it reveals the potential correlation between retinal images and the health status of premature infants, and predicts the probability of the occurrence of retinopathy; Risk Assessment and Intervention Module: Combining the results of data association analysis and real-time monitored clinical data, it dynamically assesses the risk of retinopathy in premature infants and provides personalized intervention suggestions.

[0034] This invention covers any alternatives, modifications, equivalent methods, and solutions made within the spirit and scope of this invention. To enable the public to have a thorough understanding of this invention, specific details are described in detail in the following preferred embodiments of this invention. However, those skilled in the art can fully understand this invention even without the description of these details. Additionally, well-known methods, processes, procedures, components, and circuits, etc. are not described in detail to avoid unnecessary confusion to the essence of this invention.

[0035] The above are only the preferred embodiments of this invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of this invention, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of this invention.

Claims

1. A method for predicting the risk of retinopathy of prematurity, characterized in that: The following steps are involved: S1, clinical data collection: clinical data of premature infants were collected, including birth weight, gestational age, blood oxygen level, respiratory rate, and heart rate; S2, retinal image data acquisition: collect retinal image data of premature infants to obtain the microstructure and blood flow status information of retinal blood vessels; S3, data preprocessing: preprocessing the collected retinal image data and clinical data, including image denoising and data normalization; S4, correlation analysis between retinal images and clinical data: correlation analysis was performed on the preprocessed retinal image data and clinical data to reveal the potential correlation between retinal images and the health status of premature infants and predict the probability of retinal disease; S5, Risk assessment and intervention recommendations: Based on the results of correlation analysis of retinal images and clinical data, combined with real-time monitored clinical data, the risk of retinopathy in premature infants is dynamically assessed and personalized intervention recommendations are provided.

2. A method for predicting the risk of retinopathy of prematurity according to claim 1, characterized in that: The clinical data collection in S1 includes: S11, Birth weight data collection: Birth weight data of premature infants were collected through the electronic medical record system; S12, gestational age data collection: gestational age data of premature infants were obtained through obstetrician records; S13, blood oxygen level data collection: Collect blood oxygen level data of premature infants through continuous monitoring equipment; S14, respiratory rate data collection: respiratory rate data of premature infants were recorded by respiratory monitoring equipment or manual counting method; S15, heart rate data collection: Collect the heart rate data of premature infants through electrocardiogram monitoring equipment.

3. A method for predicting the risk of retinopathy of prematurity according to claim 1, characterized in that: The retinal image data acquisition in S2 includes: S21, retinal image acquisition: using ophthalmic OCT equipment to obtain retinal image data of premature infants; S22, retinal vascular microstructure acquisition: Based on the acquired retinal image data, the microstructure of retinal blood vessels, including vessel diameter, branching, and curvature, is acquired through the imaging system of the ophthalmic equipment; S23, retinal blood flow status monitoring: A laser Doppler blood flow meter is used to monitor the retinal blood flow status in real time, including blood flow velocity and blood volume.

4. The method for predicting the risk of retinopathy of prematurity according to claim 1, characterized in that: The data preprocessing in S3 includes: S31, image denoising: using median filtering algorithm to denoise the retinal image data; S32, normalization processing: using a minimum-maximum normalization algorithm to normalize the collected clinical data and the retinal image data after image denoising.

5. The method for predicting the risk of retinopathy of prematurity according to claim 1, characterized in that: The retinal image and clinical data association analysis in S4 includes: S41, retinal image feature extraction: a convolutional neural network model is used to extract retinal image features from preprocessed retinal image data, including vessel diameter, branch density, curvature change, and texture features of the lesion area; S42, clinical data feature dimensionality reduction: use principal component analysis to reduce the dimensionality of preprocessed clinical data and map multi-dimensional clinical data into a low-dimensional feature space; S43, Association analysis between images and clinical data: Based on the extracted retinal image features and the clinical data after dimensionality reduction, association analysis is performed through a graph neural network model to reveal the potential correlation between retinal images and the health status of premature infants and predict the probability of retinal lesions.

6. A method for predicting the risk of retinopathy of prematurity according to claim 5, characterized in that: The retinal image feature extraction in S41 includes: S411, convolution feature extraction: performing multi-layer convolution operations on the preprocessed retinal image data to extract high-dimensional feature vectors including vascular diameter, branch density, curvature change and texture features; S412, blood vessel diameter calculation: based on the result of convolution feature extraction, the blood vessel region is detected by a segmentation algorithm, and the diameter of each blood vessel is calculated to obtain the statistics of the blood vessel diameter; S413, branch density calculation: using the binary image after blood vessel segmentation, calculate the ratio between the number of blood vessel branches and the total area, and define the branch density; S414, curvature change calculation: based on the result of convolution feature extraction, calculate the local curvature of the blood vessel; S415, Texture feature analysis and classification of lesion area: Classify the output high-dimensional feature vector through the fully connected layer to identify the texture characteristics of the lesion area.

7. A method for predicting the risk of retinopathy of prematurity according to claim 6, characterized in that: The clinical data feature dimensionality reduction in S42 includes: S421, data standardization: standardize the preprocessed multidimensional clinical data; S422, covariance matrix calculation: calculating the covariance matrix of the standardized clinical data; S423, eigenvalue decomposition: performing eigenvalue decomposition on the covariance matrix to obtain eigenvalues ​​and eigenvectors; S424, Dimensionality reduction mapping: Select the largest The eigenvalues ​​corresponding to the eigenvectors constitute the dimension reduction transformation matrix , projecting the standardized clinical data into a low-dimensional space.

8. A method for predicting the risk of retinopathy of prematurity according to claim 7, characterized in that: The image and clinical data association analysis in S43 includes: S431, Data Graph Construction: Construct a data graph based on the extracted retinal image features and the clinical data after dimensionality reduction ,in, is a node set, is the edge set; S432, graph convolution operation: message passing and feature update on the data graph; S433, output classification layer: After the graph convolution operation, the high-dimensional feature representation of each node is finally obtained. , input to the fully connected layer and Softmax activation function, and output the probability of retinal lesions.

9. A method for predicting the risk of retinopathy of prematurity according to claim 8, characterized in that: The risk assessment and intervention recommendations in S5 include: S51, Dynamic risk score calculation: Dynamically update the risk score of retinopathy of prematurity based on the results of correlation analysis of retinal image data and clinical data and real-time monitored clinical data ; S52, Risk level classification: Risk scoring The upper limit of the preset risk threshold , Risk threshold lower limit For comparison, When , it means the risk level is low risk. When , it means the risk level is medium risk. When , it indicates that the risk level is high risk; S53, generation of personalized intervention suggestions: Generate personalized intervention suggestions based on the risk levels, including: When the risk level is low, regular monitoring and maintenance intervention are recommended; When the risk level is medium, it is recommended to increase the monitoring frequency and adjust the treatment plan; When the risk level is high, intensive intervention is recommended, including emergency medical attention or laser treatment.

10. A system for predicting the risk of retinopathy of prematurity, used to implement a method for predicting the risk of retinopathy of prematurity as claimed in any one of claims 1 to 9, characterized in that: Includes the following modules: Clinical data collection module: collects clinical data of premature infants, including birth weight, gestational age, blood oxygen level, respiratory rate, and heart rate; Retinal image data acquisition module: collects retinal image data of premature infants to obtain the microstructure and blood flow status information of retinal blood vessels; Data preprocessing module: preprocess the collected retinal image data and clinical data, including image denoising and data normalization; Data association analysis module: Based on the preprocessed retinal image data and clinical data, the association analysis method is used to reveal the potential correlation between retinal images and the health status of premature infants, and to predict the probability of retinal lesions; Risk assessment and intervention module: Combines the results of data association analysis with real-time monitored clinical data to dynamically assess the risk of retinopathy in premature infants and provide personalized intervention recommendations.

Images