AI-assisted postoperative complication warning and rehabilitation guidance method and system for glaucoma

Abstract

This invention discloses an AI-assisted method and system for early warning and rehabilitation guidance of postoperative glaucoma complications, belonging to the field of postoperative glaucoma rehabilitation guidance technology. The method includes: collecting multi-dimensional time-series data on postoperative intraocular pressure, retinal nerve fiber layer thickness, and ocular surface inflammation through a retinal nerve intelligent monitoring and analysis platform; obtaining intraocular pressure feature vectors, quantified results of nerve fiber layer integrity, and inflammation prediction data through feature extraction and evaluation model analysis; fusing these data to construct a complication risk assessment matrix; and generating targeted early warning information and phased rehabilitation guidance plans by combining a personalized parameter library. This method and system achieve comprehensive integration and dynamic evaluation of multi-dimensional data, improving the accuracy and timeliness of complication early warning, meeting patients' personalized rehabilitation needs, optimizing clinical postoperative management processes, reducing the incidence of complications, and ensuring the long-term effectiveness of surgical treatment.

Description

Technical Field

[0001] This invention relates to the field of postoperative rehabilitation guidance technology for glaucoma, and in particular to an AI-assisted method and system for early warning and rehabilitation guidance of postoperative glaucoma complications. Background Technology

[0002] Glaucoma, an irreversible cause of blindness, requires long-term monitoring of the eye after surgery to mitigate the risk of complications. However, the complexity of postoperative physiological changes, such as fluctuations in intraocular pressure, progression of nerve fiber layer damage, and inflammatory responses, makes real-time and accurate assessment difficult with traditional monitoring methods. With the advancement of medical intelligence, the clinical demand for personalized and efficient postoperative management plans is increasingly urgent. There is a pressing need to develop an auxiliary system that can integrate multi-dimensional ocular data and dynamically track the recovery process, providing precise decision support for medical staff and scientific rehabilitation guidance for patients, thereby reducing the incidence of complications and ensuring the effectiveness of surgical treatment.

[0003] Existing technologies have two significant shortcomings: First, they lack the ability to systematically integrate multi-dimensional postoperative physiological data, often monitoring only a single indicator, which fails to fully reflect the correlation between eye recovery and the occurrence of complications, resulting in one-sided assessment results and insufficient timeliness of early warnings. Second, rehabilitation guidance programs are mostly general suggestions that do not fully take into account individual differences in patient recovery, basic eye conditions, and other factors, making it difficult to meet the personalized needs of different patients, thereby affecting rehabilitation outcomes and failing to effectively adapt to diverse postoperative management scenarios in clinical practice. Summary of the Invention

[0004] In order to overcome the shortcomings and deficiencies of existing technologies, this invention provides an AI-assisted method and system for early warning and rehabilitation guidance of postoperative complications in glaucoma.

[0005] The technical solution adopted in this invention is an AI-assisted method for early warning and rehabilitation guidance of postoperative glaucoma complications, comprising the following steps: S1, collecting dynamic intraocular pressure data, retinal nerve fiber layer thickness distribution data, and ocular surface inflammation-related physiological indicator data at different postoperative time points through a retinal nerve intelligent monitoring and analysis platform to construct a multi-dimensional time-series dataset; S2, using an intraocular pressure time-series feature extraction algorithm to mine trend features, fluctuation features, and mutation features of the intraocular pressure data to generate a high-dimensional intraocular pressure feature vector; S3, inputting the retinal nerve fiber layer thickness data into a retinal nerve fiber layer assessment model to obtain the quantification results of nerve fiber layer integrity and damage progression rate parameters; S4, using a postoperative inflammation response prediction model to analyze ocular surface inflammation indicator data and output inflammation severity grading results and inflammation regression trend prediction values; S5, using the retinal nerve intelligent monitoring and analysis platform to integrate the intraocular pressure feature vector, nerve fiber layer assessment results, and inflammation prediction data to construct a complication risk assessment matrix; S6, based on the risk assessment matrix and combined with a personalized parameter library for AI-assisted early warning and rehabilitation guidance of postoperative glaucoma complications, generating targeted complication early warning information and a phased rehabilitation guidance plan.

[0006] Furthermore, the expression for the intraocular pressure temporal feature extraction algorithm is as follows: , in, This represents the temporal comprehensive characteristic value of intraocular pressure. The trend feature weighting coefficient, Let be the intraocular pressure value at time t. This represents the total duration of data collection. The time decay coefficient, This is the baseline time point after surgery. The fluctuation characteristic weighting coefficient, To detect the number of data points, Let i be the intraocular pressure value from the i-th measurement. Mean intraocular pressure Let be the credibility weight of the i-th detection data.

[0007] Furthermore, the expression for the retinal nerve fiber layer assessment model is as follows: , in, This is a quantitative value for the integrity of the nerve fiber layer. For thickness weighting coefficient, To determine the number of zones in the nerve fiber layer, This represents the thickness value of the nerve fiber layer in the j-th region. Let be the functional importance coefficient of the j-th partition. For zone attenuation coefficient, Let be the blood vessel distribution density of the j-th partition. The progress rate weighting coefficient, The difference in thickness between two consecutive measurements. This represents the time interval between adjacent detections.

[0008] Furthermore, the expression for the postoperative inflammatory response prediction model is as follows: , in, This is a predictive value for the degree of inflammation. The overall weight of inflammatory markers, For the number of inflammation detection indicators, The value of the k-th inflammatory marker is... Let be the inflammation correlation coefficient of the k-th indicator. This is the coefficient for the rate of inflammation resolution. This is the peak time of postoperative inflammation. This is the weighting coefficient for inflammation fluctuations. This represents the mean of inflammatory markers.

[0009] Further, step S1 includes the following sub-steps: S11, using the intraocular pressure sensing module of the retinal nerve intelligent monitoring and analysis platform, intraocular pressure data is collected from 1 day to 90 days post-surgery at preset time intervals, and the timestamp and detection location information of each collection are recorded; S12, using the optical coherence tomography module of the platform, the thickness data of the retinal nerve fiber layer in the four quadrants of the nasal, temporal, superior, and inferior sides are obtained, and a layer thickness distribution map is generated; S13, using the ocular surface analyzer of the platform, tear secretion, corneal fluorescein staining degree, conjunctival hyperemia index, and inflammation-related physiological indicators are collected to form a standardized indicator dataset; S14, the intraocular pressure data, nerve fiber layer thickness data, and inflammation indicator data are aligned by timestamp to construct a time-series dataset including basic patient information, detection time, and multi-dimensional physiological data.

[0010] Further, step S2 includes the following sub-steps: S21, performing time-series segmentation processing on the intraocular pressure data, dividing it into three stages: postoperative days 1-7, 8-30, and 31-90, and extracting data for each stage; S22, using the sliding window method to extract trend features from the intraocular pressure data of each stage, calculating the rate of change of the mean intraocular pressure and the slope of the linear fit within the window; S23, calculating the standard deviation, coefficient of variation, and range of the intraocular pressure data of each stage through statistical analysis methods, and extracting fluctuation features; S24, using an outlier detection algorithm to identify abrupt changes in the intraocular pressure data, calculating the difference in intraocular pressure before and after the abrupt change and the frequency of the abrupt change, and extracting abrupt change features; S25, concatenating the trend features, fluctuation features, and abrupt change features dimensionally to generate a high-dimensional intraocular pressure feature vector.

[0011] Further, S3 includes the following sub-steps: S31, performing zonal calibration on the retinal nerve fiber layer thickness data, and determining the standard range and coordinate mapping relationship of each detection zone based on anatomical landmarks; S32, inputting the calibrated thickness data of each zone into the retinal nerve fiber layer assessment model, and calculating the thickness compliance rate and deviation rate from the preoperative baseline value of each zone; S33, analyzing the temporal change pattern of the thickness data of each zone through the model to obtain the nerve fiber layer damage progression rate parameter; S34, fusing the compliance rate, deviation rate and progression rate parameter of each zone to generate a quantitative result of nerve fiber layer integrity.

[0012] Further, S4 includes the following sub-steps: S41, standardizing the range of physiological indicators related to ocular surface inflammation and mapping each indicator to a unified data range; S42, inputting the standardized indicator data into the postoperative inflammation response prediction model and calculating the inflammation contribution of each indicator; S43, outputting the inflammation severity grading result based on the inflammation contribution and the relationship between the indicators; S44, analyzing the temporal change trend of the inflammation indicators through the model and combining the characteristics of the postoperative recovery cycle to generate a predicted value for the inflammation regression trend.

[0013] Further, S5 includes the following sub-steps: S51, establishing a mapping relationship table between intraocular pressure feature vector, nerve fiber layer assessment results, and inflammation prediction data, and clarifying the correlation weight of each data dimension; S52, using a weighted fusion algorithm to fuse multi-dimensional data and calculate the comprehensive contribution value of each data dimension; S53, constructing a two-dimensional risk assessment matrix based on the comprehensive contribution value, with the row dimension representing complication type and the column dimension representing risk level; S54, verifying the rationality of the risk assessment matrix through a retinal nerve intelligent monitoring and analysis platform, eliminating abnormal data interference, and optimizing the matrix accuracy.

[0014] Further, S6 includes the following sub-steps: S61, retrieving personalized parameters such as patient age, surgical method, and preoperative disease severity from the personalized parameter library for AI-assisted glaucoma postoperative complication early warning and rehabilitation guidance; S62, determining the early warning thresholds and priorities for various complications based on the risk assessment matrix and personalized parameters; S63, generating early warning information including complication type, risk level, and early warning time window based on the early warning thresholds and priorities; S64, developing a phased rehabilitation guidance plan by combining the recovery pattern of the nerve fiber layer, the trend of inflammation resolution, and the intraocular pressure control target, including medication guidance, recommendations for follow-up examination frequency, and standardized eye care procedures.

[0015] An AI-assisted system for early warning and rehabilitation guidance of postoperative glaucoma complications is implemented. This system comprises: a multi-dimensional temporal data acquisition and integration unit, an intelligent mining and vector generation unit for intraocular pressure temporal features, a quantitative assessment unit for retinal nerve fiber layer integrity, a dynamic prediction and grading unit for ocular surface inflammation, a multi-source data fusion and complication risk matrix construction unit, and a rehabilitation guidance program output unit. The output of the multi-dimensional temporal data acquisition and integration unit is connected to the inputs of the intelligent mining and vector generation unit for intraocular pressure temporal features, the quantitative assessment unit for retinal nerve fiber layer integrity, and the dynamic prediction and grading unit for ocular surface inflammation, respectively, for transmitting the constructed multi-dimensional temporal dataset to these three units. The intelligent mining and vector generation unit for intraocular pressure temporal features... The outputs of the data mining and vector generation unit, the quantification assessment unit for the integrity of the retinal nerve fiber layer, and the dynamic prediction and grading unit for ocular surface inflammation are all connected to the input of the multi-source data fusion and complication risk matrix construction unit. These units transmit their respective generated high-dimensional intraocular pressure feature vectors, quantification results of nerve fiber layer integrity, damage progression rate parameters, inflammation severity grading results, and predicted inflammation regression trends. The output of the multi-source data fusion and complication risk matrix construction unit is connected to the input of the rehabilitation guidance scheme output unit, transmitting the constructed complication risk assessment matrix. The rehabilitation guidance scheme output unit calls upon the personalized parameter library for AI-assisted postoperative glaucoma complication early warning and rehabilitation guidance to generate and output targeted complication early warning information and phased rehabilitation guidance schemes.

[0016] Beneficial Effects: This invention proposes an AI-assisted method and system for early warning and rehabilitation guidance of postoperative glaucoma complications. By systematically integrating multi-dimensional time-series data on postoperative intraocular pressure, retinal nerve fiber layer, and ocular surface inflammation, it breaks through the limitations of single-indicator monitoring, comprehensively captures dynamic changes during the eye recovery process, and accurately establishes the correlation between various physiological parameters and the occurrence of complications, significantly improving the timeliness and accuracy of early warning and solving the problems of one-sided assessment and delayed early warning in traditional technologies. Simultaneously, through a personalized parameter library, combined with individual characteristics such as differences in postoperative recovery and basic ocular conditions, it generates highly targeted, phased rehabilitation guidance plans, replacing traditional general recommendations, fully meeting the personalized rehabilitation needs of different patients, and significantly optimizing rehabilitation outcomes. Furthermore, through multi-source data fusion and dynamic evaluation mechanisms, it provides comprehensive and accurate decision support for medical staff, simplifies postoperative management processes, improves clinical work efficiency, effectively reduces the incidence of complications, ensures the long-term effectiveness of surgical treatment, and provides a scientific and efficient rehabilitation management solution for postoperative glaucoma patients. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the overall process of the method of the present invention. Figure 2 This is a flowchart of the method step S1 of the present invention; Figure 3 This is a flowchart of method step S2 of the present invention; Figure 4 This is a flowchart of step S3 of the method of the present invention. Figure 5 This is a flowchart of method step S4 of the present invention; Figure 6 This is a flowchart of step S5 of the method of the present invention; Figure 7 This is a flowchart of step S6 of the method of the present invention; Figure 8 This is a diagram showing the system unit composition of the present invention. Detailed Implementation

[0018] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. The application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0019] like Figure 1 As shown, the AI-assisted method for early warning and rehabilitation guidance of postoperative glaucoma complications includes the following steps: S1, collecting dynamic intraocular pressure data, retinal nerve fiber layer thickness distribution data, and ocular surface inflammation-related physiological indicator data at different postoperative time points through a retinal nerve intelligent monitoring and analysis platform to construct a multi-dimensional time-series dataset; S2, using an intraocular pressure time-series feature extraction algorithm to mine trend features, fluctuation features, and mutation features of the intraocular pressure data to generate a high-dimensional intraocular pressure feature vector; S3, inputting the retinal nerve fiber layer thickness data into a retinal nerve fiber layer assessment model to obtain the quantitative results of nerve fiber layer integrity and damage progression rate parameters; S4, using a postoperative inflammation response prediction model to analyze ocular surface inflammation indicator data and output inflammation severity grading results and inflammation regression trend prediction values; S5, using the retinal nerve intelligent monitoring and analysis platform to integrate the intraocular pressure feature vector, nerve fiber layer assessment results, and inflammation prediction data to construct a complication risk assessment matrix; S6, based on the risk assessment matrix and combined with the personalized parameter library of AI-assisted early warning and rehabilitation guidance for postoperative glaucoma complications, generating targeted complication early warning information and phased rehabilitation guidance plans.

[0020] Step S1 involves multi-dimensional data collection and integration using a retinal nerve intelligent monitoring and analysis platform. Specifically, at fixed time points of 1, 3, 7, 14, 30, 60, and 90 days post-surgery, dynamic intraocular pressure data, retinal nerve fiber layer thickness distribution data, and ocular surface inflammation-related physiological indicators are collected simultaneously. Intraocular pressure data is collected hourly for 5 minutes each time, and the average value is taken as the valid data for that time point. Retinal nerve fiber layer thickness data is divided into four quadrants: nasal, temporal, superior, and inferior. Each quadrant has 20 detection points, and the thickness value at each point is recorded. Ocular surface inflammation-related physiological indicators include four core indicators: tear secretion, corneal fluorescein staining degree, conjunctival hyperemia index, and aqueous humor flare value. Data for each indicator is collected according to a unified testing standard. During the data collection process, basic information such as the patient's age, surgical method, and preoperative disease severity are recorded simultaneously. All data are aligned according to timestamps, and invalid and abnormal data are removed. A three-dimensional time-series dataset is constructed, including time dimension, indicator dimension, and patient individual dimension. This step provides comprehensive and standardized data support for subsequent feature extraction, model analysis, and risk assessment, ensuring the accuracy and reliability of the analysis results in each subsequent stage.

[0021] Step S2 employs an intraocular pressure (IOP) temporal feature extraction algorithm to perform deep feature mining on the collected IOP data. During implementation, the IOP temporal data from 1 to 90 days post-surgery are processed in stages, divided into three phases: early postoperative period (1-7 days), mid-postoperative period (8-30 days), and late postoperative period (31-90 days). Trend features, fluctuation features, and mutation features are extracted for each phase. Trend features are obtained by calculating the linear fitting slope and mean change rate of the IOP data for each phase, with an absolute slope value greater than 0.5 considered a significant trend change. Fluctuation features are obtained by calculating the standard deviation, coefficient of variation, and range of the IOP data for each phase; a standard deviation greater than 3 and a coefficient of variation greater than 0.15 are considered abnormal fluctuations. Mutation features are determined by setting a threshold; when the absolute value of the difference between a single IOP measurement and the previous measurement is greater than 5, a mutation point is identified, and the time of occurrence, magnitude of numerical change, and duration of the mutation point are recorded. After normalizing the parameters under the three feature dimensions, the parameters are weighted and fused according to the following ratios: trend feature weight 0.4, fluctuation feature weight 0.35, and mutation feature weight 0.25. This generates a high-dimensional intraocular pressure feature vector with 30 dimensions. This vector comprehensively reflects the dynamic changes in intraocular pressure at different recovery stages and provides key intraocular pressure-related evidence for complication risk assessment.

[0022] Step S3 inputs the retinal nerve fiber layer thickness data into the retinal nerve fiber layer assessment model for quantitative analysis. During implementation, the thickness data of 20 detection points in each of the four quadrants are first statistically analyzed by region, calculating the average thickness, minimum thickness, standard deviation, and coefficient of variation for each quadrant. This is then compared with the baseline thickness data of the corresponding quadrant before surgery to obtain the thickness change rate and trend for each quadrant. The model integrates the thickness parameters, change rate parameters, and pre- and post-operative difference data for each quadrant, employing a hierarchical assessment mechanism. First, the integrity of individual detection points is rated, categorized into four levels: intact, mild damage, moderate damage, and severe damage. Then, a comprehensive rating is performed by quadrant, finally generating a quantitative result for the overall nerve fiber layer integrity. The quantitative result ranges from 0 to 100, with higher values ​​indicating better integrity. Simultaneously, based on the thickness data at two adjacent time points, the damage progression rate parameter for each quadrant and the overall model is calculated. The progression rate is calculated based on the daily thickness change; a daily thickness decrease greater than 0.2 is considered rapid progression. This step provides core neural structure-related data support for assessing the eye's recovery status and the risk of complications by precisely quantifying the integrity and damage progression of the nerve fiber layer.

[0023] Step S4 utilizes a postoperative inflammatory response prediction model to systematically analyze ocular surface inflammation index data. During implementation, four indicators—tear secretion, corneal fluorescein staining degree, conjunctival hyperemia index, and aqueous humor flare value—are first standardized. Each indicator is converted into a quantitative score of 0-10 according to a preset scoring standard. Specifically, tear secretion greater than 10 is 10 points, 5-10 is 5 points, and less than 5 is 0 points; corneal fluorescein staining is 10 points for no staining, 6 points for mild staining, 3 points for moderate staining, and 0 points for severe staining; conjunctival hyperemia index is 10 points for no hyperemia, 7 points for mild hyperemia, 4 points for moderate hyperemia, and 0 points for severe hyperemia; and aqueous humor flare value is 10 points for no flare, 6 points for mild flare, 3 points for moderate flare, and 0 points for severe flare. The model obtains a comprehensive inflammation score through weighted summation, and classifies the degree of inflammation into four levels according to the score range: no inflammation (0-2 points), mild inflammation (3-5 points), moderate inflammation (6-8 points), and severe inflammation (9-10 points), outputting the inflammation severity classification result. Simultaneously, based on the comprehensive inflammation score at different postoperative time points, an inflammation change curve is fitted, the inflammation resolution rate is calculated, and the inflammation resolution trend is predicted within 180 days postoperatively, yielding a predicted value for the inflammation resolution trend. This predicted value is expressed as the number of days required for complete resolution of remaining inflammation, providing crucial inflammation-related evidence for assessing the ocular recovery process and the risk of complications.

[0024] Step S5 involves using the retinal nerve intelligent monitoring and analysis platform to perform multi-source data fusion and risk assessment matrix construction. First, an association mapping relationship is established between the intraocular pressure feature vector, nerve fiber layer assessment results, and inflammation prediction data. A fusion weight system is set with the following weights: intraocular pressure feature vector weight 0.35, nerve fiber layer integrity quantification result weight 0.35, damage progression rate parameter weight 0.15, inflammation severity grading result weight 0.1, and inflammation regression trend prediction value weight 0.05. A weighted average algorithm is used to fuse the multi-dimensional data, obtaining a comprehensive assessment value for each time point. The comprehensive assessment value ranges from 0 to 100, with higher values ​​indicating better recovery and lower complication risk. Based on the comprehensive assessment value and the complication type classification, a two-dimensional risk assessment matrix is ​​constructed, including three categories of complications: intraocular pressure abnormality-related complications, nerve fiber layer damage-related complications, and inflammation-related complications, with three risk levels: low risk (0-30 points), medium risk (31-60 points), and high risk (61-100 points). During the matrix construction process, the rationality is verified by comparing with historical clinical data, and abnormal data that deviates from clinical patterns are eliminated to ensure that the matrix can accurately reflect the risk of complications under different recovery stages and different combinations of data characteristics, providing a direct basis for the generation of subsequent early warning information and the formulation of rehabilitation guidance programs.

[0025] Step S6, based on the constructed complication risk assessment matrix and combined with a personalized parameter library for AI-assisted early warning and rehabilitation guidance of postoperative glaucoma complications, generates targeted complication early warning information and a phased rehabilitation guidance plan. During implementation, personalized parameters such as patient age, surgical method, preoperative disease severity, basic ocular conditions, and overall health status are retrieved from the personalized parameter library and matched with the results of the risk assessment matrix. For high-risk items in the risk assessment matrix, early warning information is generated, including complication type, risk level, early warning time window, and key risk indicators. The early warning time window is divided into short-term (1-7 days), medium-term (8-30 days), and long-term (31-90 days), clearly defining the key monitoring indicators and monitoring frequency for different stages. The phased rehabilitation guidance plan is formulated in three stages: 1-30 days post-surgery, 31-90 days post-surgery, and 91-180 days post-surgery. Each stage includes three parts: medication guidance, recommended follow-up frequency, and eye care operation guidelines. The medication guidance specifies the type, dosage, frequency of use, and adjustment points for eye drops. The recommended follow-up frequency is set according to risk level: high-risk patients should be followed up every 3 days, medium-risk patients every 7 days, and low-risk patients every 14 days. The eye care operation guidelines specify cleaning methods, limits on eye use time, dietary and exercise precautions, etc., to ensure the personalization, scientific nature, and operability of the rehabilitation guidance plan, helping patients recover smoothly and reducing the incidence of complications.

[0026] Preferably, the expression for the intraocular pressure temporal feature extraction algorithm is: , in, This represents the temporal comprehensive characteristic value of intraocular pressure. The trend feature weighting coefficient, Let be the intraocular pressure value at time t. This represents the total duration of data collection. The time decay coefficient, This is the baseline time point after surgery. The fluctuation characteristic weighting coefficient, To detect the number of data points, Let i be the intraocular pressure value from the i-th measurement. Mean intraocular pressure Let be the credibility weight of the i-th detection data.

[0027] Specifically, the intraocular pressure (IOP) temporal feature extraction algorithm is based on the temporal characteristics of postoperative IOP data. It combines the influence weights of trend changes, fluctuation amplitude, and abrupt changes on complication warning. The algorithm quantifies the dynamic trend of IOP over time through integral calculations, introduces an exponential decay term to correct the correlation between data at different time points, and uses statistical methods to calculate fluctuation characteristics. Finally, a weighted fusion is used to achieve a comprehensive representation of multi-dimensional features. First, the weight coefficients of trend and fluctuation characteristics are determined through clinical data statistical analysis. The weight of trend characteristics ranges from 0.6 to 0.8, and the weight of fluctuation characteristics ranges from 0.2 to 0.4. The time decay coefficient is set to 0.01 to 0.05 based on the postoperative recovery period to ensure a higher contribution of recent data to feature extraction. In implementation, the total data collection period is determined to be 90 days, the postoperative baseline is set to day 1 postoperatively, and the number of data points is determined to be 30 based on a collection frequency of once every 3 days. The reliability weight is set to 0.8-1.0 based on the accuracy of the detection equipment and the degree of operational standardization. This formula integrates the trend rate of change of intraocular pressure, the time decay effect, and the degree of fluctuation dispersion, which can comprehensively capture the key information in the intraocular pressure time series data. The trend feature reflects the overall direction of intraocular pressure change, the fluctuation feature reflects the stability of intraocular pressure, and the mutation feature identifies abnormal fluctuation nodes. The parameter values ​​have been verified by a large amount of clinical data to ensure the accuracy of feature extraction, providing comprehensive and accurate intraocular pressure feature support for subsequent complication risk assessment.

[0028] Preferably, the expression for the retinal nerve fiber layer assessment model is: , in, This is a quantitative value for the integrity of the nerve fiber layer. For thickness weighting coefficient, To determine the number of zones in the nerve fiber layer, This represents the thickness value of the nerve fiber layer in the j-th region. Let be the functional importance coefficient of the j-th partition. For zone attenuation coefficient, Let be the blood vessel distribution density of the j-th partition. The progress rate weighting coefficient, The difference in thickness between two consecutive measurements. This represents the time interval between adjacent detections.

[0029] Specifically, the retinal nerve fiber layer assessment model is based on the zonal characteristics of nerve fiber layer thickness distribution. It combines the differences in functional importance of each zone and the influence of vascular distribution on nerve fiber layer integrity. The model quantifies the thickness contribution of each zone through stratified weighted summation, introducing the nonlinear effects of several modified vascular distribution densities, and incorporating the thickness change rate to reflect damage progression. Anatomical studies were used to determine the functional importance coefficients of each zone: 0.3 for the nasal and temporal quadrants, and 0.2 for the superior and inferior quadrants. The thickness weighting coefficient ranged from 0.7 to 0.9, the zone attenuation coefficient from 0.1 to 0.3, and the progression rate weighting coefficient from 0.4 to 0.6. During implementation, the nerve fiber layer detection zone was set to four quadrants, with 20 detection points in each quadrant. The time interval between adjacent detections was set at 14 days post-surgery. The thickness difference between two adjacent detections was calculated to obtain damage progression data. This formula integrates the thickness, functional importance, vascular distribution, and damage progression rate of each region to accurately quantify the integrity of the nerve fiber layer. The product of the thickness and functional importance of each region reflects the contribution of the region. The vascular distribution density avoids interference from extreme values ​​by adjusting several factors. The damage progression rate is directly related to the risk of complications. The parameter values ​​are based on a large number of clinical case statistics to ensure the scientificity and reliability of the assessment results, providing a core basis for neural structure assessment for complication early warning.

[0030] Preferably, the expression for the postoperative inflammatory response prediction model is: , in, This is a predictive value for the degree of inflammation. The overall weight of inflammatory markers, For the number of inflammation detection indicators, The value of the k-th inflammatory marker is... Let be the inflammation correlation coefficient of the k-th indicator. This is the coefficient for the rate of inflammation resolution. This is the peak time of postoperative inflammation. This is the weighting coefficient for inflammation fluctuations. This represents the mean of inflammatory markers.

[0031] Specifically, the postoperative inflammation prediction model is based on the temporal variation of inflammatory indicators, combined with the correlation differences between various inflammatory indicators and the degree of inflammation. It quantifies the overall inflammation level through weighted summation, introduces an exponential function to correct the rate of inflammation regression at different postoperative stages, and uses statistical methods to calculate the degree of inflammation fluctuation, achieving a comprehensive prediction of the degree of inflammation and its regression trend. The correlation coefficients of each inflammatory indicator were determined through clinical experimental data analysis: tear secretion and conjunctival hyperemia index were set at 0.3, corneal fluorescein staining degree and aqueous humor flare value were set at 0.2, the overall weight of inflammatory indicators ranged from 0.8 to 1.0, the inflammation regression rate coefficient ranged from 0.02 to 0.06, the postoperative inflammation peak time was set at 7 days postoperatively based on clinical statistics, and the inflammation fluctuation weight ranged from 0.1 to 0.3. During implementation, four inflammatory indicators were detected, and the detection frequency was consistent with the intraocular pressure data collection. The degree of fluctuation was quantified by calculating the sum of squared deviations of each indicator from the mean. This formula, by integrating the weighted contributions, time decay effects, and fluctuation dispersion of inflammatory indicators, can accurately predict the degree and regression trend of inflammation. The weighted summation reflects the comprehensive impact of each indicator on inflammation, the exponential term simulates the regression pattern of inflammation over time, the fluctuation characteristics reflect the stability of inflammation, and the parameter values ​​have been clinically validated to suit the recovery status of different patients, providing key inflammatory assessment basis for complication early warning and rehabilitation guidance.

[0032] Preferred, such as Figure 2 The step S1 includes the following sub-steps: S11, using the intraocular pressure sensing module of the retinal nerve intelligent monitoring and analysis platform, collecting intraocular pressure data from 1 day to 90 days post-surgery at preset time intervals, and recording the timestamp and detection location information for each collection; S12, using the platform's optical coherence tomography module, acquiring the thickness data of the retinal nerve fiber layer in the four quadrants of the nasal, temporal, superior, and inferior sides, and generating a layer thickness distribution map; S13, using the platform's ocular surface analyzer to collect tear secretion, corneal fluorescein staining degree, conjunctival hyperemia index, and inflammation-related physiological indicators, forming a standardized indicator dataset; S14, aligning the intraocular pressure data, nerve fiber layer thickness data, and inflammation indicator data according to the timestamps to construct a time-series dataset including basic patient information, detection time, and multi-dimensional physiological data.

[0033] Specifically, step S1 involves constructing a multi-dimensional time-series dataset: S11 uses the intraocular pressure sensing module of the retinal nerve intelligent monitoring and analysis platform to collect intraocular pressure data at preset intervals, spanning from 1 day to 90 days post-surgery. Each collection process lasts 5 minutes, simultaneously recording the precise timestamp of each collection and specific information from the two detection locations—central and peripheral cornea—to ensure the temporal integrity and positional accuracy of the data. S12 utilizes the platform's optical coherence tomography (OCT) module to scan and image the retinal nerve fiber layer in four quadrants: nasal, temporal, superior, and inferior. Twenty evenly distributed detection points are set in each quadrant. After acquiring the thickness data at each point, an intuitive slice thickness distribution map is generated, clearly presenting the spatial distribution of the nerve fiber layer thickness. Features; S13: Using the ocular surface analyzer integrated into the platform, three core inflammation-related physiological indicators—tear secretion, corneal fluorescein staining degree, and conjunctival hyperemia index—are collected according to unified detection standards. Each indicator is collected three times and the average value is taken to form a standardized indicator dataset, ensuring data reliability. S14: The collected intraocular pressure data, nerve fiber layer thickness data, and inflammation indicator data are precisely aligned according to time stamps, and associated with basic information such as the patient's age, surgical method, and preoperative disease level. Invalid data and outliers that occur during the collection process are removed, and finally, a time-series dataset including time, indicators, and individual patient dimensions is constructed. This provides a comprehensive, standardized, and high-quality data foundation for subsequent analysis and processing steps, ensuring the accuracy and effectiveness of subsequent technical operations.

[0034] Preferred, such as Figure 3 The S2 step includes the following sub-steps: S21, performing time-series segmentation processing on the intraocular pressure data, dividing it into three stages: postoperative days 1-7, 8-30, and 31-90, and extracting data for each stage; S22, using the sliding window method to extract trend features from the intraocular pressure data of each stage, calculating the rate of change of the mean intraocular pressure and the slope of the linear fit within the window; S23, calculating the standard deviation, coefficient of variation, and range of the intraocular pressure data of each stage through statistical analysis methods, and extracting fluctuation features; S24, using an outlier detection algorithm to identify abrupt changes in the intraocular pressure data, calculating the difference in intraocular pressure before and after the abrupt change and the frequency of the abrupt change, and extracting abrupt change features; S25, concatenating the trend features, fluctuation features, and abrupt change features to generate a high-dimensional intraocular pressure feature vector.

[0035] Specifically, step S2 involves extracting intraocular pressure (IOP) temporal features, generating a high-dimensional feature vector through five sub-steps: S21 first divides the complete IOP temporal data from 1 to 90 days post-surgery into stages, clearly defining three key stages: early postoperative period (1-7 days), mid-postoperative period (8-30 days), and late postoperative period (31-90 days). Raw IOP data for each stage is extracted, laying the foundation for staged feature mining. S22 uses a sliding window method with a window size of 3 days to process the IOP data segment by segment, calculating the rate of change of the mean IOP and the slope of the linear fit within each window. The slope value is used to determine the upward or downward trend of IOP, accurately extracting trend features. S23 uses statistical analysis methods to calculate the standard deviation, coefficient of variation, and range of the IOP data for each stage. Key parameters include standard deviation (reflecting data dispersion), coefficient of variation (reflecting relative volatility), and range (characterizing data fluctuation range), comprehensively extracting volatility features. S24 employs a threshold-based outlier detection algorithm, setting the intraocular pressure (IOP) mutation threshold to 5. When the absolute value of the difference between a single detection value and the previous detection value exceeds this threshold, it is identified as a mutation point. The occurrence time, magnitude of numerical change, and duration of the mutation point are recorded simultaneously, completing mutation feature extraction. S25 integrates the extracted trend features, volatility features, and mutation features, sequentially splicing them to form a high-dimensional IOP feature vector with 30 dimensions. This vector fully encompasses the dynamic changes in IOP at different recovery stages, providing accurate and comprehensive IOP feature support for subsequent complication risk assessment.

[0036] Preferred, such as Figure 4 The S3 step includes the following steps: S31, performing zonal calibration on the retinal nerve fiber layer thickness data, and determining the standard range and coordinate mapping relationship of each detection zone based on anatomical landmarks; S32, inputting the calibrated thickness data of each zone into the retinal nerve fiber layer assessment model, and calculating the thickness compliance rate and deviation rate from the preoperative baseline value of each zone; S33, analyzing the temporal change pattern of the thickness data of each zone through the model to obtain the nerve fiber layer damage progression rate parameter; S34, fusing the compliance rate, deviation rate and progression rate parameter of each zone to generate a quantitative result of nerve fiber layer integrity.

[0037] Specifically, step S3 revolves around the assessment of the retinal nerve fiber layer, achieving quantification of integrity and analysis of damage progression through four sub-steps: S31 first calibrates the collected retinal nerve fiber layer thickness data by region, clarifying the boundaries of the four quadrants (nasal, temporal, superior, and inferior) and the coordinate mapping relationship of each detection point according to the standard markings of ocular anatomy, ensuring the accuracy and consistency of the data for each region; S32 inputs the calibrated thickness data of each region into the assessment model, calculating the thickness attainment rate (the attainment thickness is set at 80% of the preoperative baseline value) and the deviation rate from the preoperative baseline value for each region, visually reflecting the recovery status of the thickness of each region through numerical values; S33 utilizes the model's built-in time... The sequence analysis function performs trend fitting on the thickness data of each stage and each region, calculates the thickness change between two adjacent detection time points (14 days apart), and then obtains the daily thickness change rate, i.e., the nerve fiber layer damage progression rate parameter. When the daily thickness reduction exceeds 0.2, it is marked as rapid progression. S34 uses a weighted fusion method to comprehensively calculate the compliance rate, deviation rate and progression rate parameter of each region according to the functional importance coefficient of each quadrant (0.3 for nasal and temporal sides, and 0.2 for superior and inferior sides), and generates a quantitative result of nerve fiber layer integrity. The result ranges from 0 to 100. The higher the value, the better the integrity, providing a core basis for the neural structural status for complication risk assessment.

[0038] Preferred, such as Figure 5 The S4 step includes the following sub-steps: S41, standardizing the range of physiological indicators related to ocular surface inflammation and mapping each indicator to a unified data range; S42, inputting the standardized indicator data into the postoperative inflammation response prediction model and calculating the inflammation contribution of each indicator; S43, outputting the inflammation severity grading result based on the inflammation contribution and the relationship between the indicator combinations; S44, analyzing the temporal change trend of inflammation indicators through the model and combining it with the characteristics of the postoperative recovery cycle to generate a predicted value for the inflammation regression trend.

[0039] Specifically, step S4 is used to predict postoperative inflammatory response. This is achieved through four sub-steps: S41 first standardizes the collected data on four inflammatory indicators—tear secretion, corneal fluorescein staining intensity, conjunctival hyperemia index, and aqueous humor flare value—transforming the raw values ​​of each indicator into a 0-10 range according to a unified mapping rule, eliminating dimensional differences between different indicators and ensuring data comparability; S42 inputs the standardized indicator data into the postoperative inflammatory response prediction model. The model calculates the inflammatory contribution of each indicator based on preset indicator weights (tear secretion and conjunctival hyperemia index = 0.3, corneal fluorescein staining intensity and aqueous humor flare value = 0.2), clarifying the inflammatory contribution of different indicators. The model assesses the degree of influence of each indicator on the overall inflammatory state. Based on the inflammatory contribution of each indicator and the comprehensive weighted score, the model categorizes the degree of inflammation according to the score range: 0-2 indicates no inflammation, 3-5 indicates mild inflammation, 6-8 indicates moderate inflammation, and 9-10 indicates severe inflammation, outputting a clear grading result for the degree of inflammation. The S44 model analyzes the comprehensive inflammation score at each time point, fits the inflammation change curve, calculates the inflammation regression rate (in units of daily score decrease), and, combined with the postoperative recovery cycle pattern, predicts the trend of inflammation changes within 180 days post-surgery, obtaining a predicted value for the inflammation regression trend. This value is expressed as the number of days required for complete resolution of the remaining inflammation, providing crucial inflammatory-related evidence for assessing the ocular recovery process and the risk of complications.

[0040] Preferred, such as Figure 6 The S5 step includes the following steps: S51, establishing a mapping relationship table between intraocular pressure feature vector, nerve fiber layer assessment results, and inflammation prediction data, and clarifying the correlation weight of each data dimension; S52, using a weighted fusion algorithm to fuse multi-dimensional data and calculate the comprehensive contribution value of each data dimension; S53, constructing a two-dimensional risk assessment matrix based on the comprehensive contribution value, with the row dimension representing complication type and the column dimension representing risk level; S54, verifying the rationality of the risk assessment matrix through a retinal nerve intelligent monitoring and analysis platform, eliminating abnormal data interference, and optimizing the matrix accuracy.

[0041] Specifically, step S5 involves multi-source data fusion and risk matrix construction, achieving comprehensive risk assessment through four sub-steps: S51 first establishes a correlation mapping table between intraocular pressure feature vectors, nerve fiber layer assessment results, and inflammation prediction data. Based on clinical data statistical analysis, the correlation weights of each data dimension are determined, with the intraocular pressure feature vector having a weight of 0.35, the nerve fiber layer integrity quantification result having a weight of 0.35, the damage progression rate parameter having a weight of 0.15, the inflammation severity grading result having a weight of 0.1, and the inflammation regression trend prediction value having a weight of 0.05, thus clarifying the relative importance of each data dimension. S52 employs a weighted fusion algorithm, multiplying the quantification results of each data dimension by their corresponding correlation weights and summing the results to calculate the comprehensive assessment value for each time point. The comprehensive assessment value ranges from 0 to 100, with higher values ​​indicating higher risk. The better the recovery status, the lower the risk of complications. S53, based on the comprehensive assessment value and combined with common clinical complication types, constructs a two-dimensional risk assessment matrix. The matrix's row dimension is divided into three categories: intraocular pressure abnormality-related complications, nerve fiber layer damage-related complications, and inflammation-related complications. The column dimension is divided into three levels: low risk (0-30 points), medium risk (31-60 points), and high risk (61-100 points), clearly presenting the risk status of different complications. S54, through the retinal nerve intelligent monitoring and analysis platform, calls the historical clinical database to compare and verify the constructed risk assessment matrix with the risk data of similar cases, eliminates abnormal data that deviates from clinical patterns, corrects unreasonable risk classifications in the matrix, optimizes the matrix accuracy, ensures the scientificity and reliability of the risk assessment results, and provides a direct basis for the generation of subsequent early warning information.

[0042] Preferred, such as Figure 7 The S6 procedure includes the following steps: S61, retrieving personalized parameters such as patient age, surgical method, and preoperative disease severity from the personalized parameter library for AI-assisted glaucoma postoperative complication early warning and rehabilitation guidance; S62, determining the early warning thresholds and priorities for various complications based on the risk assessment matrix and personalized parameters; S63, generating early warning information including complication type, risk level, and early warning time window based on the early warning thresholds and priorities; and S64, developing a phased rehabilitation guidance plan by combining the recovery pattern of the nerve fiber layer, the trend of inflammation resolution, and the intraocular pressure control target. The plan includes medication guidance, recommendations for follow-up examination frequency, and standardized eye care procedures.

[0043] Specifically, step S6 involves personalized early warning and rehabilitation guidance, generating a targeted plan through four sub-steps: S61: From the personalized parameter library of AI-assisted glaucoma postoperative complication early warning and rehabilitation guidance, key personalized parameters such as the patient's age, surgical method (e.g., trabeculectomy, glaucoma drainage device implantation), preoperative disease severity (mild, moderate, severe), basic ocular conditions (e.g., corneal thickness, anterior chamber depth), and overall health status are precisely retrieved to provide an individual basis for plan customization; S62: Based on the risk level classification of the risk assessment matrix and the retrieved personalized parameters, a weighted analysis method is used to determine the early warning thresholds for various complications. The early warning threshold for high-risk patients is reduced by 20%, for medium-risk patients it is set to the standard threshold, and for low-risk patients it is increased by 10%. Simultaneously, the early warning priority is determined according to the severity and probability of the complication, with severe and frequently occurring complications set as first priority, decreasing sequentially; S63: Based on the set early warning thresholds and priorities, a plan is generated including the complication type, specific risk level, and early warning information. Early warning information on time windows (1-7 days in the near term, 8-30 days in the medium term, and 31-90 days in the long term) and key risk indicators ensures that medical staff and patients are promptly aware of the risk situation. S64, combining the recovery pattern of the nerve fiber layer, the trend of inflammation resolution, and intraocular pressure control targets (normal intraocular pressure range is 10-21), develops phased rehabilitation guidance plans for three stages: 1-30 days post-surgery, 31-90 days, and 91-180 days. The plans specify the type of eye drops, dosage (e.g., 1 drop per use), and frequency of use. (e.g., 3 times daily) and adjustment points (e.g., reduce medication frequency when inflammation subsides to mild). The frequency of follow-up examinations is recommended to be set according to risk level (once every 3 days for high risk, once every 7 days for medium risk, and once every 14 days for low risk). The eye care operation specifications clearly define the cleaning method (e.g., wiping around the eyes with sterile cotton swabs), the time limit for using the eyes (no more than 40 minutes at a time), and dietary and exercise precautions (e.g., avoiding spicy foods and not engaging in strenuous exercise). This ensures that the rehabilitation guidance plan is personalized, scientific, and operable, helping patients recover smoothly.

[0044] Preferred, such as Figure 8The AI-assisted glaucoma postoperative complication early warning and rehabilitation guidance system is characterized in that it is applied to an AI-assisted glaucoma postoperative complication early warning and rehabilitation guidance method, comprising: a multi-dimensional time-series data acquisition and integration unit, an intraocular pressure time-series feature intelligent mining and vector generation unit, a retinal nerve fiber layer integrity quantitative assessment unit, an ocular surface inflammation response dynamic prediction and grading unit, a multi-source data fusion and complication risk matrix construction unit, and a rehabilitation guidance plan output unit; the output end of the multi-dimensional time-series data acquisition and integration unit is connected to the input ends of the intraocular pressure time-series feature intelligent mining and vector generation unit, the retinal nerve fiber layer integrity quantitative assessment unit, and the ocular surface inflammation response dynamic prediction and grading unit, respectively, for transmitting the constructed multi-dimensional time-series dataset to the three units; the intraocular pressure time... The outputs of the sequence feature intelligent mining and vector generation unit, the retinal nerve fiber layer integrity quantitative assessment unit, and the ocular surface inflammation response dynamic prediction and grading unit are all connected to the input of the multi-source data fusion and complication risk matrix construction unit. These units transmit their respective generated high-dimensional intraocular pressure feature vectors, nerve fiber layer integrity quantitative results and damage progression rate parameters, inflammation severity grading results, and inflammation regression trend prediction values. The output of the multi-source data fusion and complication risk matrix construction unit is connected to the input of the rehabilitation guidance scheme output unit, transmitting the constructed complication risk assessment matrix. The rehabilitation guidance scheme output unit calls upon the personalized parameter library of AI-assisted glaucoma postoperative complication early warning rehabilitation guidance to generate and output targeted complication early warning information and phased rehabilitation guidance schemes.

[0045] An AI-assisted method and system for early warning and rehabilitation guidance of postoperative glaucoma complications has completely changed the limitations of traditional single-indicator monitoring through multi-dimensional data integration and dynamic evaluation mechanisms. The system comprehensively collects postoperative intraocular pressure, retinal nerve fiber layer, and ocular surface inflammation-related physiological data, achieving in-depth correlation analysis of multi-dimensional time-series information. It can fully capture subtle changes and potential risk signals during the eye's recovery process, effectively overcoming the shortcomings of traditional technologies in terms of one-sided assessment and delayed early warning. This significantly improves the accuracy and timeliness of complication early warning, providing medical staff with comprehensive and forward-looking basis for disease assessment.

[0046] Meanwhile, this method and system, through a personalized parameter library, fully considers individual patient recovery differences and specific factors such as basic ocular conditions, generating phased rehabilitation guidance plans tailored to different patients. This completely eliminates the drawbacks of traditional universal recommendations, accurately meeting patients' personalized rehabilitation needs and significantly optimizing postoperative recovery outcomes. Its closed-loop management model, constructed through data fusion and intelligent analysis, simplifies clinical postoperative management processes, improves medical service efficiency, reduces complication rates through scientific guidance, and ensures the long-term effectiveness of surgical treatment, providing a more efficient and precise solution for postoperative rehabilitation management of glaucoma.

[0047] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various equivalent changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An AI-assisted postoperative complication early warning rehabilitation guidance method, characterized in that, Includes the following steps: S1. The system collects dynamic intraocular pressure data, retinal nerve fiber layer thickness distribution data, and ocular surface inflammation-related physiological index data at different time points after surgery through the retinal nerve intelligent monitoring and analysis platform to construct a multi-dimensional time-series dataset. S2, an intraocular pressure time series feature extraction algorithm is used to mine trend features, fluctuation features and mutation features of intraocular pressure data to generate a high-dimensional intraocular pressure feature vector; S3, input the retinal nerve fiber layer thickness data into the retinal nerve fiber layer assessment model to obtain the quantitative results of nerve fiber layer integrity and damage progression rate parameters; S4. The postoperative inflammatory response prediction model is used to analyze ocular surface inflammatory index data and output the grading results of inflammatory severity and the predicted value of inflammatory regression trend. S5 integrates intraocular pressure feature vectors, nerve fiber layer assessment results, and inflammation prediction data through a retinal nerve intelligent monitoring and analysis platform to construct a complication risk assessment matrix; S6, based on a risk assessment matrix and combined with a personalized parameter library for AI-assisted early warning and rehabilitation guidance of postoperative glaucoma complications, generates targeted complication warning information and phased rehabilitation guidance plans.

2. The AI-assisted post-glaucoma-surgery complication warning and rehabilitation guidance method of claim 1, wherein, The expression for the intraocular pressure temporal feature extraction algorithm is as follows: ， wherein, is an intraocular pressure time series comprehensive characteristic value, is a trend characteristic weight coefficient, is an intraocular pressure detection value at time t, is a total data collection duration, is a time decay coefficient, is a postoperative reference time point, is a fluctuation characteristic weight coefficient, is a detection data point number, is an i-th intraocular pressure detection value, is an intraocular pressure mean value, is a reliability weight of i-th detection data.

3. The AI-assisted post-glaucoma-surgery complication warning and rehabilitation guidance method of claim 1, wherein, The expression for the retinal nerve fiber layer assessment model is as follows: ， wherein, is a nerve fiber layer integrity quantification value, is a thickness weight coefficient, is a number of nerve fiber layer detection partitions, is a nerve fiber layer thickness value for the jth partition, is a functional importance coefficient for the jth partition, is a partition decay coefficient, is a blood vessel distribution density for the jth partition, is a progression rate weight coefficient, is a thickness difference value between adjacent detections, is an adjacent detection time interval.

4. The AI-assisted method for early warning and rehabilitation guidance of postoperative glaucoma complications according to claim 1, characterized in that, The expression for the postoperative inflammatory response prediction model is as follows: ， in, This is a predictive value for the degree of inflammation. The overall weight of inflammatory markers, For the number of inflammation detection indicators, The value of the k-th inflammatory marker is... Let be the inflammation correlation coefficient of the k-th indicator. This is the coefficient for the rate of inflammation resolution. This is the peak time of postoperative inflammation. This is the weighting coefficient for inflammation fluctuations. This represents the mean of inflammatory markers.

5. The AI-assisted method for early warning and rehabilitation guidance of postoperative glaucoma complications according to claim 1, characterized in that, Step S1 includes the following sub-steps: S11 uses the intraocular pressure sensing module of the retinal nerve intelligent monitoring and analysis platform to collect intraocular pressure data from 1 day to 90 days after surgery at preset time intervals, and records the timestamp and detection location information of each collection. S12, using the platform's optical coherence tomography module, acquires the thickness data of the retinal nerve fiber layer in the four quadrants of the nasal side, temporal side, superior side, and inferior side, and generates a layer thickness distribution map. S13 uses the platform's ocular surface analyzer to collect tear secretion, corneal fluorescein staining intensity, conjunctival hyperemia index, and inflammation-related physiological indicators to form a standardized indicator dataset. S14. The intraocular pressure data, nerve fiber layer thickness data, and inflammatory index data are aligned by timestamps to construct a time-series dataset that includes basic patient information, detection time, and multi-dimensional physiological data.

6. The AI-assisted method for early warning and rehabilitation guidance of postoperative glaucoma complications according to claim 1, characterized in that, S2 includes the following steps: S21. Time-series segmentation processing was performed on the intraocular pressure data, dividing it into three stages: postoperative days 1-7, 8-30, and 31-90, and data for each stage were extracted. S22, the sliding window method is used to extract trend features from intraocular pressure data at each stage, and the mean change rate of intraocular pressure and the slope of linear fitting within the window are calculated. S23, calculate the standard deviation, coefficient of variation and range of intraocular pressure data at each stage through statistical analysis methods, and extract fluctuation characteristics; S24, an outlier detection algorithm is used to identify mutation points in intraocular pressure data, calculate the intraocular pressure difference before and after the mutation point and the mutation frequency, and extract mutation features; S25 concatenates trend features, fluctuation features, and mutation features to generate a high-dimensional intraocular pressure feature vector.

7. The AI-assisted method for early warning and rehabilitation guidance of postoperative glaucoma complications according to claim 1, characterized in that, S3 includes the following steps: S31, perform zonal calibration on the retinal nerve fiber layer thickness data, and determine the standard range and coordinate mapping relationship of each detection zone based on anatomical landmarks; S32, input the calibrated thickness data of each zone into the retinal nerve fiber layer assessment model, and calculate the thickness compliance rate of each zone and the deviation rate from the preoperative baseline value; S33, by analyzing the temporal variation of thickness data in each region through model analysis, the parameters of nerve fiber layer damage progression rate are obtained; S34 integrates the compliance rate, deviation rate, and progression rate parameters of each region to generate a quantitative result of the integrity of the nerve fiber layer.

8. The AI-assisted method for early warning and rehabilitation guidance of postoperative glaucoma complications according to claim 1, characterized in that, S4 includes the following sub-steps: S41, standardize the range of physiological indicators related to ocular surface inflammation and map each indicator to a unified data range; S42, input the standardized indicator data into the postoperative inflammatory response prediction model to calculate the inflammatory contribution of each indicator; S43, based on the relationship between the contribution of inflammation and the combination of indicators, outputs the grading results of the degree of inflammation; S44 uses a model to analyze the temporal trends of inflammatory indicators and, combined with postoperative recovery cycle characteristics, generates a predicted value for the trend of inflammation resolution.

9. The AI-assisted method for early warning and rehabilitation guidance of postoperative glaucoma complications according to claim 1, characterized in that, S5 includes the following steps: S51, establish a mapping relationship table between intraocular pressure feature vector, nerve fiber layer assessment results and inflammation prediction data, and clarify the correlation weight of each data dimension; S52 uses a weighted fusion algorithm to fuse multi-dimensional data and calculate the comprehensive contribution value of each data dimension; S53. Construct a two-dimensional risk assessment matrix based on the comprehensive contribution value. The row dimension of the matrix represents the complication type, and the column dimension represents the risk level. S54 uses a retinal nerve intelligent monitoring and analysis platform to verify the rationality of the risk assessment matrix, eliminate abnormal data interference, and optimize the matrix accuracy.

10. An AI-assisted early warning and rehabilitation guidance system for postoperative glaucoma complications, characterized in that, The system is applied to the AI-assisted glaucoma postoperative complication early warning and rehabilitation guidance method as described in claim 1, including: a multi-dimensional time-series data acquisition and integration unit, an intraocular pressure time-series feature intelligent mining and vector generation unit, a retinal nerve fiber layer integrity quantitative assessment unit, an ocular surface inflammation response dynamic prediction and grading unit, a multi-source data fusion and complication risk matrix construction unit, and a rehabilitation guidance program output unit. The output of the multi-dimensional temporal data acquisition and integration unit is connected to the input of the intraocular pressure temporal feature intelligent mining and vector generation unit, the retinal nerve fiber layer integrity quantitative assessment unit, and the ocular surface inflammation response dynamic prediction and grading unit, respectively, for transmitting the constructed multi-dimensional temporal dataset. The outputs of the intraocular pressure temporal feature intelligent mining and vector generation unit, the retinal nerve fiber layer integrity quantitative assessment unit, and the ocular surface inflammation response dynamic prediction and grading unit are all connected to the input of the multi-source data fusion and complication risk matrix construction unit, for transmitting the generated high-dimensional intraocular pressure feature vector, nerve fiber layer integrity quantitative results and damage progression rate parameters, inflammation severity grading results, and inflammation regression trend prediction values, respectively. The output of the multi-source data fusion and complication risk matrix construction unit is connected to the input of the rehabilitation guidance scheme output unit, for transmitting the constructed complication risk assessment matrix. The rehabilitation guidance scheme output unit calls the personalized parameter library of AI-assisted glaucoma postoperative complication early warning rehabilitation guidance to generate and output targeted complication early warning information and phased rehabilitation guidance schemes.

Images