A method of ocular disease risk assessment incorporating ophthalmic genetic information

Abstract

The application provides an eye disease risk assessment method combined with ophthalmic genetic information, comprising: from the enhanced variant-structure association data set, extracting high-impact variant type classification criteria according to the variant record screening rule, performing deep comparison on the association strength of high-impact variants and structures, constructing an eye tissue-specific damage distribution map, and judging the potential damage area distribution characteristics; according to the eye tissue-specific damage distribution map, using a similar mode grouping technology combined with a paired mapping time window analysis, dividing the damage risk mode, calculating the tissue damage probability and risk weight, and obtaining a refined eye tissue risk distribution map; if the damage probability of a certain tissue in the refined eye tissue risk distribution map exceeds the preset warning line, a microscopic level disease mechanism path diagram for the tissue is generated through inter-mode difference comparison technology combined with data pairing error range analysis, and the specific affected tissue site is determined.

Description

Technical Field

[0001] This invention relates to the field of information technology, and in particular to a method for assessing the risk of eye diseases by incorporating ophthalmic genetic information. Background Technology

[0002] As a crucial field of medical research, ophthalmic disease risk assessment bears the important mission of preventing and intervening in the early stages of eye health problems. With the interdisciplinary integration of genetics and clinical medicine, risk assessment based on genetic information is gradually becoming a key pathway to improve diagnostic accuracy. However, research and application in this field still face many challenges and urgently need to overcome the limitations of existing technologies to meet the needs of personalized medicine.

[0003] While current risk assessment methods can identify the association between certain gene mutations and eye diseases, they often remain at a relatively broad level, lacking in-depth analysis of the disease's pathogenesis. Particularly regarding the correspondence between gene mutations and specific eye structures, existing technologies struggle to accurately reveal the impact of mutations on different tissue layers of the eye. This limitation prevents doctors from accurately determining the specific sites that a disease might affect based on a patient's genetic information, thus impacting the targeted nature of treatment plans.

[0004] A deeper technical challenge lies in pinpointing the extent of the impact of gene mutations to the microscopic structure of the eye. The effects of gene mutations are not uniformly distributed across the entire organ; rather, they may act only on specific regions or layers. For example, a gene mutation might only affect a single layer of the lens, rather than the entire lens. This lack of precise localization prevents risk assessment from reaching specific tissue levels. Furthermore, this localization challenge leads to another related challenge: establishing a multi-level correlation between gene mutations and microscopic structures. Without this correlation, when analyzing a patient's genetic data, doctors struggle to determine whether a particular mutation will cause lesions in a specific area. For instance, it's impossible to determine whether a gene change only affects the structure and function of a single membrane layer in the macula, rather than the entire macula.

[0005] Therefore, establishing a precise correspondence between gene variations and the microstructure of the eye, and refining the accuracy of risk assessment through this correspondence, has become a key issue that current research urgently needs to address. Solving this problem will directly impact the depth of doctors' assessment of patients' disease risks and will also provide important evidence for the development of personalized treatment plans. Summary of the Invention

[0006] This invention provides a method for assessing the risk of ocular diseases by incorporating ophthalmic genetic information, mainly including:

[0007] By collecting optical coherence tomography images and genetic information data of the patient's eye, the feature information of the eye microstructure is extracted using layered image analysis technology, the correlation of structural damage markers is finely annotated, and irrelevant interference is removed by data cleaning and noise filtering methods, resulting in a preliminary quantitative set of eye structures. Based on the preliminary quantitative set of eye structure and the gene variation sequence in the genetic information, we performed variation site localization accuracy analysis and variation frequency statistical range division, and used the support vector classification boundary method to evaluate the quantitative index of the influence of gene variation on microstructure, and constructed an initial variation-structure association matrix. If the association score of a gene variant in the initial variant and structure association matrix is ​​lower than the pairing accuracy threshold, the association score is dynamically adjusted by iteratively adjusting the step size control and the update frequency of the validation parameters, and the convergence condition is optimized based on the accuracy to generate an enhanced variant and structure association dataset. From the enhanced variant and structure association dataset, high-impact variant type classification criteria are extracted according to the variant record screening rules. In-depth comparison is performed on the association strength between high-impact variants and structures to construct an ocular tissue-specific damage distribution map and determine the distribution characteristics of potential damage areas. Based on the distribution map of specific eye tissue damage, the damage risk patterns are divided by similar pattern grouping technology combined with paired mapping time window analysis. The probability of tissue damage and risk weight are calculated to obtain a detailed distribution map of eye tissue risk. If the probability of damage to a certain tissue in the detailed risk distribution map of eye tissue exceeds the preset warning line, a microscopic disease mechanism pathway map for that tissue is generated by using inter-pattern difference comparison technology combined with data pairing error range analysis to determine the specific affected tissue location.

[0008] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects:

[0009] This invention discloses a method for risk assessment of ocular diseases based on optical coherence tomography (OCT) images and genetic information data of the eye. The method involves collecting patient eye images and genetic data, extracting microstructural features using hierarchical image analysis technology, and constructing a quantitative dataset of ocular structures through data cleaning and noise filtering. Support vector classification is then used to assess the impact of gene variations on structure, generating a variation-structure correlation matrix. For cases where the correlation score is insufficient, parameters are dynamically adjusted to optimize the dataset, thereby screening for high-impact variation types, constructing a tissue-specific damage distribution map, and calculating damage probability and risk weights. If the damage probability exceeds a certain threshold, a disease mechanism pathway map is generated to identify the specific affected tissue sites. This invention achieves a complete integration of the entire process from data acquisition to risk assessment and identification of specific affected tissue sites, significantly improving the accuracy of ocular disease diagnosis and providing a scientific basis for clinical decision-making. Attached Figure Description

[0010] Figure 1 This is a flowchart of a method for assessing the risk of eye diseases by incorporating ophthalmic genetic information, according to the present invention. Figure 2 This is a schematic diagram of an ophthalmic disease risk assessment method that incorporates ophthalmic genetic information according to the present invention; Figure 3 This is another schematic diagram of an eye disease risk assessment method that incorporates ophthalmic genetic information according to the present invention. Detailed Implementation

[0011] To further understand the content of this invention, a detailed description of the invention is provided in conjunction with the accompanying drawings and embodiments. The specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention. It should also be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0012] like Figures 1-3 This embodiment of an ophthalmic genetic information-based method for assessing the risk of eye diseases may specifically include:

[0013] Step S101 involves collecting optical coherence tomography images and genetic information data of the patient's eye, extracting the feature information of the eye's microstructure using layered image analysis technology, finely annotating the correlation of structural damage markers, and removing irrelevant interference by combining data cleaning and noise filtering methods to obtain a preliminary quantitative set of eye structures.

[0014] pass Optical coherence tomography equipmentEye images of patients were collected, and corresponding genetic information records were obtained from a genetic information database to construct an initial joint dataset of eye images and genetic information. For this initial joint dataset, a hierarchical analysis technique was used to decompose the images layer by layer, extracting relevant information about the microstructure of the eye, such as layer thickness, curvature and flatness, abnormal echo points, discontinuities, reflectance intensity distribution, and damage area and volume. This resulted in a set of decomposed microstructural features. The hierarchical analysis technique refers to the automatic identification and separation of highly anatomically related eye tissue images according to their physical layers using deep learning or digital signal processing algorithms on optical coherence tomography (OCT) images of the eye. Based on the decomposed set of microstructural features, classification and labeling were performed using preset damage marking standards to identify potential structural damage areas and determine the distribution information of damage markers. If noise interference exists in the distribution information of damage markers, data cleaning techniques were used to process the labeling results, and irrelevant data was removed using noise filtering methods to obtain a cleaned damage marker dataset. For the cleaned lesion marker dataset, genetic information records were used for association annotation. The correspondence between genetic information and lesion markers was analyzed to construct a comprehensive dataset with association annotation. Structural analysis was performed on the comprehensive dataset with association annotation. A support vector machine algorithm was used to classify the data and extract key quantitative features of ocular structures, resulting in the final quantitative dataset. Based on the final quantitative dataset, joint analysis results of ocular structure and genetic information were generated, completing a comprehensive assessment of the patient's ocular microstructure and outputting a structural quantitative analysis file.

[0015] In one possible implementation, the decomposed set of microstructural features is classified and labeled using preset damage marking standards, such as structural integrity standards and impact feature standards. For example, microstructural feature data, such as the thickness of each layer and reflectivity, are compared one by one with preset standards. A support vector machine algorithm is used to classify the anomalies according to their types, such as grouping structural changes into one category and exudative changes into another. The identified anomaly markers are then bound to their three-dimensional coordinates or tissue layers in the eye image to determine the precise location of these markers in the eye space. Furthermore, areas that meet the damage marking standards are identified and labeled as potential structural damage areas, and the distribution information of the damage markers is determined, including damage type codes, spatial coordinates, and damage area / volume.

[0016] In one possible implementation, the cleaned lesion marker dataset is associated with genetic information records for annotation. For example, a unique association ID is assigned to each patient to ensure a one-to-one correspondence between the lesion markers extracted from their ocular OCT images (such as coordinates, type, and area) and the patient's genetic information records (such as gene loci and mutation type). Then, through multidimensional label mapping, the cleaned lesion marker data (such as "macular RPE layer fracture") is used as the target feature, and the corresponding genetic information records (such as "CFH gene mutation") are used as the association attribute. Finally, the two are encapsulated into a "feature vector pair" to construct a comprehensive dataset after association annotation. This dataset not only contains "where the damage occurred" but also records "what mutation the patient carries," providing underlying data support for the subsequent construction of the association matrix.

[0017] In one possible implementation, the constructed, annotated, comprehensive dataset is first formatted. An algorithm is then used to perform anatomically significant measurements on the decomposed tissue layers, such as calculating the vertical distance from the internal limiting membrane (ILM) to the retinal pigment epithelium (RPE), analyzing the spatial relationships between tissue layers, identifying any abnormal structural overlaps or layer separations, comparing the structural appearance of the same patient in different scan sections, and eliminating abnormal data fluctuations caused by acquisition angles or nystagmus. This completes the structural analysis of the annotated comprehensive dataset. Then, a support vector machine algorithm is used to classify the data, for example, into regions significantly affected by variation or regions with conventional damage. Key quantitative features of ocular structures are extracted, including hierarchical quantitative indicators (such as the mean thickness, standard deviation, and local thickness mutation rate of each anatomical layer), pathological reflectance parameters (such as the reflectance intensity of specific areas in the image data), damage geometric features (such as the precise area, volume, aspect ratio, and distribution density of identified structural damage areas in the fundus coordinate system), and correlation strength scores. Step S102: Based on the preliminary quantitative set of eye structure and the gene variation sequence in the genetic information, perform variation site localization accuracy analysis and variation frequency statistical range division, use the support vector classification boundary method to evaluate the quantitative index of the influence of gene variation on microstructure, and construct the initial variation-structure association matrix.

[0018] Gene variation sequence data were obtained from genetic information databases. Batch processing was used to clean and standardize the sequence data, resulting in a standardized gene variation dataset. Based on this standardized dataset, each variation site was compared and analyzed individually. A pre-established localization algorithm was applied to determine the specific location information of each variation site, obtaining a distribution record of the variation sites. Using this distribution record, variation frequency data in different regions was statistically analyzed. An interval partitioning method was used to classify the frequency distribution range, yielding a classification result. Based on the frequency distribution classification result, combined with microstructural parameters from eye structure data, a support vector machine method was applied to evaluate the correlation between gene variation and microstructure, determining the grading of influence levels. Using this grading data, a correlation matrix between gene variation and eye microstructure was constructed. Matrix imputation techniques, such as singular value thresholding, singular value projection, or least-rank approximate atomic decomposition, were used to fill in missing data, resulting in a complete correlation matrix. Based on the complete correlation matrix, if the influence level of a variation site exceeds a preset threshold, its corresponding microstructural parameters are marked, obtaining a marked focus dataset. By using the labeled key datasets, a priority ranking of the associations between variant sites and microstructures is generated. Data storage technology is used to save the ranking results and determine the focus of subsequent analysis.

[0019] In one possible implementation, the process involves performing a one-by-one alignment analysis on each variant site based on a standardized gene variant dataset. A pre-established localization algorithm is applied to determine the specific location information of each variant site, obtaining a distribution record of the variant sites. Specifically, a coordinate mapping algorithm based on reference genome alignment, such as the Burrows-Wheeler transform (BWT) alignment algorithm, is employed. This algorithm aligns the patient's original sequencing reads or variant data with ophthalmology-related standard genomic reference sequences (such as GRCh38 / hg38) to determine the absolute location of the variant site on a specific chromosome and its relative location among ocular tissue-related genes (such as genes related to the retinal pigment epithelium). The resulting distribution record of the variant sites includes physical coordinates, regional classification, and frequency characteristics.

[0020] In one possible implementation, the distribution records of variant sites are used to statistically analyze the frequency data of variants in different regions. An interval division method, such as threshold segmentation based on frequency distribution, is employed to classify the frequency distribution range, resulting in a frequency distribution classification. The specific classification process is as follows: A classification threshold is set, i.e., pre-establishing classification criteria for the frequency distribution (e.g., rare variants, low-frequency variants, common variants): rare variant intervals with a frequency below 0.1% are marked as "high potential pathogenicity risk level"; low-frequency variant intervals with a frequency between 0.1% and 5% are marked as "moderate associated risk level"; and common variant intervals with a frequency exceeding 5% are marked as "basic genetic background level". Mapping is then performed by substituting the frequency value of each variant site into the aforementioned intervals to obtain the classification result for each site. Finally, a list with category labels is obtained, such as "site A: belongs to the high-frequency interval; site B: belongs to the rare interval". Through this interval division, the system can effectively screen out key factors that truly have a significant impact on the microstructure of the eye from a massive amount of genetic variations.

[0021] In one possible implementation, the classification results of frequency distributions, such as rare variants, low-frequency variants, and common variants, are paired with microstructural parameters in ocular structural data, such as tissue layer thickness, reflectivity and density, damage geometry, and topological association attributes. An analysis matrix is ​​constructed, and a support vector machine (SVM) method is applied to map gene variant data and microstructural parameters to a high-dimensional space. By finding the optimal classification boundary, it is identified which changes in microstructural parameters (e.g., thinning of a certain layer) are significantly correlated with the frequency of a specific gene variant. The association score of each variant site relative to microstructural features is calculated, and a preset influence threshold is set for grading to determine the level of influence (e.g., high, medium, low influence). In step S103, if the association score of a gene variant in the initial variant-structure association matrix is ​​lower than the pairing accuracy threshold, the association score is dynamically adjusted by iteratively adjusting the step size and the update frequency of the validation parameters. The convergence condition is optimized based on accuracy to generate an enhanced variant-structure association dataset.

[0022] The process begins by acquiring initial gene variant and structural association matrix data. For each gene variant, an initial association score is compared. If the score is below a preset accuracy threshold, it is marked as an object to be processed, resulting in a set of variant scores to be optimized. For this set, an iterative adjustment method is used. By gradually changing the step size control parameter, the score change trend after each iteration is calculated to determine the direction of score improvement. Based on the direction of score improvement, the parameter update frequency is adjusted. Multiple calculations are performed on the adjusted parameter values ​​to obtain updated association score data, and it is determined whether the updated association scores are close to the preset accuracy threshold. If the updated association scores still do not reach the preset accuracy threshold, a dynamic enhancement mechanism combined with a support vector machine algorithm is used to classify the score data, resulting in classified score groups. For each classified score group, the score distribution characteristics are obtained. By comparing the matching degree between the distribution characteristics and the convergence condition, the optimal score adjustment scheme is determined. Based on the optimal score adjustment scheme, the original gene variant and structural association matrix is ​​processed to generate an enhanced variant and structural association dataset, and it is determined whether it meets the final accuracy requirements. By comparing the augmented dataset with the initial matrix, the specific changes in data augmentation are recorded, and the final optimization result is determined.

[0023] In one possible implementation, the initial gene variant and structural association matrix data is obtained. A preliminary comparison is performed on the association score of each gene variant. If the score is lower than a preset accuracy threshold, it is marked as an object to be processed, resulting in a set of variant scores to be optimized. The association score is a quantitative confidence assessment of the correspondence between gene variants and ocular microstructural damage. The calculation logic is as follows:

[0024] Residual analysis: Calculates the deviation between the predicted values ​​of the association model established in S102 and the actual observation values ​​extracted in S101. If a gene variant is predicted to cause structural damage, but the deviation is large in the actual image, the score will be low.

[0025] Support Vector Machine (SVM) probability output: The classification probability value or functional distance from the sample point to the hyperplane output by the SVM algorithm. The greater the distance (or the closer the probability value is to 1), the higher the certainty of the association, and the higher the score.

[0026] Example of calculation formula: The score is usually derived by weighting statistical significance, effect size and sample consistency.

[0027] (where w is the weighting coefficient)

[0028] The preset accuracy threshold is typically based on confidence level thresholds such as 0.85 or 0.9, error tolerance such as 15%-20%, and dynamic adjustment attributes.

[0029] In one possible implementation, for the set of variant scores to be optimized, an iterative adjustment method is used. This involves gradually changing step size control parameters, such as the displacement increment of the support vector classification boundary or the correction ratio of the spatial coordinate mapping, calculating the score change trend after each iteration, and determining the direction of score improvement. For example, the current iteration score can be calculated using differential calculation. Compared with the previous generation selection score The difference Slope / gradient analysis The first derivative (slope) varies with step size, if And the continued increase indicates that the score is trending upwards, and the current adjustment is effective. If The decreasing or negative values ​​indicate that the score is converging or declining, and the association accuracy is deteriorating. To prevent interference from noise in single data iterations, a moving average method is typically used to calculate the average rate of change over the most recent k iterations. Based on the calculated score change trend, the direction with the largest gradient is identified. If increasing a certain weight parameter leads to a score increase, this direction is determined as the direction of improvement; if the score decreases, the step size parameter is immediately adjusted in the opposite direction. The next large-scale validation parameter update is performed along the direction of the fastest gradient increase (score increase). When the score increase becomes gradual (i.e., the change trend approaches zero), and the score exceeds a preset pairing accuracy threshold, it is determined that the direction has reached the optimal mapping, and the generational selection process for this site is completed.

[0030] In one possible implementation, adjusting the parameter update frequency based on the direction of score improvement follows the following algorithmic path:

[0031] 1. Directional Consistency Detection: Calculate the angle between the direction vectors. If the angle is very small (approaching 0 degrees), it indicates that the search direction is very stable. Then, multiply the update frequency by a magnification factor (e.g., F=Fx1.2).

[0032] 2. Oscillation Feedback Suppression: If the direction of score improvement changes drastically (e.g., from positive to negative), it indicates that the previous update frequency was too fast, leading to a "leapfrog" deviation. In this case, the frequency should be immediately reset and the weights reduced, forcing the system to stop and perform directional calibration.

[0033] In one possible implementation, if the updated association score still does not reach the preset accuracy threshold, a dynamic boosting mechanism is used. This mechanism, combined with a Support Vector Machine (SVM) algorithm, divides the scoring data into a convergent high-precision association region and a region requiring further iteration. The SVM applies a kernel function (such as a Gaussian kernel RBF) to map the data to a higher-dimensional space. Based on the SVM classification results, the system assigns a new label to each data point, resulting in classified score groups such as high-scoring samples and low-scoring but potentially valuable samples. This dynamic boosting mechanism is an adaptive feedback optimization algorithm that monitors score trends in real time. If the score is improving, the mechanism automatically maintains or increases the optimization intensity; if the score stagnates, it changes the search strategy (e.g., adjusting the step size).

[0034] In one possible implementation, for the classified scoring groups, the scoring distribution characteristics within each group are obtained. The optimal scoring adjustment scheme is determined by comparing the matching degree between the distribution characteristics and the convergence criteria. The scoring distribution characteristics refer to the statistical morphology and clustering attributes exhibited by all associated scoring data points within a specific scoring group, specifically including: central tendency (median / mean), dispersion (variance / standard deviation), probability density morphology, and local maxima. The convergence criteria are the termination criteria used to measure whether iterative optimization can stop and whether an "optimal solution" has been found. Specifically, they refer to:

[0035] Score increment threshold: the score increase between two consecutive generations ( Less than a very small preset value (e.g.) This indicates that continued iteration can no longer significantly improve accuracy.

[0036] Accuracy assessment: The center value of the score distribution has stabilized above the "pairing accuracy value".

[0037] Gradient approaching zero: When the gradient of the loss function of the Support Vector Machine (SVM) decreases to near zero at the classification boundary, it means that the globally optimal parameter update direction has been found.

[0038] Stability constraint: The dispersion (variance) of the rating distribution is reduced to a preset stable range, and no more violent oscillations occur.

[0039] In one possible implementation, the process of processing the original gene variation and structural association matrix according to the optimal scoring adjustment scheme, generating an enhanced variation and structural association dataset, and determining whether the final accuracy requirement is met typically includes the following three levels of indicators:

[0040] Global association confidence: requires that the association scores of the vast majority (e.g., more than 95%) of high-impact variant sites in the dataset must remain stable above a preset threshold.

[0041] Residual convergence: In the last iteration, the average error (residual) between the predicted value of the structure quantization feature and the actual image observation value must be less than a preset minimum value (e.g., <5%).

[0042] Logical consistency verification: Based on the logical verification rules mentioned in step S105, determine whether the generated associated data conforms to medical logic (e.g., whether structural damage caused by specific gene mutations conforms to known pathological characteristics of eye diseases).

[0043] The judgment process typically follows the following logical loop:

[0044] 1. Individual Compliance Check: Check the scores of all sites optimized by S103. If some sites still fail to meet the criteria, the system will assess whether the "maximum number of iterations" has been reached. If the number of iterations has not been reached, optimization continues; if the number of iterations has been reached and the site still fails to meet the criteria, the site may be marked as low confidence or removed.

[0045] 2. Distribution Feature Matching: The characteristics of the rating distribution are compared with the convergence criteria. If the rating distribution of the entire group exhibits a high mean and low variance, it is considered to meet the accuracy requirements.

[0046] 3. Result Output Judgment: If the requirements are met, a formal enhanced variant and structure association dataset is generated and handed over to step S104 to extract classification criteria for high-impact variant types. If the requirements are not met, the process goes back to S102 or S101 to check whether the initial data cleaning and noise filtering were incomplete or whether there is a systematic bias in the localization algorithm.

[0047] Step S104: From the enhanced variant and structure association dataset, extract the classification criteria for high-impact variant types according to the variant record screening rules, perform in-depth comparison of the association strength between high-impact variants and structures, construct an ocular tissue-specific damage distribution map, and determine the distribution characteristics of potential damage areas.

[0048] From a pre-established variant and structural association database, a raw dataset containing variant types and structural associations is obtained. High-impact labels are initially labeled to obtain a preliminary set of variant records. For this preliminary set, variant types are classified using screening rules, and high-impact variants are stratified according to classification criteria to determine a core subset of high-impact variants. From this core subset, structural association data related to ocular tissues are extracted, and the strength of the associations is analyzed. If the association strength exceeds a preset threshold, it is marked as a key association point, resulting in a set of key association points. The preset threshold is a quantitative standard for measuring the coupling degree between gene variants and tissue structural damage, typically determined based on a statistical confidence threshold (requiring an association score of 0.85 or 0.9 or higher), a bias explanatory power threshold (0.75), and clinicopathological constants. Based on the set of key association points, a specific damage distribution model of ocular tissues is constructed, mapping the distribution of specific damage in different regions and generating a preliminary distribution map. Analysis of the preliminary distribution map identifies the clustering locations of damage areas, determines areas with high potential risk, and outputs a description of the distribution characteristics of the damage areas. The distribution characteristics of the damaged areas are obtained, and combined with the structural correlation data of ocular tissues, a support vector machine algorithm is used to predict potential risk areas and determine the boundary range of high-risk areas. Based on the boundary range of high-risk areas, the damage area labels in the distribution map are updated to generate the final ocular tissue-specific damage distribution map.

[0049] In one possible implementation, the initially screened variant record set is categorized using screening rules. High-impact variants are stratified according to classification criteria, specifically into damage severity stratification, tissue specificity stratification, and variant type stratification. Variants that pass the screening rules are compared matrixwise along two dimensions: damage severity and association strength. Variants with high-weight intersections—those simultaneously satisfying high association scores and causing damage to key tissue sites—are selected into the core subset. The explanatory power of these variants for the overall ocular injury risk in patients is assessed. If a combination of several variants can explain more than 80% of structural abnormalities, this group of variants constitutes the core subset. Each variant within the core subset is assigned a specific risk weight, thus obtaining the core subset of high-impact variants. The preset screening rules specifically include: an association strength rule: setting a minimum association score threshold (based on the S103 optimized score); only variants with scores higher than this value are included in the screening range. A functional region rule: preferentially retaining variants located in gene exons, splicing sites, or important regulatory regions (promoters, enhancers). Population frequency rule: Based on the interval division results of S102, rare or low-frequency variants are screened out. In ophthalmic genetic diseases, low-frequency variants often have a higher pathogenicity. Consistency rule: The variant is required to show a stable structural damage association across multiple samples or generations.

[0050] In one possible implementation, the specific damage distribution model of ocular tissue is constructed based on a set of key correlation points. Specifically, this is achieved by establishing a high-level mapping between gene mutations and anatomical space. For example, using the ocular microstructural coordinates extracted in S101 (such as three-dimensional spatial points of various retinal layers) and the mutation site locations determined in S102, a "genetic-anatomical" joint coordinate system is established. For each key correlation point, the influence weight of the mutation on specific tissue units (such as the macula RPE layer and the optic disc RNFL layer) is calculated based on its correlation strength, thus establishing a specific damage distribution model. ,in Represents anatomical location. This represents the characteristic of variation. The model is used to describe the predicted probability of damage to various tissue sites in the eye when a specific subset of core variations is present.

[0051] The mapping process for the distribution of specific damage in different regions involves transforming the abstract numerical values ​​calculated by the model into data with geographical distribution characteristics. Specifically, ocular tissues (such as the retina) are divided into several gridded feature units. Through probability intensity mapping, the association strength values ​​of key association points are filled into the corresponding spatial units. That is, high-association areas are mapped to high-density / high-intensity damage prediction values; low-association areas are mapped to background noise or low-risk values. The "specificity" of different gene variations to different regions is analyzed. For example, some variations specifically map to the fovea of ​​the macula, while others map to the peripheral retina.

[0052] The distribution map is a visual representation of risk assessment. Its generation process is as follows: Using a similar pattern grouping technique, regions with similar damage characteristics are clustered to generate continuous risk contour lines or heatmap distributions. Damage prediction maps from different anatomical layers (such as the nerve fiber layer and photoreceptor cell layer) are vertically overlaid to form a preliminary three-dimensional damage extent map. By setting an effective threshold, regions exceeding this threshold are delineated to determine the preliminary damage extent, thus generating the preliminary distribution map.

[0053] In one possible implementation, the analysis of the preliminary distribution map identifies the clustering locations of damaged areas, i.e., finding clusters with abnormally high probability density of damage in the ocular tissue space, determining areas with high potential risk, and outputting a description of the distribution characteristics of the damaged areas. In determining areas with high potential risk, the assessment considers not only the current degree of damage clustering but also the weighting of genetic influences for a comprehensive evaluation.

[0054] Association strength weighting determination: The system will retrieve the influence degree quantification indicators determined in S102 and S103. If the variant site corresponding to a cluster location belongs to the core subset of high-impact variants and has an extremely high association score (correlation), then the region is determined to be an extremely high-risk area.

[0055] Damage concordance analysis: Comparing atlases from multiple anatomical layers. If a region shows clusters of damage across multiple retinal layers (such as the inner and outer layers), its potential risk level is significantly increased.

[0056] Logical verification rule prediction: Apply the pre-established logical verification rules mentioned in S105 to determine whether the clustered area is located on a known disease evolution path (e.g., a lesion cluster located near the fovea of ​​the macula, even if small in size, will be judged as a high-risk area due to its huge impact on vision).

[0057] Finally, the system will output a detailed distribution feature report, which describes the distribution features of the damaged area. The report includes: cluster point coordinates (the three-dimensional location where the damage is most concentrated), risk density value (quantifying the degree of risk clustering in the area), and main driving factors (specific gene mutations that cause the high risk in the area).

[0058] In one possible implementation, the distribution characteristics of the acquired damage area are described, and combined with structural correlation data of ocular tissue, a support vector machine (SVM) algorithm is used to predict potential risk areas and determine the boundary range of high-risk areas. Specifically, a feature vector is constructed for each spatial unit in the atlas, containing: spatial coordinates (x, y, z), correlation strength score, damage density value (i.e., the distribution frequency of surrounding clusters), and anatomical weight (i.e., the importance coefficient of the tissue layer to which it belongs). The classification model is trained by using data labeled as key correlation points as positive samples and background areas as negative samples. The SVM algorithm automatically learns the distribution characteristics of these points in multidimensional space. Since ocular damage boundaries are often irregular (e.g., the edges of macular degeneration are map-like), the system applies RBF (radial basis function) to map the data to a high-dimensional space to identify complex nonlinear boundaries and complete the prediction of potential risk areas. Furthermore, the SVM outputs a confidence score (i.e., the probability that the point belongs to the high-risk category) for each pixel / voxel in the atlas. The system seeks an optimal hyperplane that maximizes the geometric interval between positive and negative samples (high-risk points and normal points). This hyperplane, when projected back onto the ocular anatomical coordinate system, forms the closed boundary line of the high-risk region. Finally, support vector locking and relaxation variable adjustment are used to dynamically correct the boundary range, thus determining the boundary extent of the high-risk region. The determined boundary extent is input into step S105 to calculate the probability of substantial lesions occurring in the tissue within this range in the future. In step S105, based on the ocular tissue-specific damage distribution map, similar pattern grouping technology combined with paired mapping time window analysis is used to classify damage risk patterns, calculate tissue damage probability and risk weight, and obtain a refined ocular tissue risk distribution map.

[0059] By collecting data from ocular tissues, raw information on damage distribution is obtained, an initial distribution map is constructed, and the preliminary damage range is determined. Based on the initial distribution map, similar pattern grouping techniques such as feature vectorization, K-Means++, hierarchical clustering, and topological matching are used to classify the damage distribution, resulting in a set of classified damage patterns. For example, Category 1 is foveal involvement (high risk of blindness), Category 2 is peripheral retinal degeneration (moderate risk of progression), and Category 3 is isolated microvascular association (risk requiring observation). For the classified damage pattern set, the changing trend within a time window is analyzed using a pairwise mapping method to determine the dynamic evolution law of the damage pattern. If the changing trend within the time window exceeds a preset changing trend threshold, the risk pattern is reclassified to obtain a more accurate risk pattern classification result. By processing the risk pattern classification results, the damage probability is calculated by fusing pattern division, dynamic evolution analysis, and quantitative indicators. The damage probability is correlated and matched with structural hierarchical risk weights to determine the potential threat level of each ocular tissue. Based on the calculated damage probability results and combined with the data analysis of risk weights, a refined ocular tissue risk distribution map is constructed, resulting in the final distribution map. By employing pre-established logical verification rules, consistency checks are performed on the final distribution map to determine the rationality of the map data and complete the refined presentation of risk distribution.

[0060] In one possible implementation, the analysis of the changing trends within a time window, using a pairing mapping method on the classified set of injury patterns—that is, extracting trends by observing differential changes in data within a preset time window—to determine the dynamic evolution of injury patterns, specifically involves mapping the boundaries of the potential risk area determined in S104 to the three-dimensional coordinate system of ocular tissue at different time points, ensuring that the analysis focuses on the same lesion area. The correlation strength, injury density, and microstructural parameters (such as thickness) within this area are compared before and after. Specific subsets of core variants are paired with the lesion response of this area over time; for example, comparing the injury manifestations of patients with the same variant background at different stages. The determination of the dynamic evolution of injury patterns essentially involves pattern classification and grade prediction for injury trends, with the specific determination logic as follows:

[0061] Evolution rate determination: linear growth type, that is, the damage increases slowly and steadily over time, usually corresponding to chronic genetic risk; exponential burst type, that is, the probability of damage increases sharply after a certain point in time, which is determined to be a high-risk evolution pattern and should be marked as an extremely high potential threat immediately.

[0062] Evolution path determination: interlayer penetration law, that is, observing whether the damage spreads longitudinally from the RPE layer to the photoreceptor layer; centripetal law, that is, observing whether the damage is moving towards the core functional areas such as the fovea of ​​the macula.

[0063] Determining the stability state: If the trend of change flattens out (slope approaches zero) in the later part of the window, it is determined to be stabilizing; if it continues to rise, it is determined to be continuing to progress.

[0064] In one possible implementation, if the trend of change within the time window exceeds a preset trend threshold, the risk pattern is reclassified to obtain a more accurate risk pattern classification result. The preset trend threshold is used to define the boundary between normal evolution and abnormal mutation. It is usually composed of a threshold matrix consisting of multiple indicators, including an area expansion rate threshold (the expansion ratio of high-risk areas exceeds 15%-20% within a unit time window), a probability slope threshold (the derivative of the damage probability with time exceeds a preset constant of 0.15 / month), an association strength drift threshold (the association score between the core variant subset and structural damage fluctuates by more than 10%), and an anatomical span threshold (the penetration rate of damage from one anatomical level (such as the RPE layer) to another level (such as the photoreceptor cell layer) exceeds the biologically predicted mean).

[0065] The risk pattern reclassification involves, specifically, when a trend exceeds a certain threshold, the system deems the original similar pattern groupings invalid and necessitates initiating a dynamic feedback-based reconstruction process.

[0066] A. Subgrouping into High-Dynamic Subgroups: The original damage patterns are further "decomposed." If, within the same pattern group A, some regions evolve extremely rapidly while others remain relatively stable, the system will isolate the rapidly evolving regions and establish a separate burst-type risk subgroup. This prevents the characteristics of high-risk regions from being masked by mediocre average values.

[0067] B. Topology Reconstruction: Re-clustering is performed based on the latest spatial distribution pattern, and SVM is applied to redefine high-risk boundaries. If the original point-like patterns have merged into patch-like patterns, the system will discard the point-like classification and redefine it as a merging, map-like shrinking pattern. K-Means++ or hierarchical clustering algorithms are then invoked again, but the time rate of change is incorporated as a core feature weight in the calculation.

[0068] C. Genetic Driver Remapping: Examine whether new variant sites have shown higher association during evolution. If new key association points are found to have a higher explanatory power for the current trend, the damage pattern is reclassified under the corresponding genetic core subset driver group.

[0069] In one possible implementation, the process of constructing a refined risk distribution map of ocular tissue based on the calculation results of the damage probability and the data analysis of risk weights to obtain the final distribution map is as follows: decompose the ocular structure into microscopic levels, use the model to determine the basic weights of each structural level, establish a correspondence matrix between the mutation type and the structural level, and dynamically adjust the risk weight of the part according to the correlation strength between the mutation and the structure.

[0070] In one possible implementation, the pre-established logical verification rules are used to perform consistency checks on the final distribution map, determine the rationality of the map data, and complete the refined presentation of the risk distribution. The pre-established logical verification rules mainly include the following three dimensions of criteria:

[0071] Variation-damage correspondence criterion: verify whether the detected tissue damage site matches the microstructural level corresponding to the high-influence variation in the genetic information.

[0072] Spatial distribution continuity criterion: Verify whether the boundary range of the damaged area (predicted by SVM) conforms to the physiological structural continuity of the eye tissue, and avoid the occurrence of harmful scattered distribution that does not conform to biological logic.

[0073] Probability-Weight Coordination Criterion: Verify whether there is a logical contradiction between the calculated damage probability and the risk weight of the organization. For example, if the core functional area (high weight) has an extremely high probability of damage, but the path diagram generation logic of S106 is not triggered, it is considered a violation of the verification rule.

[0074] The consistency check of the final distribution map to determine the rationality of the map data specifically involves scanning the map using pre-established logical verification rules. If the data meets the rules (e.g., the damage probability does not exceed the preset warning line and the distribution conforms to anatomical characteristics), it is deemed reasonable. The rationality judgment criteria are: deviation range judgment, whether the feature deviation is within the "data pairing error range"; convergence condition judgment, whether the adjustment of the correlation score meets the convergence condition of accuracy optimization; and consistency passing. The map data is deemed reasonable only when there is no logical conflict between the precisely located tissue site, the microscopic disease mechanism path map, and the future trend prediction data. In step S106, if the damage probability of a certain tissue in the refined ocular tissue risk distribution map exceeds the preset warning line, a microscopic disease mechanism path map for that tissue is generated by using inter-pattern difference comparison technology combined with data pairing error range analysis to determine the specific affected tissue site.

[0075] By analyzing the risk distribution data of ocular tissues, tissue units with damage probabilities exceeding a preset threshold are identified, thus pinpointing key areas of initial interest. Based on the damage probability data of the identified areas, an inter-pattern difference comparison method is used to extract feature deviations between different tissue units, identifying potential abnormal patterns. If the extracted feature deviations exceed a preset range, data pairing technology is used to calibrate the error range of the deviation data, resulting in a calibrated abnormal pattern distribution. Data pairing technology is a coordinate alignment and correlation analysis technique used to improve data consistency and accuracy, primarily used to match data from different sources or of different properties (such as structural features extracted from OCT images and records of genetic variation) in a one-to-one spatial or logical dimension. For the calibrated abnormal pattern distribution, a microscopic-level disease mechanism analysis framework is constructed. This framework aims to simulate and analyze the logical relationships between pathological features in microscopic tissues, providing an anatomical and pathological reference system for subsequent localization. Corresponding path diagrams are drawn to determine the specific affected tissue sites. Key node data is obtained from the drawn path diagrams to analyze the propagation path of the disease mechanism at the microscopic level, determining the distribution of the core damaged area. By analyzing the distribution of the core damaged areas, a targeted detection plan is generated to identify tissue sites requiring further analysis, thus accurately locating the risk areas. If the data on the accurately located tissue sites deviates from the initial risk distribution data, the risk distribution map is updated using information comparison technology to obtain the final damage distribution result.

[0076] In one possible implementation, based on the damage probability data of the locked area, a pattern difference comparison method is used to extract the feature deviations between different tissue units and determine potential abnormal patterns. Specifically, a horizontal comparison is performed between different ocular tissue units (such as different layers of the retina or adjacent anatomical regions), and feature data of microstructures are extracted from these units. For example, the root mean square error or correlation coefficient of adjacent tissue unit data sequences is calculated, the feature deviations between different tissue units are quantified, and tissue units with feature deviations significantly greater than the normal statistical range are identified as potential abnormal patterns.

[0077] In one possible implementation, key node data, such as pathological initiation points, tissue association intersections, characteristic mutation points, and propagation crossover points, are obtained from the drawn path diagram. The system utilizes these key nodes in the path diagram to analyze the propagation trajectory of the disease mechanism at the microscopic level (e.g., how damage to cells in one layer logically leads to structural degradation in adjacent layers). By analyzing the density and directionality of the propagation paths, the system identifies the areas with the most concentrated pathological impact and the strongest correlation, defining these as the core damaged areas, thereby determining the distribution of the core damaged areas.

[0078] In one possible implementation, the targeted detection plan is generated based on the distribution of the core damaged areas to pinpoint the tissue sites requiring further analysis and achieve precise location of risk areas. This targeted detection plan is a dynamically generated instruction set based on the core damaged area distribution map generated in the preceding steps. Specifically, the generation process involves: using the coordinates of the core damaged areas determined in the path map to set the spatial range for a high-precision secondary scan; setting the sensitivity parameters of the detection algorithm based on pathological features extracted from key node data (such as specific structural fractures, leakage, or atrophy); and generating instructions to pixel-level align the microscopic disease mechanism path map with the original OCT image to ensure the detection plan can directly guide clinical observation.

[0079] The process of identifying tissue sites requiring further analysis employs a precise localization logic that moves from speculation to empirical evidence. Specifically, the system analyzes the transmission path of the disease mechanism at the microscopic level, identifies the most severely affected logical checkpoints, and thus identifies the tissue targets requiring further analysis. The generated detection scheme is then applied to perform highly refined data extraction on the target areas, completing the precise localization of the risk regions. The system compares the precisely located tissue site data with the original data in the initial risk distribution map in real time. If there are spatial or feature discrepancies, the system will force an update to the risk distribution map through information comparison technology, thereby correcting and ultimately identifying the most accurate damaged tissue site.

[0080] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for assessing the risk of ocular diseases by incorporating ophthalmic genetic information, characterized in that, The method includes: By collecting optical coherence tomography images and genetic information data of the patient's eye, the characteristic information of the eye microstructure is extracted, the distribution information of damage markers is determined, and data cleaning and noise filtering are performed. Combined with genetic information, association annotation and analysis are carried out to obtain a preliminary quantitative set of eye structures. Based on the preliminary quantitative set of eye structure and the gene variation sequence in the genetic information, we performed variation site localization accuracy analysis and variation frequency statistical range division, and used the support vector classification boundary method to evaluate the quantitative index of the influence of gene variation on microstructure, and constructed an initial variation-structure association matrix. If the association score of a gene variant in the initial variant and structure association matrix is ​​lower than the pairing accuracy threshold, the association score is dynamically adjusted by iteratively adjusting the step size control and the update frequency of the validation parameters, and the convergence condition is optimized based on the accuracy to generate an enhanced variant and structure association dataset. From the enhanced variant and structure association dataset, high-impact variant type classification criteria are extracted according to the variant record screening rules. In-depth comparison is performed on the association strength between high-impact variants and structures to construct an ocular tissue-specific damage distribution map and determine the distribution characteristics of potential damage areas. Based on the distribution map of specific eye tissue damage, the damage risk patterns are divided by similar pattern grouping technology combined with paired mapping time window analysis. The probability of tissue damage and risk weight are calculated to obtain a detailed distribution map of eye tissue risk. If the probability of damage to a certain tissue in the detailed risk distribution map of eye tissue exceeds the preset warning line, a microscopic disease mechanism pathway map for that tissue is generated by using inter-pattern difference comparison technology combined with data pairing error range analysis to determine the specific affected tissue location.

2. The method for assessing the risk of ocular diseases by combining ophthalmic genetic information according to claim 1, characterized in that, The process involves acquiring optical coherence tomography (OCT) images and genetic information data of the patient's eye, extracting feature information of the eye's microstructure, determining the distribution information of damage markers, and performing data cleaning and noise filtering. Combined with genetic information, association annotation and analysis are then performed to obtain a preliminary quantitative set of eye structures, including: By collecting eye image data from patients and combining it with the corresponding genetic information records obtained from the genetic information database, an initial joint dataset of eye images and genetic information is constructed. For the joint dataset, a hierarchical parsing technique is used to decompose the image layer by layer, extract relevant information on the microstructure of the eye, and obtain a set of decomposed microstructure features. Based on the decomposed set of microstructural features, classification and labeling are performed using a pre-defined damage labeling standard to identify potential structural damage areas and determine the distribution information of damage labels. If there is noise interference in the distribution information of damage markers, the annotation results are processed by data cleaning techniques and noise is filtered to obtain a cleaned damage marker dataset. For the cleaned damage marker dataset, we combined genetic information records for association annotation, analyzed the correspondence between genetic information and damage markers, and constructed a comprehensive dataset after association annotation. By performing structural analysis on the comprehensive dataset after association and annotation, the support vector machine algorithm is used to classify the data and extract key quantitative features of eye structure to obtain the final quantitative dataset. Based on the final quantified dataset, a joint analysis result of eye structure and genetic information is generated, and a structural quantification analysis file is output.

3. The method for assessing the risk of ocular diseases by combining ophthalmic genetic information according to claim 1, characterized in that, Based on the preliminary quantitative set of eye structure and the gene variation sequences in the genetic information, the method performs variation site localization accuracy analysis and variation frequency statistical range division, uses the support vector classification boundary method to evaluate the quantitative index of the impact of gene variation on microstructure, and constructs an initial variation-structure association matrix, including: By obtaining gene variation sequence data from genetic information databases, and using batch processing to clean and standardize the format of the sequence data, a standardized gene variation dataset is obtained. Based on the standardized gene variation dataset, a comparative analysis was performed on each variation site, and a pre-established localization algorithm was applied to determine the location information of each variation site and obtain the distribution record of the variation sites. By recording the distribution of variant sites, statistical analysis of the variation frequency data in different regions is conducted. The frequency distribution range is then classified using an interval division method to obtain the classification results of the frequency distribution. Based on the classification results of frequency distribution, combined with the microstructural parameters in the eye structure data, the support vector machine method is applied to evaluate the correlation between gene variation and microstructure, and to determine the graded data of the degree of influence. By using graded data on the degree of impact, a correlation matrix between gene variations and eye microstructures was constructed. Matrix imputation techniques were then used to fill in the missing data to obtain complete correlation matrix data. Based on the complete association matrix data, if the influence of a certain mutation site exceeds a preset threshold, its corresponding microstructural parameters are marked to obtain the marked key focus dataset. By using the labeled, focused datasets, a priority ranking of the associations between variant sites and microstructures is generated and saved.

4. The method for assessing the risk of ocular diseases by combining ophthalmic genetic information according to claim 1, characterized in that, If the association score of a gene variant in the initial variant-structure association matrix is ​​lower than the pairing accuracy threshold, then the association score is dynamically adjusted by iteratively adjusting the step size control and validation parameter update frequency, and the convergence condition is optimized based on accuracy to generate an enhanced variant-structure association dataset, including: Obtain the initial gene variation and structural association matrix data. If the association score of each gene variation is lower than the preset accuracy threshold, it is marked as an object to be processed, and a set of variation scores to be optimized is obtained. For the set of variant scores to be optimized, an iterative adjustment method is adopted. By gradually changing the step size control parameter, the score change trend after each iteration is calculated to determine the direction of score improvement. Based on the direction of score improvement, adjust the parameter update frequency, perform multiple calculations on the parameter values ​​after frequency adjustment, obtain updated related score data, and determine whether it is close to the preset accuracy threshold. If the updated associated scores still do not reach the preset accuracy threshold, a dynamic boosting mechanism is used to classify the score data using a support vector machine algorithm to obtain the classified score groups. For the classified rating groups, the rating distribution characteristics within each group are obtained. By comparing the matching degree between the distribution characteristics and the convergence conditions, the optimal rating adjustment scheme is determined. Based on the optimal scoring adjustment scheme, the original gene variation and structure association matrix is ​​processed to generate an enhanced variation and structure association dataset, and it is determined whether it meets the final accuracy requirements. By comparing the augmented dataset with the initial matrix, the specific changes in data augmentation are recorded, and the final optimization result is determined.

5. The method for assessing the risk of ocular diseases by combining ophthalmic genetic information according to claim 1, characterized in that, The process involves extracting high-impact variant type classification criteria from the enhanced variant and structure association dataset based on variant record screening rules, performing in-depth comparisons of the association strength between high-impact variants and structures, constructing an ocular tissue-specific damage distribution map, and determining the distribution characteristics of potential damage areas, including: From a pre-established database of variant and structural associations, a raw dataset containing variant types and structural associations is obtained. High-impact labels are initially labeled to obtain a preliminary set of variant records. For the initially screened set of variant records, the variant types are classified using screening rules, and high-impact variants are stratified according to the classification criteria to determine the core subset of high-impact variants. From the core subset of high-impact variants, structural association data related to ocular tissues are extracted. If the association strength exceeds a preset threshold, it is marked as a key association point, thus obtaining a set of key association points. Based on the set of key correlation points, a specific damage distribution model of ocular tissue is constructed to map the distribution of specific damage in different regions and generate a preliminary distribution map. By analyzing the preliminary distribution map, the clustering locations of damaged areas are identified, areas with high potential risk are determined, and a description of the distribution characteristics of the damaged areas is output. The distribution characteristics of the damaged area are obtained, and combined with the structural correlation data of the eye tissue, the support vector machine algorithm is used to predict the potential risk area and determine the boundary range of the high-risk area. Based on the boundary range of high-risk areas, the labeling of damaged areas in the distribution map is updated to generate the final distribution map of ocular tissue-specific damage.

6. The method for assessing the risk of ocular diseases by combining ophthalmic genetic information according to claim 1, characterized in that, Based on the ocular tissue-specific injury distribution map, similar pattern grouping technology combined with paired mapping time window analysis is used to classify injury risk patterns, calculate tissue injury probability and risk weight, and obtain a detailed ocular tissue risk distribution map, including: By collecting data from ocular tissues, we can obtain raw information on the distribution of damage, construct an initial distribution map, and determine the preliminary extent of damage. Based on the initial distribution map, the damage distribution is classified using a similar pattern grouping technique to obtain a set of classified damage patterns. For the classified set of damage patterns, the changing trends within the time window are analyzed using the pairing mapping method to determine the dynamic evolution of the damage patterns. If the trend of change within the time window exceeds the preset threshold, the risk pattern will be reclassified to obtain a more accurate risk pattern classification result. By processing the risk pattern classification results, the probability of damage is calculated, and the potential threat level of each ocular tissue is determined. Based on the calculation results of the damage probability and combined with the data analysis of risk weights, a refined risk distribution map of ocular tissues is constructed, resulting in the final distribution map.

7. The method for assessing the risk of ocular diseases by combining ophthalmic genetic information according to claim 1, characterized in that, If the probability of damage to a certain tissue in the refined ocular tissue risk distribution map exceeds a preset warning line, then through inter-pattern difference comparison technology combined with data pairing error range analysis, a microscopic disease mechanism pathway map for that tissue is generated to determine the specific affected tissue sites, including: By analyzing the risk distribution data of eye tissues, tissue units with a damage probability exceeding a preset threshold are identified, and key areas of initial concern are identified. Based on the damage probability data of the locked area, the inter-pattern difference comparison method is used to extract the characteristic deviations between different tissue units and identify potential abnormal patterns. If the extracted feature deviation exceeds the preset range, the deviation data is calibrated by combining data pairing technology to obtain the calibrated abnormal pattern distribution. Based on the calibrated distribution of abnormal patterns, a disease mechanism analysis framework at the microscopic level is constructed, corresponding path diagrams are drawn, and the specific affected tissue sites are identified. Key node data are obtained from the drawn path diagram to analyze the transmission path of the disease mechanism at the micro level and determine the distribution of the core damaged area. By analyzing the distribution of the core damaged areas, a targeted detection plan is generated to identify the tissue sites that require further analysis and to accurately locate the risk areas. If there is a discrepancy between the precisely located tissue location data and the initial risk distribution data, the risk distribution map is updated using information comparison technology to obtain the final damage distribution result.

Images