Holographic eye multi-dimensional disease screening system and method

Abstract

The invention discloses a holographic eye multi-dimensional disease screening system and method, and relates to the technical field of ophthalmology medical treatment, and the system comprises a data collection and preprocessing module which is used for collecting multi-dimensional clinical data through professional equipment, optimizing image data through an image processing algorithm, and combining dynamic numerical value normalization and abnormal value processing to provide standardized data; according to the invention, deep fusion analysis is carried out on multi-dimensional clinical data through the holographic fusion analysis and prediction module by means of a deep learning algorithm, a data association relationship model is established, an eye health assessment system is formed, a causal relationship model can be constructed based on the multi-dimensional data, and disease development trend prediction is realized. More comprehensive eye health information of the patient is provided for doctors, personalized treatment schemes and prevention measures can be formulated, disease development can be intervened in advance, and the treatment effect and life quality of the patient are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of ophthalmology, in particular to a holographic eye multi-dimensional disease screening system and method. BACKGROUND

[0002] In the process of continuous development of medical technology, the diagnosis and screening of ophthalmic diseases have always been an important research direction in the medical field. With the increasing emphasis on eye health and the rising incidence of ophthalmic diseases, the demand for accurate, efficient and comprehensive ophthalmic disease diagnosis technology is increasingly urgent. In recent years, medical data has shown a diversified development trend, and the types of data available in the field of ophthalmology have become increasingly rich, covering multi-dimensional clinical data such as intraocular pressure, retinal nerve fiber layer thickness, blood glucose level, and visual acuity, as well as image data such as fundus images and optical coherence tomography images. These multi-dimensional data provide a rich information base for more accurate diagnosis of ophthalmic diseases. At the same time, deep learning technology has shown great application potential in image recognition and data processing, providing technical support for integrating these multi-dimensional data for comprehensive diagnosis.

[0003] In the field of ophthalmic disease diagnosis, traditional diagnostic methods have many limitations. On the one hand, for example, only fundus images or intraocular pressure measurements are used to determine the disease. This single data approach cannot fully and accurately reflect the true health status of the eye, which can lead to missed or misdiagnosed diseases. In existing clinical assessments, the integration of multi-dimensional data is low, and the correlation model between different data types has not been established, which cannot fully utilize the synergistic effect of multi-source data. On the other hand, in the analysis of fundus images and OCT images, current methods mainly rely on manual analysis, which is greatly influenced by individual differences and physician experience. Different doctors may have different judgments on the same image, resulting in poor consistency of the diagnosis results. Moreover, manual analysis is inefficient and cannot meet the growing demand for ophthalmic disease diagnosis, and cannot provide timely diagnosis results and treatment recommendations for patients. In addition, existing OCT examinations have limitations in scanning area and scanning method, and quantitative analysis is mainly focused on basic indicators such as retinal nerve fiber layer thickness, and lacks quantitative evaluation of subtle structures such as the ganglion cell layer and the inner plexiform layer. Furthermore, it is impossible to mine the spatial relationships and gradient change characteristics between the layers of the retina through algorithms. In addition, although deep learning technology has certain applications in image recognition and data processing, there is currently a lack of mature disease screening systems and methods for deep fusion analysis of multi-dimensional eye data, which cannot fully utilize the advantages of multi-dimensional data and achieve more accurate disease screening. SUMMARY

[0004] The holographic eye multi-dimensional disease screening system and method can realize comprehensive, accurate and efficient screening of various eye diseases, and provide strong support for early diagnosis and treatment of eye diseases.

[0005] The holographic eye multi-dimensional disease screening system and method can realize comprehensive, accurate and efficient screening of various eye diseases, and provide strong support for early diagnosis and treatment of eye diseases. The data acquisition and preprocessing module: the professional equipment is used for collecting multi-dimensional clinical data including fundus images and OCT images, the image data is optimized through an image processing algorithm, dynamic numerical normalization and abnormal value processing are combined, and standardized data is provided; The deep learning model construction and analysis module: a multi-dimensional disease screening model is built, the features of the fundus images and the OCT images are extracted through dynamic adjustment of the convolution kernel quantity and a hybrid optimization algorithm, multi-modal data is fused by dynamically distributing weights through reinforcement learning, and intelligent screening of eye diseases is realized; The holographic fusion analysis and prediction module: deep learning algorithms are used, multi-dimensional data correlation is mined based on a graph neural network and a knowledge graph, a long short-term memory network and a medical knowledge updating mechanism are combined, an eye health evaluation system is constructed, and disease development trends are predicted; The result output and report generation module: the eye disease risk is evaluated by comprehensively considering disease inheritance, environmental factors and data uncertainty, the screening results are visually presented, and detailed reports including data fusion processes and disease trend predictions are generated to assist clinical decision-making.

[0006] Furthermore, the data acquisition and preprocessing module uses professional equipment to collect multi-dimensional clinical data, including fundus images, OCT images, intraocular pressure, retinal nerve fiber layer (RNFL), blood sugar level, visual acuity, and medical history information. Among them, the medical history information collection covers disease history, family genetic history, symptom duration, attack frequency, and previous treatment plans. The OCT detection site includes a 3.4mm annular area around the optic disc and the macular area. A combination of horizontal scanning, vertical scanning, and 45° / 135° oblique cross scanning is used. Before scanning, the pupil is dilated based on the transparency of the refractive media and the pupil diameter. Those who meet the conditions are treated with 1% compound tropicamide eye drops for pupil dilation. The image data is optimized by an image processing algorithm, and the algorithm formula is as follows: ,in, Represents enhanced fundus images or OCT images, optimized image data for subsequent analysis. Represents the original image collected, is the image enhancement power adjustment parameter, which is determined by comparing clinical fundus images and OCT images. is the brightness adjustment coefficient, which is determined based on the brightness distribution statistics of normal eye images by minimizing the brightness error between the enhanced image and the standard template. It is the difference between the average brightness of the original image and the preset standard brightness.

[0007] Furthermore, the deep learning model construction and analysis module builds a multidimensional disease screening model based on a deep learning framework, which includes an input layer, an intermediate layer and an output layer. The input layer receives preprocessed multi-source data, and the intermediate layer constructs a deep convolutional neural network for fundus images, extracts image features through multi-layer convolution and pooling operations, and uses a convolutional neural network combined with a three-dimensional convolutional neural network for OCT images to analyze image structure and spatial relationship features. The image features and other numerical data are then input into subsequent neural network layers, and comprehensive analysis is performed using feature fusion and decision fusion technology. The output layer outputs the screening results of ophthalmic diseases to achieve intelligent diagnosis of diseases.

[0008] Furthermore, the middle layer of the deep learning model construction and analysis module constructs a deep convolutional neural network for fundus images, and extracts image features through multi-layer convolution and pooling operations. The extraction formula is: ,in, For the The feature map output by the convolutional layer is used to identify multidimensional eye disease lesions. It is the Swish activation function, which improves the response to minor lesions by adaptively adjusting parameters. is the number of convolution kernels in this layer, which is dynamically optimized during training according to the complexity of the fundus image and the diversity of lesions. It is Layer convolution kernels, * represents the convolution operation, is the feature map of the previous layer, It is Layer bias term.

[0009] Furthermore, in the deep learning model construction and analysis module, for OCT images, a convolutional neural network combined with a three-dimensional convolutional neural network structure is used to calculate the OCT image quality. Extract features, where It is The feature volume output by the three-dimensional convolution layer is used to capture the spatial information of the retinal structure at each layer. The number of three-dimensional convolution kernels is dynamically set according to the difference in inter-layer structure of OCT images and the depth of the lesion. The inter-layer structure recognition accuracy under the given value (ELM / EZ recognition accuracy>95% in the macula, and cup-disc boundary recognition accuracy>93% in the ONH) is determined. It is Layer Three-dimensional convolution kernels whose parameters are optimized by combining adversarial training and meta-learning. represents a three-dimensional convolution operation, is the feature body of the previous layer, It is Layer bias term.

[0010] Furthermore, in the deep learning model construction and analysis module, the feature extraction results of fundus images and OCT images are combined with numerical data including intraocular pressure, RNFL, blood sugar level and visual acuity through the formula To integrate, is the comprehensive feature vector after fusion, is the fundus image feature vector, is the OCT image feature vector, is a numerical data feature vector, and The fusion weight coefficient is determined by a dynamic weight adjustment mechanism based on reinforcement learning, and the weight distribution is optimized in real time according to the degree of dependence of different diseases on each modality data.

[0011] Furthermore, the holographic fusion analysis and prediction module performs deep fusion processing on the input multi-dimensional clinical data, and uses the formula Mining the relationship between data, among which, Indicates the degree of data association, It is The basic correlation between different data types and diseases is calculated based on the statistical analysis of historical cases, and the Pearson correlation coefficient of different data types is taken as the normalized value. is the interaction coefficient of the data type, which is determined by fitting the interaction coefficient of the logistic regression model with reference to the weight of multi-index collaborative diagnosis in the ophthalmology clinical guidelines. It is The dynamic weights of various data types are determined by learning from the data through the graph attention network, thus forming a comprehensive eye health assessment system. At the same time, a prediction model is built based on multi-dimensional data to explore the potential connections behind the data, so as to predict the development trend of the disease.

[0012] Furthermore, the holographic fusion analysis and prediction module predicts the development trend of the disease, and its prediction formula is: ,in, is the predicted disease development trend, is a multi-dimensional clinical data time series, It is a long short-term memory network model combined with the attention mechanism. is the model training parameter set, It is a trend correction coefficient that is dynamically updated based on the latest medical research results. It is the difference between the current data characteristics and the historical data characteristics of similar cases, so as to achieve accurate prediction of the development trend of the disease.

[0013] Furthermore, the result output and report generation module comprehensively considers disease genetics, environmental factors and data uncertainty to assess the risk of eye diseases. The assessment formula is: ,in, is the disease risk value, is the output of the deep learning model. Disease screening scores, It is the comprehensive impact weight determined by the genetic probability of the disease and environmental risk factors. It is a risk fluctuation adjustment coefficient that is dynamically adjusted based on the data uncertainty assessment results. It is determined by statistically testing the uncertainty of the data and controlling the confidence interval of the risk assessment. The mean score for all disease screening tests is shown in Table 2.

[0014] In another aspect, a holographic multidimensional eye disease screening method is provided, comprising the following specific steps: Data acquisition: Start the data acquisition and preprocessing module to sequentially collect the patient's fundus images, OCT images, intraocular pressure data, RNFL data, blood sugar level data, and visual acuity data, and complete data classification and storage; Data preprocessing: the original data is processed, including image data denoising, enhancement, format arrangement, numerical data normalization and outlier processing, to generate preprocessed data; Model analysis: the preprocessed data is input into the multi-dimensional disease screening model of the deep learning model construction and analysis module, the model extracts features from the fundus image and OCT image respectively, combines other numerical data, and outputs the preliminary screening result through feature fusion and decision fusion analysis; Holographic fusion analysis: the preliminary screening result and related data are transmitted to the holographic fusion analysis and prediction module, and data fusion analysis is performed to establish a correlation model and a causal relationship model, and a disease diagnosis result and a disease development trend prediction are obtained; Result output: through the result output and report generation module, the screening result, the diagnosis conclusion, the disease risk level, the data contribution analysis and the detailed screening report are displayed on the visual interface.

[0015] Compared with the prior art, the holographic eye multi-dimensional disease screening system and method has the following beneficial effects: I. The holographic fusion analysis and prediction module of the present application performs deep fusion analysis on multi-dimensional clinical data by means of deep learning algorithm, not only establishes a correlation model between data and forms an eye health evaluation system, but also constructs a causal relationship model based on multi-dimensional data, realizes disease development trend prediction, provides more comprehensive patient eye health information for doctors, helps to develop personalized treatment plan and preventive measures, intervenes in disease development in advance, and improves patient treatment effect and quality of life.

[0016] II. The present application integrates intraocular pressure, retinal nerve fiber layer thickness (RNFL), blood glucose level, vision and other multi-dimensional clinical data with fundus image and optical coherence tomography (OCT) image, and uses artificial neural network and deep learning technology for comprehensive analysis, which can more comprehensively and accurately reflect the eye health status, effectively avoid misdiagnosis and misdiagnosis, at the same time, the deep learning model can automatically extract and analyze the features of data, reduce the influence of individual difference and experience on artificial analysis, greatly improve the diagnosis efficiency, and quickly provide the diagnosis result for the patient.

[0017] Other advantages, objects and features of the present application will be described in the subsequent specification to some extent, and to some extent, it will be obvious to those skilled in the art based on the study of the following text or can be taught from the practice of the present application. BRIEF DESCRIPTION OF DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained from these drawings without any creative effort.

[0019] Figure 1 A structural schematic diagram of a holographic eye multi-dimensional disease screening system; Figure 2 A flowchart of a holographic eye multi-dimensional disease screening method. DETAILED DESCRIPTION

[0020] In order to further illustrate the technical means and effects adopted by the present application to achieve the predetermined application purposes, the following will combine the drawings and preferred embodiments to specifically describe the specific embodiments, structures, features and effects according to the present application.

[0021] Embodiment one A 55-year-old female patient was screened, and her mother had glaucoma, belonging to a high-risk group of glaucoma, and she often felt eye swelling and visual fatigue recently.

[0022] In the data acquisition and preprocessing link, after the operator assists the patient to complete the basic information registration, the following contents are collected: Medical history information: collected through the built-in structured questionnaire, including disease history (denying metabolic diseases such as diabetes and hypertension, no previous surgery in ophthalmology), family genetic history (mother diagnosed with glaucoma, father with normal vision), symptom characteristics (eye swelling degree score 4 / 10, visual fatigue lasting about 2 hours / day), frequency of onset (3-4 times per week) and previous treatment (once self-use eye drops to relieve, specific drugs unknown); OCT detection: strictly according to the set position, scanning the 3.4mm annular area around the optic disc (with the optic disc center as the center) and the macular area (with the macular fovea as the center with a diameter of 6mm); the combination of horizontal scanning (128 scanning lines), vertical scanning (128 scanning lines) and 45° / 135° diagonal cross scanning (each 64 scanning lines) is adopted; before scanning, it is found that the pupil diameter of the patient is 2.5mm (≤3mm), and the lens is slightly turbid (refractive interval transparency score 2 / 5), which meets the mydriasis condition, so 1% compound tropicamide eye drops are used for mydriasis treatment, 2 times are dropped at an interval of 5 minutes, and the pupil diameter is confirmed to be ≥6mm after 30 minutes, then scanning is performed; Other data: synchronous acquisition of fundus images, intraocular pressure (22 mmHg in the right eye and 21 mmHg in the left eye), RNFL data, fasting blood glucose (5.3 mmol / L), and visual acuity (0.7 in the left eye and 0.8 in the right eye).

[0023] For fundus images and OCT images, the formula is used for processing, where the average brightness of the original OCT image is 65 (the preset standard brightness is 90), so that ; the brightness adjustment coefficient is calculated based on 1200 clinical image data, and the calculation formula is , which gives ; the image enhancement power adjustment parameter is dynamically set according to the original image entropy value (original entropy value 4.1, standard entropy value 5.2), and when the entropy value is lower than the standard, the value is 1.05 (for every 0.5 decrease in entropy value, 0.05 increase), so that ; and the final image enhancement formula is , which improves the clarity of the optic disc edge in the processed image ; for numerical data such as intraocular pressure, RNFL, blood glucose level, and visual acuity, normalization and dynamic adjustment are performed, and abnormal value detection and correction are performed on RNFL data.

[0024] As shown in Figure 1 , the preprocessed data is input into the deep learning model construction and analysis module, and for fundus images, the formula is used to extract features, and because there is a suspected lesion in the optic disc area of the patient's fundus, the number of convolution kernels K is set to 40 according to the complexity of the lesion (trained on 300 glaucoma cases) (K = 24 when there is no obvious lesion, and K = 56 when there is a complex lesion), the parameter of the activation function σ is optimized to 0.9 through model training, the bias term , and the weight is trained using a hybrid optimization algorithm, automatically extracting features such as blurred optic disc boundaries and increased cup-disc ratio in the patient's both eyes, with the left eye cup-disc ratio reaching 0.6 and the right eye cup-disc ratio reaching 0.55, which are beyond the normal range; for OCT images, the formula is used, and because the interlayer contrast around the optic disc is moderate (contrast value 0.52), the number of three-dimensional convolution kernels N is set to 24 according to the 500 OCT data calibration rule (N = 16 when the contrast is > 0.6, and N = 32 when the contrast is < 0.4); the bias term is determined through testing of 100 verification sets to reduce edge noise interference, and the weight , analysis of both eyes in the temporal and nasal retinal nerve fiber layer appears thin, thickness values decreased to 70 μm and 75 μm, significantly lower than the normal lower limit of the same age 80 μm, and then through the formula , the image features and numerical data features such as intraocular pressure, RNFL are fused and Through the dynamic weight adjustment mechanism based on reinforcement learning, the OCT image features (closely related to the optic nerve structure) weight , the fundus image feature weight , the numerical data (intraocular pressure, RNFL) weight , so the fusion formula is , the system preliminarily judges that the patient has a high probability of glaucoma.

[0025] Data into holographic fusion analysis and prediction module, through the formula , the internal relationship between multi-dimensional data is mined, and 3 types of key data are set: OCT structure parameters , family history , intraocular pressure , among which The basic correlation between each data type and glaucoma is obtained by analyzing the historical data relationship graph with a graph neural network, (OCT and glaucoma correlation trained from 1000 cases), (glaucoma family history correlation), (intraocular pressure and glaucoma correlation), reflect the synergistic effect between different data, (OCT and intraocular pressure synergistic effect), (family history and OCT synergistic effect), (intraocular pressure and family history interaction is weak), Through the combination of reinforcement learning and knowledge graph reasoning, the dynamic weight is optimized , , , the calculation is , the system finally evaluates the probability of the patient having open-angle glaucoma to reach 85%, and then uses the formula , based on the LSTM attention mechanism to analyze multi-dimensional clinical data time series , combined with the trend correction coefficient dynamically updated according to the current medical research new results and expert experience knowledge base , and the difference between the current data features and the historical similar case data features , it is predicted that in the next 1-2 years, if no intervention treatment is performed, the patient's vision may continue to decline, and the visual field defect range will further expand.

[0026] Finally, the result output and report generation module uses the formula Calculate the prevalence value, where the screening score :OCT structural abnormalities , family history Abnormal intraocular pressure , comprehensive impact weight Based on genetic and environmental factors: 、 、 , mean score , data uncertainty adjustment coefficient Because the data integrity is high, it is set to 0.07; Substituting into the calculation, we get , suggesting a high suspicion of open-angle glaucoma.

[0027] Based on the calculation results, a pop-up window will appear on the risk page of the system interface to remind you that "open-angle glaucoma is highly suspected, and further examination and treatment are recommended immediately." The generated screening report, which contains comprehensive analysis content, lists detailed comparative analysis charts of various examination data, intuitively presenting the differences between the patient's eye data and normal reference values. It also includes daily precautions for glaucoma, such as avoiding prolonged use of the eyes in a dark environment, maintaining emotional stability, and regularly checking intraocular pressure and visual field. These suggestions provide detailed and practical information for doctors to formulate treatment plans and for patients to understand their condition.

[0028] Example 2 A 45-year-old male patient with a 10-year history of diabetes was selected. His daily blood sugar control was unstable and he had recently experienced blurred vision. During the data collection and preprocessing phase, the operator guided the patient to sit in front of a high-resolution fundus camera equipped with autofocus, adjusted the seat height and head frame position to ensure that the patient's eyes were aligned with the camera window, and in a darkroom environment, the camera was controlled by professional shooting software to capture fundus images of the patient's eyes from multiple angles. A total of 8 high-definition images with different fields of view were obtained, completely covering the retinal area. The OCT image and retinal nerve fiber layer thickness (RNFL) data were obtained using an optical coherence tomography instrument. The patient fixed the chin on the instrument bracket, the forehead against the headrest, kept the eyeball stable and gazed at the fixation point in the instrument. The 3.4 mm annular area around the optic disc and the 6 mm area of the macular region were scanned by using the horizontal, vertical and 45° / 135° oblique cross scanning combination method. The examination found that the pupil diameter of the patient was 3.2 mm, the refractive lenticular transparency score was 3 (mild lens opacity), and the image quality requirements could be met without mydriasis. Three sets of three-dimensional OCT image data were generated, and the thickness values of RNFL at multiple key parts such as the macular region and the optic disc were accurately measured. The intraocular pressure was measured by a non-contact tonometer. Before measurement, the instrument probe was disinfected with a medical alcohol cotton swab. The patient looked straight ahead, and the tonometer completed three intraocular pressure measurements within 10 seconds by the air puff method, and the average value was taken as the final result. The fasting blood glucose and postprandial two-hour blood glucose data were obtained from the patient's recent blood test report, and the distance vision and near vision of the patient were detected using an automatic optometry instrument. All raw data were transmitted and stored in the system database in real time.

[0029] Subsequently, the system automatically preprocessed the fundus image and the OCT image, removed the speckle and stripe noise in the image by using a denoising algorithm, made the retinal blood vessels and subtle lesion areas more clear by using an adaptive contrast enhancement technique, and normalized the numerical type data such as intraocular pressure, RNFL, blood glucose level and vision, and corrected the abnormal value of the RNFL data. For example, when the RNFL thickness value of a part deviates from the normal range of the same age by more than 20%, the system automatically marks and corrects it in combination with the data of adjacent parts.

[0030] As shown in Figure 2 , the preprocessed data were input into a deep learning model construction and analysis module. For the fundus image, the deep convolutional neural network automatically extracted the features of the retinal microvessels through multiple convolution and pooling operations. The lesion recognition unit in the model quickly detected that there were 5 microvessels and 3 small patchy hemorrhagic points on the left eye retina, and 2 microvessels were also found on the right eye retina. For the OCT image, the convolutional neural network combined with three-dimensional image processing technology analyzed the structure of each layer of the retina in detail, found that the retinal thickness of the macular region of both eyes was slightly increased, and the ganglion cell layer appeared local thinning. At the same time, the data such as intraocular pressure, RNFL, blood glucose level and vision were fused and analyzed with the image features. The system noticed that the patient's blood glucose was at a high level for a long time, and the thickness value of RNFL on the temporal side of the optic disc was significantly lower than the normal range.

[0031] The data enters the holographic fusion analysis and prediction module, and the internal relationship between the multidimensional data is mined through a deep learning algorithm. The patient's high blood sugar state, the blood vessel lesion in the fundus image, and the abnormal RNFL data are combined, and the system comprehensively judges that the patient is in the moderate non-proliferative stage of diabetic retinopathy. At the same time, based on historical case data and current disease characteristics, it is predicted that if not treated in time, the lesion has a high probability of progressing to the proliferative stage in the next 6-12 months, and may develop serious complications such as neovascularization and retinal detachment.

[0032] Finally, the result output and report generation module presents the screening results in a visual interface, marks the diagnosis conclusion of diabetic retinopathy (moderate non-proliferative stage) and high risk of disease on the screen, and displays the contribution of each dimension data to the diagnosis result in the form of a radar chart. The influence of blood glucose level and fundus image characteristics is clearly shown, and a detailed report is generated. The report not only contains the specific analysis results of each data, the fundus image and OCT image with lesion area annotation, but also includes personalized treatment suggestions such as strict blood glucose control, fundus review every 3 months, and laser photocoagulation treatment when necessary.

[0033] The above is only a preferred embodiment of the present application, and is not intended to limit the present application in any form. Although the present application has been disclosed as above with a preferred embodiment, it is not intended to limit the present application. Any person skilled in the art can make some changes or modifications to the above disclosed technical content without departing from the scope of the present application, and any equivalent embodiments with equivalent changes are equivalent. Any modification, change, and modification of the above embodiments, which does not depart from the technical solution of the present application, is still within the scope of the present application.

Claims

1. A holographic multi-dimensional eye disease screening system, characterized in that: The system includes: Data acquisition and preprocessing module: uses professional equipment to collect multi-dimensional clinical data, optimizes image data through image processing algorithms, and combines dynamic numerical normalization and outlier processing to provide standardized data; Deep Learning Model Construction and Analysis Module: This module builds a multidimensional disease screening model, extracts fundus image and OCT image features by dynamically adjusting the number of convolution kernels and using a hybrid optimization algorithm. It also uses reinforcement learning to dynamically assign weights to fuse multimodal data, enabling intelligent screening for ophthalmic diseases. Holographic fusion analysis and prediction module: Using deep learning algorithms, it mines multi-dimensional data associations based on graph neural networks and knowledge graphs, combines long-term and short-term memory networks with medical knowledge update mechanisms, builds an eye health assessment system, and predicts disease development trends; Result output and report generation module: Comprehensively considers disease genetics, environmental factors and data uncertainty, evaluates the risk of eye diseases, visualizes screening results, and generates detailed reports including data fusion process and disease trend prediction to assist clinical decision-making.

2. The holographic multi-dimensional eye disease screening system according to claim 1, characterized in that: The data acquisition and preprocessing module uses professional equipment to collect multi-dimensional clinical data, including fundus images, OCT images, intraocular pressure, retinal nerve fiber layer (RNFL), blood sugar level, visual acuity, and medical history information. The medical history information collection covers disease history, family genetic history, symptom duration, attack frequency, and previous treatment plans. The OCT detection site includes a 3.4 mm annular area around the optic disc and the macular area. A combination of horizontal scanning, vertical scanning, and 45° / 135° oblique cross scanning is used. Before scanning, mydriasis is determined based on the transparency of the refractive media and the pupil diameter. Those who meet the conditions are treated with 1% compound tropicamide eye drops for mydriasis. The image data is optimized using an image processing algorithm, and the algorithm formula is as follows: ,in, Represents enhanced fundus images or OCT images, optimized image data for subsequent analysis. Represents the original image collected, is the power adjustment parameter for image enhancement, is the brightness adjustment coefficient, which is calculated based on the difference between the average brightness of the image and the standard brightness. It is the difference between the average brightness of the original image and the preset standard brightness.

3. The holographic multi-dimensional eye disease screening system according to claim 1, characterized in that: The deep learning model construction and analysis module builds a multidimensional disease screening model based on a deep learning framework, which includes an input layer, an intermediate layer and an output layer. The input layer receives preprocessed multi-source data. The intermediate layer constructs a deep convolutional neural network for fundus images, extracts image features through multi-layer convolution and pooling operations, and uses a convolutional neural network combined with a three-dimensional convolutional neural network for OCT images to analyze image structure and spatial relationship features. The image features and other numerical data are then input into subsequent neural network layers, and comprehensive analysis is performed using feature fusion and decision fusion technology. The output layer outputs the screening results of ophthalmic diseases to achieve intelligent diagnosis of diseases.

4. The holographic multi-dimensional eye disease screening system according to claim 3, characterized in that: The middle layer of the deep learning model construction and analysis module constructs a deep convolutional neural network for fundus images and extracts image features through multi-layer convolution and pooling operations. The extraction formula is: ,in, For the The feature map output by the convolutional layer is used to identify multidimensional eye disease lesions. It is the Swish activation function, which improves the response to minor lesions by adaptively adjusting parameters. is the number of convolution kernels in this layer, which is dynamically optimized during training according to the complexity of the fundus image and the diversity of lesions. It is Layer convolution kernels, * represents the convolution operation, is the feature map of the previous layer, It is Layer bias term.

5. The holographic multi-dimensional eye disease screening system according to claim 3, characterized in that: In the deep learning model construction and analysis module, for OCT images, a convolutional neural network is used in combination with a three-dimensional convolutional neural network structure, and the formula Extract features, where It is The feature volume output by the three-dimensional convolution layer is used to capture the spatial information of the retinal structure at each layer. is the number of three-dimensional convolution kernels, which is dynamically set according to the structural differences between OCT image layers and the depth of the lesion. It is Layer Three-dimensional convolution kernels whose parameters are optimized by combining adversarial training and meta-learning. represents a three-dimensional convolution operation, is the feature body of the previous layer, It is Layer bias term.

6. The holographic multi-dimensional eye disease screening system according to claim 3, characterized in that: In the deep learning model construction and analysis module, the feature extraction results of fundus images and OCT images are combined with numerical data including intraocular pressure, RNFL, blood sugar level and visual acuity through the formula To integrate, is the comprehensive feature vector after fusion, is the fundus image feature vector, is the OCT image feature vector, is a numerical data feature vector, and The fusion weight coefficient is determined by a dynamic weight adjustment mechanism based on reinforcement learning, and the weight distribution is optimized in real time according to the degree of dependence of different diseases on each modality data.

7. The holographic multi-dimensional eye disease screening system according to claim 1, characterized in that: The holographic fusion analysis and prediction module performs deep fusion processing on the input multi-dimensional clinical data, and uses the formula Mining the relationship between data, among which, Indicates the degree of data association, It is The basic correlation between the data types and diseases, is the interaction coefficient of data type, It is Dynamic weights of various data types are calculated to form a comprehensive eye health assessment system. At the same time, a prediction model is built based on multi-dimensional data to explore the potential connections behind the data and predict the development trend of the disease.

8. The holographic multi-dimensional eye disease screening system according to claim 7, characterized in that: The holographic fusion analysis and prediction module predicts the development trend of the disease, and its prediction formula is: ,in, is the predicted disease development trend, is a multi-dimensional clinical data time series, It is a long short-term memory network model combined with the attention mechanism. is the model training parameter set, It is a trend correction coefficient that is dynamically updated based on the latest medical research results. It is the difference between the current data characteristics and the historical data characteristics of similar cases, so as to achieve accurate prediction of the development trend of the disease.

9. The holographic multi-dimensional eye disease screening system according to claim 1, characterized in that: The result output and report generation module comprehensively considers disease genetics, environmental factors and data uncertainty to assess the risk of eye diseases. The assessment formula is: ,in, is the disease risk value, is the output of the deep learning model. Disease screening scores, It is the comprehensive impact weight determined by the genetic probability of the disease and environmental risk factors. It is the risk volatility adjustment coefficient that is dynamically adjusted based on the data uncertainty assessment results. The mean score for all disease screening tests is shown.

10. A holographic multidimensional eye disease screening method, which is applicable to a holographic multidimensional eye disease screening system according to claims 1-9, characterized in that: The method comprises the following specific steps: Data acquisition: Start the data acquisition and preprocessing module to sequentially collect the patient's fundus images, OCT images, intraocular pressure data, RNFL data, blood sugar level data, and visual acuity data, and complete data classification and storage; Data preprocessing: Processing the raw data, including image data denoising, enhancement, formatting, numerical data normalization and outlier processing, to generate preprocessed data; Model analysis: The preprocessed data is input into the multidimensional disease screening model of the deep learning model construction and analysis module. The model extracts features from fundus images and OCT images respectively, combines them with other numerical data, and outputs preliminary screening results through feature fusion and decision fusion analysis. Holographic fusion analysis: The preliminary screening results and related data are transmitted to the holographic fusion analysis and prediction module, and data fusion analysis is performed to establish correlation models and causal relationship models to obtain disease diagnosis results and disease development trend predictions; Result output: Through the result output and report generation module, the screening results, diagnostic conclusions, disease risk level, data contribution analysis are displayed in a visual interface, and a detailed screening report is generated.

Images