Intelligent analysis method for diabetic corneal neuropathy based on confocal microscope

Abstract

This invention discloses an intelligent analysis method for diabetic corneal neuropathy based on confocal microscopy, comprising: acquiring raw images of corneal nerve tissue using a confocal microscope; performing format conversion and preliminary feature extraction via an intelligent diagnostic platform for diabetic neuropathy; calling a multi-scale neuromorphic analysis model to perform multi-dimensional analysis of nerve fiber structure and obtain a morphological parameter matrix; conducting lesion risk correlation analysis using a corneal neuropathy early warning model to generate a feature mapping map; and optimizing image quality and completing structurally defective areas using a confocal image denoising and repair algorithm. The core steps of multi-scale neuromorphic analysis, lesion early warning, and denoising and repair all include processes such as layered capture, feature fusion, risk calculation, and noise filtering. This invention achieves accurate extraction of corneal nerve morphological parameters and objective assessment of lesion risk, providing a systematic intelligent analysis solution for the early diagnosis of diabetic corneal neuropathy.

Description

Technical Field

[0001] This invention relates to the field of diabetic corneal neuropathy analysis technology, and in particular to an intelligent analysis method for diabetic corneal neuropathy based on confocal microscopy. Background Technology

[0002] Diabetic corneal neuropathy, a common chronic complication of diabetes, has an insidious onset and slow progression, often lacking typical clinical symptoms in its early stages. Delayed diagnosis can easily lead to severe eye damage such as corneal ulcers and perforations. Traditional diagnostic methods rely on invasive examinations or subjective morphological observation, which struggle to accurately capture subtle structural changes in nerve fibers. While confocal microscopy provides technical support for non-invasive visualization of corneal nerve morphology, the key challenge in clinical diagnosis lies in transforming the acquired image data into objective and accurate lesion analysis results. Therefore, there is an urgent need to develop a comprehensive technical solution integrating image acquisition, morphological analysis, risk warning, noise reduction and restoration, and intelligent diagnosis to achieve early identification and accurate assessment of diabetic corneal neuropathy, filling the technological gap in intelligent and systematic analysis in current clinical diagnosis.

[0003] Existing technologies suffer from two core drawbacks: First, the precision of neuromorphic analysis is insufficient. Traditional analysis methods rely heavily on single-scale feature extraction, failing to comprehensively capture the morphological parameters of corneal nerve fibers at different levels. They also lack multi-dimensional collaborative analysis capabilities for key information such as nerve branch nodes and direction vectors, limiting the completeness and accuracy of the morphological parameter matrix. Second, the integration of lesion analysis is inadequate. Existing technologies exhibit independent modules for image denoising and restoration, morphological analysis, and risk warning, lacking an effective collaborative scheduling mechanism and failing to form a closed-loop analysis process from image acquisition to result output. This results in low analysis efficiency and makes it difficult to improve the consistency and reliability of lesion diagnosis through collaborative optimization of various modules, thus failing to meet the actual clinical needs for early warning and accurate assessment of diabetic corneal neuropathy. Summary of the Invention

[0004] In order to overcome the shortcomings and deficiencies of existing technologies, this invention provides an intelligent analysis method for diabetic corneal neuropathy based on confocal microscopy.

[0005] The technical solution adopted in this invention is an intelligent analysis method for diabetic corneal neuropathy based on confocal microscopy, comprising the following steps: S1, acquiring nerve tissue image data of the corneal calibration area of ​​diabetic patients using a confocal microscope to obtain a raw image set including nerve fiber morphological features, distribution density, and texture information; S2, inputting the raw image set into an intelligent diagnostic platform for diabetic neuropathy, and completing image data format conversion and preliminary feature extraction through the platform's built-in data interface; S3, calling a multi-scale neural morphology analysis model to perform multi-dimensional analysis of the nerve fiber structure in the converted image data, obtaining nerve branch node coordinates, fiber orientation vectors, and morphological parameter matrices; S4, utilizing the angle... The corneal neuropathy early warning model performs lesion risk correlation analysis on the analyzed morphological parameter matrix to generate a feature map including potential lesion areas; S5, a confocal image denoising and repair algorithm is used to enhance the signal and complete the structure of areas with noise interference and structural defects in the feature map, resulting in an optimized neural tissue image; S6, the processing results from S3 to S5 are integrated through the intelligent diagnostic platform for diabetic neuropathy to output a morphological analysis report of corneal neuropathy and a set of lesion correlation parameters. In this process, the multi-scale neural morphology analysis model, the corneal neuropathy early warning model and the confocal image denoising and repair algorithm are used in synergy to complete the intelligent analysis of the entire process of data acquired by confocal microscopy.

[0006] Furthermore, the expression for the multi-scale neuromorphic analytical model is: ,in, The results are from multi-scale neuromorphic analysis. Image pixel coordinates, The angle of nerve fiber direction. This is a parameter representing the length of nerve fibers. The number of analytical scales, For the first Scale weighting coefficients For the first Gaussian kernel function of scale, For the first Gaussian kernel standard deviation of the scale This is the convolution operator. For neuromorphic feature functions, For the first Scale-based morphological adjustment parameters. For the first Feature enhancement coefficient at scale This is the error correction term.

[0007] Furthermore, the expression for the corneal neuropathy early warning model is: ,in, As a disease early warning index, , , These are the feature weight coefficients. For abnormal nerve density parameters, For neural branch variation parameters, For parameters of neural texture disorder, It is a natural constant. This is the adjustment factor for the early warning threshold. This is a risk correction factor. These are parameters representing individual differences.

[0008] Furthermore, the expression for the confocal image denoising and restoration algorithm is: ,in, To optimize the image in Pixel value at that location, For the original image in Pixel value at that location, The number of neighboring pixels. For the first The denoising coefficients of each neighborhood Original image First The pixel value of each neighboring area. This is the range for calculating the second derivative. For the first The repair weight for each offset For the Laplace operator.

[0009] Furthermore, the collaborative scheduling model of the intelligent diagnostic platform for diabetic neuropathy is as follows: ,in, The comprehensive analysis results output by the platform The integration coefficient is the result. , , These are the weights of the model output. Output of a multi-scale neuromorphic analysis model. Output for the early warning model of corneal neuropathy This is the output of the confocal image denoising and restoration algorithm. As a collaborative correction factor, This is a set of platform system parameters.

[0010] Furthermore, the image acquisition parameter optimization model for the confocal microscope is as follows: ,in, The optimized acquisition and control parameters, For parameter adjustment coefficients, This is the correlation function between corneal region depth and acquisition resolution. To collect depth parameters, Set parameters for resolution. For the adaptation function of laser intensity and scanning speed, For laser intensity parameters, For scanning speed parameters, For interference compensation coefficient, Let f be the effect function of ambient light and equipment noise. For ambient light intensity parameters, This represents the equipment noise figure.

[0011] Further, S3 includes the following sub-steps: S31, based on the difference in gray-level distribution of nerve fibers in the confocal image, a multi-scale pyramid structure is constructed, and the edge contours and internal textures of nerve fibers are captured layer by layer through feature extraction operators at different scales; S32, the feature information extracted at each scale is reorganized dimensionally through a feature fusion matrix to establish a mapping relationship between neural morphological features and analytical scales; S33, the multi-scale neural morphology analysis model is called to perform iterative calculations on the reorganized feature data, gradually refining the calculation accuracy of neural branch node coordinates, fiber orientation vectors, and morphological parameters; S34, the iterative calculation results are filtered to remove redundant information unrelated to neural morphology and retain the calibrated morphological parameter matrix.

[0012] Further, S4 includes the following sub-steps: S41, performing feature normalization processing on the morphological parameter matrix output by S3, mapping parameters of different dimensions to a unified analysis interval; S42, constructing a risk association rule base based on the pathological characteristics of corneal neuropathy, clarifying the correspondence between abnormal morphological parameters and lesion types; S43, inputting the normalized morphological parameters into the corneal neuropathy early warning model, calculating the lesion risk index of each region; S44, dividing high, medium, and low risk regions according to the risk index threshold, generating a feature mapping map including risk level annotations.

[0013] Further, step S5 includes the following sub-steps: S51, performing noise detection on the feature map and identifying the regions where salt-and-pepper noise and Gaussian noise are located through the statistical distribution of pixel grayscale values; S52, calling the confocal image denoising and repair algorithm to filter the signal in the noise region and replace abnormal pixel values ​​by calculating the correlation of neighboring pixels; S53, for the broken or missing areas of nerve fibers in the image, performing structural completion based on the morphological features of adjacent regions to construct a continuous nerve fiber structure; S54, performing feature verification on the repaired image to ensure the consistency between the neural morphological parameters and the original analysis results.

[0014] A confocal microscopy-based intelligent analysis method for diabetic corneal neuropathy is implemented through different units, including: a high-precision confocal image acquisition unit, used to capture raw image data of corneal nerve tissue from diabetic patients, achieve clear imaging of nerve fiber details through optical parameter adjustment, and transmit the acquired data to the next-level unit; an image format conversion and feature preprocessing unit, which receives the data output from the acquisition unit, completes format standardization conversion and initial feature extraction, and establishes an interface for adaptation with subsequent analysis models; a multi-scale neuromorphic intelligent analysis unit, which calls a multi-scale neuromorphic analysis model to analyze the neural structure of the preprocessed image data and outputs a morphological parameter matrix; a lesion risk warning and feature mapping unit, based on a corneal neuropathy warning model, performs risk analysis on the morphological parameter matrix and generates a feature mapping map annotating lesion risk areas; an image denoising and repair optimization unit, which uses a confocal image denoising and repair algorithm to filter noise and complete the structure of the feature mapping map, and outputs an optimized nerve tissue image; and a comprehensive analysis result output unit, which receives the output data from the optimization unit, integrates the data through a diabetic neuropathy intelligent diagnostic platform, generates a final result including a morphological analysis report and lesion-related parameters, and performs a closed-loop output of the entire intelligent analysis process.

[0015] Beneficial Effects: This invention proposes an intelligent analysis method for diabetic corneal neuropathy based on confocal microscopy. Leveraging the high-precision image acquisition capabilities of confocal microscopy and the collaborative scheduling function of an intelligent diagnostic platform for diabetic neuropathy, a multi-scale neuromorphic analysis model is used to capture and analyze the morphological features of corneal nerve fibers in a layered and multi-dimensional manner. This comprehensively obtains key parameters such as nerve branch nodes and direction vectors, effectively solving the problems of incomplete morphological parameter analysis and insufficient accuracy caused by traditional single-scale analysis. A corneal neuropathy early warning model establishes a correlation analysis mechanism between morphological parameters and lesion risk. Combined with a confocal image denoising and repair algorithm, image noise and structural defect areas are optimized. Simultaneously, the platform enables collaborative model invocation, forming a closed-loop analysis process from image acquisition to result output, completely overcoming the shortcomings of existing technologies where functional modules are independent and lack integration. This method does not rely on subjective observation. Through a systematic intelligent analysis process, it significantly improves the completeness and accuracy of corneal nerve morphology analysis, enhances the objectivity and reliability of lesion risk assessment, provides comprehensive technical support for the early identification and accurate assessment of diabetic corneal neuropathy, and greatly improves the efficiency and quality of clinical diagnosis. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the overall process of the method of the present invention. Figure 2 This is a flowchart of method step S3 of the present invention; Figure 3This is a flowchart of method step S4 of the present invention; Figure 4 This is a flowchart of step S5 of the method of the present invention; Figure 5 This is a diagram showing the system unit composition of the present invention. Detailed Implementation

[0017] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. The application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0018] like Figure 1 As shown, the intelligent analysis method for diabetic corneal neuropathy based on confocal microscopy includes the following steps: S1, acquires neural tissue image data of corneal calibrated areas of diabetic patients through confocal microscopy, and obtains a set of raw images including nerve fiber morphological characteristics, distribution density and texture information; Specifically, in step S1, neural tissue image data of a specific region of the cornea of ​​a diabetic patient is acquired using a confocal microscope. During the process, the central region of the cornea and four quadrants within 3 mm from the limbus are first identified as the target acquisition area. The laser wavelength of the confocal microscope is adjusted to 488 nanometers, the scanning resolution is set to 1024×1024 pixels, the scanning speed is controlled to 8 frames per second, and the single acquisition time is set to 15 seconds to ensure that data acquisition is completed while the patient's cornea remains stable. During the acquisition process, the corneal nerve tissue was magnified 400 times using the optical zoom system of the microscope, focusing on the subepithelial nerve plexus and stromal nerve fibers of the cornea. The morphological characteristics of the nerve fibers, including diameter, curvature, and number of branches, were recorded simultaneously. The distribution density was statistically measured using the length and number of nerve fibers per square millimeter. Texture information, including key indicators such as grayscale uniformity, contrast, and texture entropy, was also recorded. Finally, a raw image set of 20 to 30 single-frame images was obtained. Each image contained complete spatial distribution and microstructural information of nerve fibers, providing comprehensive and high-precision raw data support for subsequent analysis. This step ensured that the raw images could truly reflect the physiological or pathological state of the corneal nerves by precisely controlling the acquisition parameters and region selection.

[0019] S2, input the original image set into the intelligent diagnostic platform for diabetic neuropathy, and complete the format conversion and preliminary feature extraction of the image data through the platform's built-in data interface; Specifically, in step S2, the original image set is input into the intelligent diagnostic platform for diabetic neuropathy. During implementation, the platform first receives the original image data through its built-in HDMI and USB 3.0 dual data interfaces, with the interface transmission rate set to 5Gbps to ensure efficient and complete data transmission. The platform then activates its image format conversion module to convert the original RAW format images to the standard DICOM format. During the conversion process, the pixel depth is maintained at 16 bits, the color mode is grayscale, and the pixel coordinates of the images are calibrated to ensure that the spatial coordinates of each image correspond one-to-one with the actual location of the cornea. The initial feature extraction module is then initiated, employing a sliding window method with a window size of 32×32 pixels and a step size of 16 pixels to traverse and scan the converted image, extracting basic features such as the mean gray level of nerve fibers, edge gradient magnitude, and texture energy. A corresponding extraction threshold is set for each feature dimension: the mean gray level threshold ranges from 80 to 200, the edge gradient magnitude threshold is set to 30, and the texture energy threshold is set to 500. Background areas and invalid pixels are removed through threshold filtering, initially identifying regions of interest including nerve fibers, and generating a preliminary feature set with a feature vector dimension of 128. This step, through standardized format conversion and targeted initial feature extraction, provides highly adaptable preprocessed data with low noise interference for subsequent model analysis, ensuring the smooth progress of the subsequent analysis process.

[0020] S3 calls a multi-scale neuromorphic analysis model to perform multi-dimensional analysis of the nerve fiber structure in the converted image data, and obtains the coordinates of nerve branch nodes, fiber orientation vectors and morphological parameter matrices. Specifically, step S3 calls a multi-scale neuromorphic analysis model to perform multi-dimensional analysis of the neural fiber structure in the converted image data. During implementation, the number of analysis scales is initially set to five levels, corresponding to image scaling ratios of 1x, 2x, 4x, 8x, and 16x, respectively. Each scale is configured with independent analysis parameters. For the image data at each scale, the model first identifies the edge contours of the neural fibers, capturing the boundary pixels of the fibers through an edge detection algorithm. Then, based on connected component analysis, it determines the continuous morphology of the neural fibers, calculates the coordinates of the neural branch nodes in pixel coordinates (x, y), accurate to the position of a single pixel, and simultaneously solves for the fiber orientation vector using vector calculation methods. With the horizontal direction as the 0-degree reference, the orientation angle of each fiber segment is calculated with an accuracy controlled within 0.1 degrees. The construction of the morphological parameter matrix includes 15 key parameters such as nerve fiber diameter, length, curvature, branch spacing, and branch angle. The diameter is measured to an accuracy of 0.01 micrometers, the length is converted to the actual physical length from the pixel distance, the curvature is calculated as the ratio of the actual fiber length to the straight-line distance, the branch spacing is the straight-line distance between two adjacent branch nodes, and the branch angle is the angle between the branch and the main fiber. After standardization, all parameters are constructed into a 15×N dimensional morphological parameter matrix, where N is the total number of nerve fibers in the image. This step, through multi-scale collaborative analysis and precise parameter calculation, comprehensively obtains detailed information on nerve morphology, providing a quantitative basis for lesion diagnosis.

[0021] S4. Using the corneal neuropathy early warning model, the morphological parameter matrix after analysis is subjected to lesion risk correlation analysis to generate a feature mapping map including potential lesion areas. Specifically, step S4 utilizes the corneal neuropathy early warning model to perform lesion risk association analysis on the parsed morphological parameter matrix. First, the 15 key parameters in the morphological parameter matrix are categorized into three main types: density-related parameters, morphological structure parameters, and texture feature parameters. Each type of parameter is assigned a corresponding weight coefficient. The model incorporates a risk association algorithm trained on a clinical case database, which includes corneal nerve data from over 10,000 patients with diabetic corneal neuropathy and healthy individuals, covering case information with different disease durations, ages, and severity levels. During the analysis, the model compares the input morphological parameter matrix with the standard parameters in the database, calculating the degree of abnormality for each parameter. The degree of abnormality is represented by the deviation rate between the actual parameter and the mean of the standard parameters, with a deviation rate threshold set at ±30%. Parameters exceeding this range are considered abnormal. Subsequently, based on the number, type, and degree of abnormal parameters, a lesion risk index is calculated using a weighted summation method. The risk index range is set from 0 to 10, where 0 to 3 represents low risk, 3 to 7 represents medium risk, and 7 to 10 represents high risk. Meanwhile, the model uses a heatmap generation algorithm to label corneal regions corresponding to different risk levels. Red indicates high-risk areas, yellow indicates medium-risk areas, and blue indicates low-risk areas. Finally, it generates a feature map that includes the location, extent, and risk level of potential lesions. This step, by combining massive clinical data with precise parameter comparison, achieves an objective assessment and visualization of lesion risk, providing key support for early lesion identification.

[0022] S5. A confocal image denoising and repair algorithm is used to enhance the signal and complete the structure of the regions with noise interference and structural defects in the feature map, so as to obtain an optimized neural tissue image. Specifically, step S5 employs a confocal image denoising and repair algorithm to enhance signals and complete structures in areas of noise interference and structural defects in the feature map. During implementation, the feature map is first subjected to noise type identification. By statistically analyzing the distribution characteristics of pixel grayscale values, salt-and-pepper noise and Gaussian noise are distinguished. Salt-and-pepper noise is determined by extreme points where pixel grayscale values ​​exceed the normal range (80 to 200). Gaussian noise is determined by the standard deviation of grayscale values; a standard deviation greater than 20 indicates the presence of Gaussian noise. For salt-and-pepper noise regions, a median filtering algorithm is used for denoising, with a filter window size of 5×5 pixels. Abnormal pixel values ​​are replaced by the median value within the window. For Gaussian noise regions, an adaptive Wiener filtering algorithm is used, with the filter kernel size dynamically adjusted according to the noise intensity, ranging from 3×3 to 9×9 pixels, ensuring that detailed information of nerve fibers is preserved while removing noise. For areas with structural defects, morphological dilation and erosion operations are used to first locate the defects. Defective areas are defined as regions with more than five missing pixels in a continuous nerve fiber structure. Then, based on the morphological parameters of nerve fibers in adjacent regions, an interpolation algorithm is used to supplement the pixel information of the defective areas. The interpolation step size is set to 1 pixel to ensure that the completed nerve fiber structure is continuous and conforms to the overall morphological characteristics. Simultaneously, a signal enhancement module is activated, using a grayscale stretching algorithm to increase the grayscale contrast between the nerve fibers and the background by 40%, enhancing the visualization of the nerve fibers. The final result is an optimized neural tissue image with noise interference of less than 5% and structural integrity greater than 95%. This step, through targeted denoising and precise structural completion, significantly improves image quality and provides highly reliable image data for subsequent result integration.

[0023] S6 integrates the processing results of S3 to S5 through the intelligent diagnostic platform for diabetic neuropathy, and outputs a morphological analysis report of corneal neuropathy and a set of lesion-related parameters. In this process, the intelligent analysis of the entire process of confocal microscope acquisition data is completed through the coordinated use of multi-scale neuromorphic analysis model, corneal neuropathy early warning model and confocal image denoising and repair algorithm.

[0024] Specifically, step S6 integrates the processing results from S3 to S5 through the intelligent diagnostic platform for diabetic neuropathy. During implementation, the platform first constructs a data integration framework, setting up three data input ports to receive the morphological parameter matrix output from S3, the feature map output from S4, and the optimized neural tissue image output from S5, respectively. Data transmission employs an encryption protocol to ensure information security. Internally, the platform activates a data fusion module, using a weighted fusion algorithm to integrate the three types of data. The weights for the morphological parameter matrix are set to 0.4, the feature map to 0.3, and the optimized image to 0.3. During the fusion process, the timestamps and spatial coordinates of each data point are calibrated to ensure spatiotemporal consistency. Subsequently, the analysis report generation module is activated. Based on a preset report template, this module presents key indicators from the morphological parameter matrix in tabular form, including core parameters such as the average nerve fiber diameter, distribution density, and number of branches. Simultaneously, it overlays the risk area annotations from the feature map with the optimized image, generating a visualized image report that marks the specific locations and ranges of high- and medium-risk areas. The generation of the lesion-related parameter set includes 10 key related parameters such as risk index, number of abnormal parameters, lesion area proportion, and degree of nerve fiber damage. Each parameter is accompanied by corresponding statistical analysis results. The final output morphological analysis report includes four modules: data acquisition information, analysis process parameters, result statistical charts, and visualization images. The lesion-related parameter set is stored in a standardized data file format. During the process, a multi-model collaborative calling scheduling mechanism ensures the synchronization of data transmission and processing across modules, realizing intelligent analysis of confocal microscopy data from raw images to final analysis results, providing comprehensive, accurate, and intuitive reference for clinical diagnosis.

[0025] Preferably, the expression for the multi-scale neuromorphic analytical model is: ,in, The results are from multi-scale neuromorphic analysis. Image pixel coordinates, The angle of nerve fiber direction. This is a parameter representing the length of nerve fibers. The number of analytical scales, For the first Scale weighting coefficients For the first Gaussian kernel function of scale, For the first Gaussian kernel standard deviation of the scale This is the convolution operator. For neuromorphic feature functions, For the first Scale-based morphological adjustment parameters. For the first Feature enhancement coefficient at scale This is the error correction term.

[0026] Specifically, the multi-scale neuromorphic analysis model is based on the morphological differences of corneal nerve fibers at different spatial scales. It is constructed by integrating the synergistic effect of multi-scale Gaussian kernel functions and neuromorphic feature functions. The derivation process first clarifies that neuromorphic analysis needs to cover multi-dimensional information from the micro-pixel level to the macro-structural level. Therefore, a multi-scale hierarchical analysis approach is introduced. Images are filtered at multiple scales using Gaussian kernel functions with different standard deviations to preserve neuromorphic details at each scale. Then, convolution operations are used to fuse multi-scale features with neuromorphic feature functions. Simultaneously, weighting coefficients and feature enhancement coefficients are introduced to adjust the contribution of features at each scale. Finally, an error correction term is added to compensate for systematic biases in the analysis process, thus establishing a complete formula. The core basis of this formula is the multi-scale distribution characteristics of corneal nerve fibers. Different scales correspond to different structures such as the trunk, branches, and terminals of nerve fibers, requiring differentiated parameters to achieve accurate analysis. Regarding parameter values, the number of analytical scales is set to 5 to 8, the weighting coefficients are allocated from 0.1 to 0.3 according to the importance of features at each scale, the standard deviation of the Gaussian kernel is set from 1 to 8 increasing with scale, the morphological adjustment parameter ranges from 0.5 to 1.2, the feature enhancement coefficient is set from 1.0 to 2.0, and the error correction term is controlled within ±0.05. In implementation, the input image is first decomposed into multiple scales, then the neuromorphic analysis results are calculated scale by scale using formulas, and finally the outputs from each scale are fused to obtain complete morphological parameters. This process, through multi-scale collaborative analysis, ensures the extraction accuracy of key parameters such as neural branch nodes and direction vectors, providing comprehensive morphological data support for subsequent lesion analysis.

[0027] Preferably, the expression for the corneal neuropathy early warning model is: ,in, As a disease early warning index, , , These are the feature weight coefficients. For abnormal nerve density parameters, For neural branch variation parameters, For parameters of neural texture disorder, It is a natural constant. This is the adjustment factor for the early warning threshold. This is a risk correction factor. These are parameters representing individual differences.

[0028] Specifically, the corneal neuropathy early warning model is based on a logistic regression model, optimized by incorporating the nonlinear correlation between corneal nerve morphological parameters and lesion risk. The derivation process first identifies three core risk factors—abnormal nerve density, branching variation, and texture disorder—through statistical analysis of clinical data. These are used as input variables in the model, and a sigmoid function maps the linear combination results to a risk range of 0 to 1. Then, an early warning threshold adjustment factor is introduced to broaden the effective risk differentiation range. Finally, an individual difference correction term is added to compensate for differences in the physiological characteristics of different patients, forming a complete formula. The core basis of this formula is the strong correlation between abnormal nerve morphological parameters and lesion occurrence in clinical cases, achieving accurate risk assessment by quantifying the degree of parameter abnormality. Regarding parameter values, the weight coefficients of the three main features are set to 0.3 to 0.5 based on clinical correlation analysis, the early warning threshold adjustment factor is set to 0.8 to 1.2, the risk correction coefficient is set to 0.1 to 0.3, and the individual difference parameter is calibrated using basic information such as patient age and disease duration, with a value range of 0.05 to 0.2. During implementation, the morphological parameters obtained from the analysis are first standardized, the degree of abnormality of each parameter is calculated, and then the parameters are substituted into the formula to calculate the lesion warning index. The risk level is divided according to the index size. This process integrates multi-dimensional risk factors and individual difference correction to achieve an objective quantitative assessment of lesion risk and provide accurate numerical basis for early warning.

[0029] Preferably, the expression for the confocal image denoising and restoration algorithm is: ,in, To optimize the image in Pixel value at that location, For the original image in Pixel value at that location, The number of neighboring pixels. For the first The denoising coefficients of each neighborhood Original image First The pixel value of each neighboring area. This is the range for calculating the second derivative. For the first The repair weight for each offset For the Laplace operator.

[0030] Specifically, the confocal image denoising and restoration algorithm is based on the principles of pixel correlation and structural continuity. It combines the noise characteristics of confocal images with the structural features of neural fibers. The derivation process first analyzes the noise types in the confocal image, determining suppression strategies for salt-and-pepper noise and Gaussian noise. Noise filtering is achieved through neighborhood pixel difference calculation. Then, a second derivative term is introduced to capture structural changes in the image, supplementing pixel information in the missing areas. Denoising and restoration functions are integrated through product and summation operations to establish a complete formula. The core basis of this formula is the difference in grayscale distribution between noise and signal in the confocal image, and the continuity of neural fiber structure. Noise is suppressed through neighborhood correlation, and defects are supplemented through structural derivatives. Regarding parameter values, the number of neighborhood pixels is set to 8 to 16, the denoising coefficient is dynamically adjusted from 0.05 to 0.2 according to the noise intensity, the second derivative calculation range is set to 1 to 3 pixel offsets, and the restoration weight is allocated from 0.1 to 0.4. During implementation, noise is first detected and located in the original image. Then, the optimized pixel value is calculated pixel by pixel using a formula. Neighborhood denoising is enhanced for noisy areas, and structural repair contribution is increased for defective areas. This process effectively reduces image noise interference, completes the missing neural fiber structure, and improves image quality and the reliability of subsequent analysis through targeted denoising and repair synergy.

[0031] Preferably, the collaborative scheduling model of the intelligent diagnostic platform for diabetic neuropathy is as follows: ,in, The comprehensive analysis results output by the platform The integration coefficient is the result. , , These are the weights of the model output. Output of a multi-scale neuromorphic analysis model. Output for the early warning model of corneal neuropathy This is the output of the confocal image denoising and restoration algorithm. As a collaborative correction factor, This is a set of platform system parameters.

[0032] Specifically, the collaborative scheduling model of the intelligent diagnostic platform for diabetic neuropathy is based on weighted fusion theory and constructed by combining the complementarity of model output results. The derivation process first clarifies the different roles of morphological analysis results, lesion warning index, and optimized images, using them as model input variables. Weight coefficients are used to adjust the contribution ratio of each input, and a result integration coefficient is introduced to optimize the fusion effect. Then, system parameter correction terms are added to compensate for errors during platform operation, establishing a complete formula. The core basis for this formula is the functional complementarity of the three core models: morphological analysis provides basic data, the warning model provides risk assessment, and the optimized image provides visualization support. Comprehensive analysis needs to be achieved through collaborative fusion. Regarding parameter values, the output weights of the three models are set to 0.3 to 0.5 according to functional importance, the result integration coefficient ranges from 0.9 to 1.1, the collaborative correction factor is set to 0.05 to 0.15, and the system parameter set is calibrated through platform hardware performance and software configuration, with a value range of 0.8 to 1.2. During implementation, the platform first receives the output data from the three major models, performs timestamp and spatial coordinate calibration, and then substitutes it into the formula for weighted fusion calculation to generate comprehensive analysis results. This process achieves efficient collaboration of data from each module through precise weight allocation and system error correction, ensuring the comprehensiveness and accuracy of the output results and providing integrated analytical support for clinical diagnosis.

[0033] Preferably, the image acquisition parameter optimization model of the confocal microscope is as follows: ,in, The optimized acquisition and control parameters, For parameter adjustment coefficients, This is the correlation function between corneal region depth and acquisition resolution. To collect depth parameters, Set parameters for resolution. For the adaptation function of laser intensity and scanning speed, For laser intensity parameters, For scanning speed parameters, For interference compensation coefficient, Let f be the effect function of ambient light and equipment noise. For ambient light intensity parameters, This represents the equipment noise figure.

[0034] Specifically, the image acquisition parameter optimization model for confocal microscopy is based on the principles of optical imaging and the physiological characteristics of corneal tissue. It is constructed by considering the interrelationships between parameters such as acquisition depth, resolution, and laser intensity. The derivation process first establishes a correlation function between corneal depth and acquisition resolution, describing the resolution adaptation requirements at different depths. Then, an adaptation function between laser intensity and scanning speed is constructed to balance image quality and acquisition efficiency. These two functions are integrated through a fractional structure, and an interference compensation coefficient is introduced to correct for the influence of environmental and equipment noise, forming a complete formula. The core basis for this formula is the optical characteristics of confocal microscopy and the special requirements of corneal neural imaging; different acquisition parameters need to be dynamically adapted to obtain clear neural images. Regarding parameter values, the parameter adjustment coefficient was set to 0.7 to 1.3, the acquisition depth parameter was set to 10 to 50 based on the corneal nerve distribution, the resolution parameter was set to 1024 to 4096, the laser intensity parameter was controlled between 0.1 and 0.8, the scanning speed parameter was set to 4 to 12, the interference compensation coefficient was set to 0.1 to 0.3, and the parameters of the ambient light and equipment noise influence function were calibrated to 0.05 to 0.2 based on the actual acquisition environment. During implementation, initial parameters were first determined based on the depth of the corneal acquisition area, and then the optimized acquisition control parameters were calculated using formulas. The microscope's control system adjusted hardware parameters such as laser intensity and scanning speed. This process, through dynamic optimization and adaptation of parameters, ensured the acquisition of high-quality neural tissue images under different acquisition conditions, providing reliable raw data for subsequent analysis.

[0035] Preferred, such as Figure 2 As shown, step S3 includes the following sub-steps: S31, based on the difference in gray-level distribution of nerve fibers in the confocal image, a multi-scale pyramid structure is constructed, and the edge contours and internal textures of nerve fibers are captured layer by layer through feature extraction operators at different scales; S32, the feature information extracted at each scale is reorganized dimensionally through a feature fusion matrix to establish a mapping relationship between neural morphological features and analytical scales; S33, the multi-scale neural morphology analysis model is called to iteratively calculate the reorganized feature data, gradually refining the calculation accuracy of neural branch node coordinates, fiber orientation vectors, and morphological parameters; S34, the iterative calculation results are filtered to remove redundant information unrelated to neural morphology and retain the calibrated morphological parameter matrix.

[0036] Specifically, step S3 is implemented through four sub-steps to achieve precise analysis of neural morphology. In S31, based on the difference in grayscale distribution between neural fibers and the background in the confocal image, a grayscale threshold range of 80 to 200 is set. Based on this, a five-level multi-scale pyramid structure is constructed, with each level corresponding to image scaling ratios of 1x, 2x, 4x, 8x, and 16x. Then, a combination of gradient and texture extraction operators is used to capture the neural fiber edge contours and internal textures at each scale, ensuring that key features at different scales are not missed. In S32, the 128-dimensional feature vectors extracted at each scale are reorganized using a 32×128-dimensional feature fusion matrix, establishing a one-to-one mapping between feature dimensions and the analysis scale, achieving orderly integration of multi-scale features. In S33, a multi-scale neuromorphic analysis model is invoked to iteratively calculate the reconstructed feature data. The number of iterations is set to 20, with a convergence threshold of 0.001 for each iteration. Through successive iterations, the calculation precision of nerve branch node coordinates, fiber orientation vectors, and morphological parameters is gradually refined, with coordinate precision controlled at a single pixel level and orientation vector angle precision controlled at 0.1 degrees. In S34, a feature filtering algorithm is used to remove redundant information from the iterative calculation results. A filtering threshold of 0.8 is set, retaining core morphological parameters whose feature contribution exceeds this threshold. Finally, a morphological parameter matrix including 15 key parameters such as nerve fiber diameter, length, and curvature is constructed. This step-by-step process, through layered capture, reconstructed fusion, iterative refinement, and filtering purification, ensures the comprehensiveness and accuracy of neuromorphic analysis, providing high-quality quantitative data for subsequent lesion risk analysis.

[0037] Preferred, such as Figure 3 As shown, S4 includes the following sub-steps: S41, performing feature normalization processing on the morphological parameter matrix output by S3, mapping parameters of different dimensions to a unified analysis interval; S42, constructing a risk association rule base based on the pathological characteristics of corneal neuropathy, clarifying the correspondence between abnormal morphological parameters and lesion types; S43, inputting the normalized morphological parameters into the corneal neuropathy early warning model, calculating the lesion risk index of each region; S44, dividing high, medium, and low risk regions according to the risk index threshold, generating a feature mapping map including risk level annotations.

[0038] Specifically, step S4 constructs a complete lesion risk association analysis process through four sub-steps, achieving accurate risk assessment and visualization. In S41, a linear transformation method is used to normalize the morphological parameter matrix output from S3, mapping the value ranges of different dimensions of parameters to an analysis interval of 0 to 1. The nerve density parameter is scaled according to the actual maximum and minimum values ​​of fiber length per square millimeter, and the morphological structure parameters are calibrated according to industry standard reference values ​​to ensure comparability. In S42, based on data from over 10,000 clinical cases, a rule base containing 300 risk association rules is constructed, clarifying the correspondence between abnormal nerve density, branching variations, texture disturbances, and mild, moderate, and severe lesions. The rule base supports dynamic updates to adapt to the accumulation of clinical data. In S43, the normalized morphological parameters are input into the corneal neuropathy early warning model. Weight coefficients are assigned according to the degree of abnormality of each parameter, and the lesion risk index for each corneal region is calculated through weighted summation. A step size of 0.01 is set during the calculation process to ensure the precision of the index calculation. In S44, risk index thresholds are set: 0 to 3 is the low-risk range, 3 to 7 is the medium-risk range, and 7 to 10 is the high-risk range. Based on these thresholds, the risk levels of each region are divided. Then, heatmap generation technology is used to mark regions of different risk levels in blue, yellow, and red, respectively, generating a feature mapping map that includes risk level labeling, region location, and range information. This step-by-step process, through the orderly advancement of standardization, rule construction, index calculation, and risk labeling, achieves objective quantification and intuitive presentation of lesion risk, providing a clear basis for early lesion identification.

[0039] Preferred, such as Figure 4 As shown, step S5 includes the following sub-steps: S51, performing noise detection on the feature map and identifying the regions where salt-and-pepper noise and Gaussian noise are located through the statistical distribution of pixel grayscale values; S52, calling the confocal image denoising and repair algorithm to filter the signal in the noise region and replace abnormal pixel values ​​by calculating the correlation of neighboring pixels; S53, for the broken or missing areas of nerve fibers in the image, performing structural completion based on the morphological features of adjacent regions to construct a continuous nerve fiber structure; S54, performing feature verification on the repaired image to ensure the consistency between the neural morphological parameters and the original analysis results.

[0040] Specifically, step S5 achieves synergistic optimization of image denoising and structural restoration through four sub-steps, significantly improving image quality. In S51, statistical analysis is used to detect the distribution of pixel grayscale values ​​in the feature map, calculating the mean and standard deviation of the grayscale values. A mean ± 3 times the standard deviation is set as the normal grayscale range; pixels exceeding this range are identified as salt-and-pepper noise, and areas with a standard deviation greater than 20 are identified as containing Gaussian noise. The coordinates and range information of the noise regions are recorded simultaneously. In S52, a 5×5 pixel median filter window is used for denoising the salt-and-pepper noise regions, replacing abnormal pixel values ​​with the median value within the window. For Gaussian noise regions, an adaptive Wiener filter algorithm is used, dynamically adjusting the filter kernel size according to the noise intensity, ranging from 3×3 to 9×9 pixels, ensuring that edge details of nerve fibers are preserved while removing noise. In S53, morphological operations are used to analyze the connectivity of the nerve fiber structure. Regions with more than 5 missing pixels in a continuous structure are identified as defective areas. Then, based on morphological parameters such as the diameter, direction, and curvature of nerve fibers in adjacent regions, a linear interpolation algorithm is used to supplement the pixel information of the defective areas. The interpolation step size is set to 1 pixel to ensure that the completed nerve fiber structure is continuous and consistent with the overall morphology. In S54, the repaired image is feature-verified, and the morphological parameter similarity between the repaired image and the original analysis result is calculated. A similarity threshold of 0.95 is set. If this threshold is not reached, the process returns to S52 to repeat the denoising and repair until the requirements are met. This step-by-step process achieves an optimization effect of less than 5% noise interference and greater than 95% structural integrity through a closed-loop operation of accurate noise detection, targeted denoising, structured repair, and effectiveness verification, providing highly reliable image data for subsequent result integration.

[0041] like Figure 5As shown, an intelligent analysis method for diabetic corneal neuropathy based on confocal microscopy is implemented through different units, including: a high-precision confocal image acquisition unit, used to capture raw image data of corneal nerve tissue in diabetic patients, achieve clear imaging of nerve fiber details through optical parameter adjustment, and transmit the acquired data to the next level unit; an image format conversion and feature preprocessing unit, which receives the data output from the acquisition unit, completes format standardization conversion and initial feature extraction, and establishes an adaptation interface with the subsequent analysis model; a multi-scale neuromorphic intelligent analysis unit, which calls a multi-scale neuromorphic analysis model to perform neural structure analysis on the preprocessed image data and outputs a morphological parameter matrix; a lesion risk warning and feature mapping unit, based on a corneal neuropathy warning model, performs risk analysis on the morphological parameter matrix and generates a feature mapping map annotating the lesion risk area; an image denoising and repair optimization unit, which uses a confocal image denoising and repair algorithm to perform noise filtering and structural completion on the feature mapping map and outputs an optimized nerve tissue image; and a comprehensive analysis result output unit, which receives the output data from the optimization unit, integrates the data through a diabetic neuropathy intelligent diagnosis platform, generates the final result including a morphological analysis report and lesion-related parameters, and performs a closed-loop output of the entire intelligent analysis process.

[0042] The formulas in this invention can integrate different scalar and vector parameters for unified calculation, constructing a collaborative computation logic based on the essential correlation and dimensional adaptation of parameters. Through standardization, weight allocation, and correlation mapping mechanisms, the computational barriers between different types of parameters are eliminated. Taking the formulas related to multi-scale neuromorphic analysis as an example, scalar parameters such as nerve fiber diameter and branch spacing reflect the quantitative characteristics of a single dimension, while vector parameters such as fiber orientation vectors and branch node coordinate vectors embody the directional and positional relationships. When constructing the formulas, scalar parameters are first mapped to a numerical range that matches the vector parameters through feature normalization. Then, based on the core requirements of corneal neuromorphic analysis, weight coefficients representing the importance of scalar parameters are assigned, and a directional consistency verification mechanism is designed for vector parameters, so that both contribute quantitative and spatial information respectively within the same computational framework. Meanwhile, based on the pathological mechanism of corneal neuropathy, an intrinsic relationship between scalar and vector parameters is established. For example, the degree of dispersion of the nerve fiber direction vector and the degree of abnormality of the diameter scalar are directly related to the risk of lesions. The formula quantifies this relationship through operations such as convolution and weighted summation, realizing the organic integration of different types of parameters and ensuring that the calculation results not only cover multi-dimensional information but also conform to clinical pathological logic.

[0043] Furthermore, the formula further ensures the rationality and accuracy of collaborative calculations of different scalar and vector parameters through dynamic adaptation and error compensation mechanisms. Taking formulas related to lesion warning and image optimization as examples, scalars such as risk correction coefficients and denoising weights are used to adjust the computational intensity, while vectors such as pixel coordinate vectors of feature maps and neuromorphic feature vectors are used to locate key areas. The adaptation function introduced in the formula dynamically adjusts the computational rules according to the parameter type. For scalars, a linear superposition method is used to integrate them into the overall calculation, while for vectors, effective information is extracted through vector dot product, dimension recombination, etc., avoiding logical conflicts caused by direct calculations of different types of parameters. At the same time, to address potential deviations in parameter fusion, the formula sets error correction terms, such as individual difference correction terms based on clinical data calibration and image noise compensation terms, to offset the impact of differences in the dimensions and value ranges of different parameters. For example, when fusing nerve density scalars and branch direction vectors, the error correction term dynamically adjusts the weight ratio according to their clinical relevance, ensuring that the calculation results can truly reflect the pathological state of corneal nerves, achieving efficient collaborative calculations of different types of parameters in the same formula, and providing comprehensive and accurate quantitative support for intelligent analysis.

[0044] The intelligent analysis method for diabetic corneal neuropathy based on confocal microscopy utilizes the high-precision image acquisition capabilities of confocal microscopy, combined with multi-scale neuromorphic analysis technology, to capture layered features and perform multi-dimensional analysis of corneal nerve fibers. This comprehensively acquires key morphological parameters such as nerve branch nodes and direction vectors, completely overcoming the parameter loss problem caused by traditional single-scale analysis. At the same time, specialized image denoising and restoration techniques are used to optimize image quality, and a lesion early warning model is combined to establish a precise correlation between morphological parameters and lesion risk. This forms a standardized process from image acquisition to feature analysis and risk assessment, significantly improving the objectivity and reliability of the analysis results.

[0045] This method addresses the insufficient accuracy of traditional morphological analysis by employing a multi-scale hierarchical analysis and feature fusion strategy to comprehensively cover the morphological features of different levels of nerve fibers, ensuring the completeness and accuracy of key parameter extraction. To address the shortcomings of existing technologies, such as independent modules and a lack of integration, this method utilizes an intelligent diagnostic platform to achieve collaborative scheduling of the analytical model, early warning model, and denoising and repair algorithms. It constructs a closed-loop system covering the entire process from image acquisition, format conversion, morphological analysis, risk warning, image optimization to result output, eliminating collaborative barriers between modules and significantly improving analysis efficiency and consistency. Furthermore, through functional adaptation and data interoperability among units, it provides comprehensive technical support for the early and accurate diagnosis of diabetic corneal neuropathy.

[0046] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," "connect," "link," and "fix" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal communication between two components. Those skilled in the art will understand the specific meaning of the above terms in this invention based on the specific circumstances.

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

Claims

1. A smart analysis method for diabetic corneal neuropathy based on confocal microscopy, characterized in that, Includes the following steps: S1, acquires neural tissue image data of corneal calibrated areas of diabetic patients through confocal microscopy, and obtains a set of raw images including nerve fiber morphological characteristics, distribution density and texture information; S2, input the original image set into the intelligent diagnostic platform for diabetic neuropathy, and complete the format conversion and preliminary feature extraction of the image data through the platform's built-in data interface; S3 calls a multi-scale neuromorphic analysis model to perform multi-dimensional analysis of the nerve fiber structure in the converted image data, and obtains the coordinates of nerve branch nodes, fiber orientation vectors and morphological parameter matrices. S4. Using the corneal neuropathy early warning model, the morphological parameter matrix after analysis is subjected to lesion risk correlation analysis to generate a feature mapping map including potential lesion areas. S5. A confocal image denoising and repair algorithm is used to enhance the signal and complete the structure of the regions with noise interference and structural defects in the feature map, so as to obtain an optimized neural tissue image. S6 integrates the processing results of S3 to S5 through the intelligent diagnostic platform for diabetic neuropathy, and outputs a morphological analysis report of corneal neuropathy and a set of lesion-related parameters. In this process, the intelligent analysis of the entire process of confocal microscope acquisition data is completed through the coordinated use of multi-scale neuromorphic analysis model, corneal neuropathy early warning model and confocal image denoising and repair algorithm.

2. The intelligent analysis method for diabetic corneal neuropathy based on confocal microscopy according to claim 1, characterized in that, The expression for the multi-scale neuromorphic analytical model is as follows: ,in, The results are from multi-scale neuromorphic analysis. Image pixel coordinates, The angle of nerve fiber direction. This is a parameter representing the length of nerve fibers. The number of analytical scales, For the first Scale weighting coefficients For the first Gaussian kernel function of scale, For the first Gaussian kernel standard deviation of the scale This is the convolution operator. For neuromorphic feature functions, For the first Scale-based morphological adjustment parameters. For the first Feature enhancement coefficient at scale This is the error correction term.

3. The intelligent analysis method for diabetic corneal neuropathy based on confocal microscopy according to claim 1, characterized in that, The expression for the corneal neuropathy early warning model is: ,in, As a disease early warning index, , , These are the feature weight coefficients. For abnormal nerve density parameters, For neural branch variation parameters, For parameters of neural texture disorder, It is a natural constant. This is the adjustment factor for the early warning threshold. This is a risk correction factor. These are parameters representing individual differences.

4. The intelligent analysis method for diabetic corneal neuropathy based on confocal microscopy according to claim 1, characterized in that, The expression for the confocal image denoising and restoration algorithm is: ,in, To optimize the image in Pixel value at that location, For the original image in Pixel value at that location, The number of neighboring pixels. For the first The denoising coefficients of each neighborhood Original image First The pixel value of each neighboring area. This is the range for calculating the second derivative. For the first The repair weight for each offset For the Laplace operator.

5. The intelligent analysis method for diabetic corneal neuropathy based on confocal microscopy according to claim 1, characterized in that, The collaborative scheduling model of the intelligent diagnostic platform for diabetic neuropathy is as follows: ,in, The comprehensive analysis results output by the platform The integration coefficient is the result. , , These are the weights of the model output. Output of a multi-scale neuromorphic analysis model. Output for the early warning model of corneal neuropathy This is the output of the confocal image denoising and restoration algorithm. As a collaborative correction factor, This is a set of platform system parameters.

6. The intelligent analysis method for diabetic corneal neuropathy based on confocal microscopy according to claim 1, characterized in that, The image acquisition parameter optimization model for the confocal microscope is as follows: ,in, The optimized acquisition and control parameters, For parameter adjustment coefficients, This is the correlation function between corneal region depth and acquisition resolution. To collect depth parameters, Set parameters for resolution. For the adaptation function of laser intensity and scanning speed, For laser intensity parameters, For scanning speed parameters, For interference compensation coefficient, Let f be the effect function of ambient light and equipment noise. For ambient light intensity parameters, This represents the equipment noise figure.

7. The intelligent analysis method for diabetic corneal neuropathy based on confocal microscopy according to claim 1, characterized in that, S3 includes the following steps: S31, based on the difference in gray-level distribution of nerve fibers in the confocal image, a multi-scale pyramid structure is constructed, and the edge contours and internal textures of nerve fibers are captured layer by layer using feature extraction operators at different scales; S32, the feature information extracted at each scale is reorganized dimensionally through a feature fusion matrix to establish a mapping relationship between neural morphological features and analytical scales; S33, the multi-scale neural morphology analysis model is called to iteratively calculate the reorganized feature data, gradually refining the calculation accuracy of neural branch node coordinates, fiber orientation vectors, and morphological parameters; S34, the iterative calculation results are filtered to remove redundant information unrelated to neural morphology and retain the calibrated morphological parameter matrix.

8. The intelligent analysis method for diabetic corneal neuropathy based on confocal microscopy according to claim 1, characterized in that, The S4 includes the following sub-steps: S41, performing feature normalization processing on the morphological parameter matrix output by S3, mapping parameters of different dimensions to a unified analysis interval; S42, based on the pathological characteristics of corneal neuropathy, a risk association rule base is constructed to clarify the correspondence between abnormal morphological parameters and lesion types; S43, input the normalized morphological parameters into the corneal neuropathy early warning model to calculate the lesion risk index of each region; S44: Divide high, medium and low risk areas according to the risk index threshold and generate a feature mapping map including risk level labels.

9. The intelligent analysis method for diabetic corneal neuropathy based on confocal microscopy according to claim 1, characterized in that, S5 includes the following sub-steps: S51, performing noise detection on the feature map, and identifying the regions where salt-and-pepper noise and Gaussian noise are located by the statistical distribution of pixel gray values; S52, invoke the confocal image denoising and repair algorithm to filter signals in noisy areas and replace abnormal pixel values ​​by calculating the correlation of neighboring pixels; S53, for broken or missing areas of nerve fibers in the image, perform structural completion based on the morphological features of adjacent areas to construct a continuous nerve fiber structure; S54, perform feature verification on the repaired image to ensure the consistency between the neural morphological parameters and the original analysis results.

10. The intelligent analysis method for diabetic corneal neuropathy based on confocal microscopy according to any one of claims 1-9, characterized in that, This method is implemented through different units, including: a high-precision confocal image acquisition unit, used to capture raw image data of corneal nerve tissue in diabetic patients, achieve clear imaging of nerve fiber details through optical parameter adjustment, and transmit the acquired data to the next-level unit; an image format conversion and feature preprocessing unit, which receives the data output from the acquisition unit, completes format standardization conversion and initial feature extraction, and establishes an interface for adaptation with subsequent analysis models; a multi-scale neuromorphic intelligent analysis unit, which calls a multi-scale neuromorphic analysis model to perform neural structure analysis on the preprocessed image data and outputs a morphological parameter matrix; a lesion risk warning and feature mapping unit, which performs risk analysis on the morphological parameter matrix based on a corneal neuropathy warning model and generates a feature mapping map annotating lesion risk areas; an image denoising and repair optimization unit, which uses a confocal image denoising and repair algorithm to perform noise filtering and structural completion on the feature mapping map and outputs an optimized nerve tissue image; and a comprehensive analysis result output unit, which receives the output data from the optimization unit, integrates the data through a diabetic neuropathy intelligent diagnosis platform, generates the final result including a morphological analysis report and lesion-related parameters, and performs a closed-loop output of the entire intelligent analysis process.

Images