Preoperative corneal epithelium thickness measuring method based on slit lamp CCD three-dimensional reconstruction

Abstract

The invention discloses a preoperative corneal epithelium thickness measuring method based on slit lamp CCD three-dimensional reconstruction, and relates to the technical field of slit lamp microscope and CCD imaging, and the method comprises the steps: extracting information from an electronic medical record based on template initialization driven by the medical history of a patient, and automatically loading an individualized epithelium thickness initial template; a modified slit lamp CCD is used for multi-angle synchronous acquisition, tear film stability is monitored in real time, and high-fidelity epithelium point cloud data is generated; an elastic deformation-Kalman coupling fitting technology is adopted, a template is adaptively adjusted according to the point cloud residual error and the curvature gradient, a high-confidence-coefficient thickness grid is generated through curvature continuity filtering and optical reprojection correction, and an error distribution label is added to each pixel; the data is packaged and then pushed to a surgical navigation platform, and when the confidence coefficient is low, re-collection is automatically prompted; the accuracy, traceability and reliability of measurement data are improved, and robust support is provided for early lesion screening and preoperative planning.

Description

Technical Field

[0001] This invention relates to the field of slit-lamp microscopy combined with CCD imaging technology, specifically a method for measuring preoperative corneal epithelial thickness based on slit-lamp CCD three-dimensional reconstruction. Background Technology

[0002] In preoperative assessments for refractive surgery and pathological screening, clinical teams are increasingly relying on charge-coupled device (CCD) multi-angle imaging and stereoscopic reconstruction algorithms with modified slit lamps to quickly generate corneal epithelial thickness maps and determine whether and how much ablation to perform. For normal corneas without a history of surgery, this process typically relies on a generic template derived from healthy individuals to perform point cloud fitting and thickness coloring.

[0003] However, once the subject has undergone corneal remodeling surgery such as excimer laser in situ keratomileusis or micro-incision lenticule extraction, or is in the early stages of diseases such as keratoconus or dry eye, the epithelial morphological remodeling caused by trauma repair and biomechanical stress will break the template hypothesis, resulting in features such as central thinning, peripheral thickening, and asymmetric undulation. These changes have been repeatedly confirmed by multiple clinical imaging studies and have been shown to be closely related to visual quality.

[0004] The following issues arise: First, this individualized remodeling phenomenon occurs because the corneal epithelium possesses the physiological characteristics of rapid proliferation and migration. Driven by postoperative optical zone concavity or changes in local stress, the epithelium will actively thicken or thin to restore the smoothness of curvature. Second, in the slit-lamp charge-coupled device stereoscopic measurement link, the algorithm, in pursuit of speed and stability, still defaults to a smooth, gradual distribution from the center to the periphery of the healthy template. During registration, it forcibly stretches or compresses the local point cloud to fit the template, resulting in a systematic drift of the thickness value. Finally, when the distorted thickness map is directly sent to the surgical planning or early lesion screening stages, doctors may misjudge the thickness safety margin, ignore subtle asymmetric lesions, or even initiate ablation in potential keratoconus eyes, increasing the risk of postoperative complications.

[0005] Multiple studies have warned that only by identifying and correcting template inaccuracies during the measurement phase can epithelial thickness be truly used for surgical safety control and early pathological screening. Therefore, this issue has become a key technical bottleneck for the widespread clinical application of slit-lamp charge-coupled device (CCD) stereoscopic reconstruction. Summary of the Invention

[0006] (a) Technical problems to be solved To address the shortcomings of existing technologies, this invention provides a method for measuring preoperative corneal epithelial thickness based on slit-lamp CCD 3D reconstruction. This method extracts information from electronic medical records and automatically loads an individualized initial epithelial thickness template. A modified slit-lamp CCD is used for multi-angle synchronous acquisition, real-time monitoring of tear film stability, and generation of high-fidelity epithelial point cloud data. Elastic deformation-Kalman coupling fitting technology is employed to adaptively adjust the template based on point cloud residuals and curvature gradients. High-confidence thickness grids are generated through curvature continuity filtering and optical reprojection correction, with an error distribution label attached to each pixel. The data is encapsulated and pushed to a surgical navigation platform, automatically prompting for re-acquisition when the confidence level is low. This improves the accuracy, traceability, and reliability of the measurement data, solving the technical problems described in the background section.

[0007] (II) Technical Solution To achieve the above objectives, the present invention is implemented through the following technical solution: a method for measuring preoperative corneal epithelial thickness based on slit-lamp CCD three-dimensional reconstruction, comprising: selecting an initial epithelial thickness template that matches the patient's ocular features from a pre-constructed template library based on extracted labels; locating the corneal center on the slit-lamp CCD imaging interface through edge detection and Hough transform; and aligning the selected template with the corneal center. By modifying the slit-lamp CCD equipment, multi-angle synchronous scanning was performed to monitor tear film stability and record the reflectance intensity spectrum; after generating epithelial point cloud data using a stereo vision algorithm, the point cloud was spatially aligned with the template. After calculating the residual elastic relaxation index and curvature gradual stability index based on the residual field and curvature gradient, the individualized fitting confidence coefficient is output using the Bayesian generalized linear deformation confidence model. Adaptive fitting is completed through dynamic coupling of elastic deformation and Kalman filtering to generate an individualized epithelial surface. Curvature continuity filtering and optical reprojection correction are performed between the fitted surface and the original point cloud to generate a high-confidence thickness raster, and an error distribution label is attached to each pixel. After encapsulating the thickness grid, error labels, and anomaly alerts, the data is pushed to the surgical navigation platform, and a prompt to re-collect data is automatically generated when the confidence level is low.

[0008] Furthermore, medical history data, including refractive surgery records and pathology labels, are extracted from electronic medical records to identify patient-specific corneal features; based on the extracted medical history data, an initial epithelial thickness template matching the patient's corneal features is selected from a pre-built template library. The corneal center was located using image processing techniques on a slit-lamp CCD imaging system, and the selected initial epithelial thickness template was aligned with the located corneal center.

[0009] Furthermore, the scanning path is planned using the geometric features of the personalized epithelial thickness initial template to ensure that the scanning angle and position match the actual morphology of the patient's cornea; Multi-angle synchronous acquisition is performed by time synchronization triggering, tear film stability monitoring and reflectance intensity spectrum recording to obtain high-quality image data at the same time baseline.

[0010] Furthermore, stereo vision algorithms and voxel filtering techniques are used to generate epithelial point cloud data to ensure the accuracy and reliability of the data. Through coarse and fine alignment, the epithelial point cloud data is spatially aligned with the personalized initial template for epithelial thickness, thereby improving the accuracy of subsequent fitting and thickness measurement.

[0011] Furthermore, after calculating the residual fields of the epithelial point cloud data and the personalized epithelial thickness initial template, the local deviation is determined by quantifying the Euclidean distance between the two, and the curvature gradient of the epithelial point cloud data and the personalized epithelial thickness initial template is analyzed to identify areas of abrupt changes in surface morphology.

[0012] Furthermore, the residual elastic relaxation index and curvature gradient stability index are calculated based on the residual field and curvature gradient to dynamically control the fitting strength and stability. The reliability of the fit is evaluated using a Bayesian generalized linear deformation confidence model, and individualized fit confidence coefficients are output. The data and smoothing terms in the energy function are optimized through the dynamic coupling of elastic deformation and Kalman filtering, and adaptive fitting is achieved through the state update formula until the residual mean converges to a preset threshold, generating an individualized epithelial surface.

[0013] Furthermore, the curvature distribution of the individualized epithelial surface is calculated to identify regions of abnormal curvature. The Laplacian smoothing algorithm is applied to adjust the vertex positions of the surface to ensure a smooth surface shape. The original epithelial point cloud data is projected onto the image plane using camera intrinsic and extrinsic parameters.

[0014] Furthermore, the rotation matrix and translation vector of the individualized epithelial surface are optimized by calculating the reprojection error to improve spatial consistency by generating a base surface below the individualized epithelial surface; The shortest distance from a point on the surface to the base surface is calculated to generate a two-dimensional thickness raster image. The fitting residuals are calculated, and the mean and standard deviation of the residuals within the pixel coverage area of ​​the thickness raster image are statistically analyzed to generate an error distribution image to evaluate the measurement reliability.

[0015] Furthermore, the thickness raster image, error distribution image, and anomaly alert information are encapsulated into objects of the Digital Imaging and Communication Medicine Standard Format and the EyeCare Extended Data Structure, a rapid medical interoperability resource, and pushed to the surgical navigation platform through the Health Level 7 protocol and the Representational State Transition Architecture application programming interface.

[0016] Furthermore, the proportion of pixels with a residual mean of less than 0.05 mm is calculated based on the error distribution image as the confidence level; When the confidence level is below 0.9, a warning message is displayed through the user interface, and the confidence level value, abnormal message, and feedback record are stored in the audit log.

[0017] Preferred, (III) Beneficial Effects This invention provides a method for measuring preoperative corneal epithelial thickness based on slit-lamp CCD three-dimensional reconstruction, which has the following beneficial effects: By extracting patients' past refractive surgeries or pathology labels from electronic medical records, an individualized initial epithelial thickness template that matches the patient's ocular characteristics is automatically loaded, improving the matching degree between the template and the patient's actual corneal morphology. The history-driven template initialization works closely with subsequent point cloud acquisition, fitting and other steps to ensure the high degree of individualization and accuracy of the entire measurement process. Based on the modification of the slit-lamp CCD equipment, multi-angle synchronous acquisition and point cloud shaping were achieved, significantly improving the quality and reliability of epithelial point cloud data. By monitoring tear film stability in real time and recording the reflectance spectrum, it was ensured that the acquired data obtained an epithelial point cloud dataset with consistent spatial coordinates under the same time baseline, effectively avoiding acquisition errors caused by eye movement or tear film fluctuations. The combination of multi-angle synchronous acquisition technology and history-driven template guidance optimized the acquisition path, making the point cloud data more refined in morphologically complex areas. The template is adaptively adjusted based on point cloud-template residual field and curvature gradient. By calculating the residual elastic relaxation index and curvature gradient stability index, and outputting individualized fitting confidence coefficients through a Bayesian generalized linear deformation confidence model, a balance between local stress release and global curvature smoothness can be achieved. The fitting strategy is dynamically adjusted according to the individualized characteristics of the patient's cornea, effectively avoiding systematic drift in thickness values. Applying curvature continuity filtering and optical reprojection correction between the fitted surface and the original point cloud significantly improves the reliability and accuracy of the thickness grid. Curvature continuity filtering ensures the physiological rationality of the surface morphology by smoothing local discontinuities and noise; optical reprojection correction reduces registration bias by optimizing the spatial consistency between the surface and point cloud data. Adding an error distribution label to each pixel quantifies the reliability of the measurement results, allowing clinicians to intuitively understand the confidence level of the data. Through a confidence assessment mechanism, the system automatically prompts for re-collection when the confidence level is low, effectively avoiding clinical misjudgments caused by data distortion. This unified interface output and navigation linkage technology not only improves data exchangeability and traceability but also enhances the reliability of clinical applications through quality self-checking functions. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the preoperative corneal epithelial thickness measurement method of the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Please see Figure 1 This invention provides a method for measuring preoperative corneal epithelial thickness based on slit-lamp CCD three-dimensional reconstruction, including: Step 1: Extract medical history data from the electronic medical record, including refractive surgery records and pathology labels, to identify patient-specific corneal features; based on the extracted medical history data, select an initial epithelial thickness template that matches the patient's corneal features from a pre-built template library; use image processing technology on slit-lamp CCD imaging to locate the corneal center to determine precise spatial coordinates; align the selected initial epithelial thickness template with the located corneal center to establish an accurate spatial reference for subsequent three-dimensional registration.

[0021] Step one includes the following: Step 101: Extraction of medical history data During the patient history data extraction process, the first step is to retrieve past refractive surgery records and pathology labels from the patient's electronic medical record system. These records include information on laser-assisted in situ keratomileusis (LASIK), micro-lens resection, keratoconus, and dry eye syndrome. Natural language processing (NLP) technology is used to parse the electronic medical record text, identify and extract key fields related to the eye, and form a structured set of medical history labels. For example, for patients who have undergone LASIK, the extracted labels indicate that their corneal epithelium may be thinner in the central region; for keratoconus patients, the labels suggest that the thickness distribution may be irregular. After extraction, the medical history labels are stored as key-value pairs for subsequent steps.

[0022] By extracting medical history data from electronic medical records, the specific characteristics of a patient's eyes can be accurately reflected, avoiding reliance on generic assumptions about healthy individuals. Structured labels provide a personalized basis for subsequent template selection, thereby improving the accuracy of corneal epithelial thickness measurements and reducing errors caused by individual differences.

[0023] Step 102: Template Library Construction During the template library construction process, a database containing epithelial thickness templates corresponding to various ocular features was generated in advance based on statistical analysis and machine learning methods using a large amount of clinical data. Specifically, corneal epithelial thickness distribution data of patients with different medical histories and pathological states were collected, such as the central thinning and peripheral thickening patterns of patients after laser-assisted in situ keratomileusis (LASIK), or the asymmetric thickness distribution patterns of patients with keratoconus. Cluster analysis was used to group the data, and a representative template was generated for each group. The templates were stored in a parametric form, including thickness values ​​and corresponding spatial coordinates. The database is updated regularly to cover more patient types.

[0024] When used, building a diverse template library can cover a wide range of patient characteristics, ensuring that there is a corresponding template for the typical distribution of epithelial thickness for each medical history or pathological condition, so that the subsequent matching process can more accurately reflect the actual situation of the patient.

[0025] Step 103, Template Selection During template selection, based on medical history tags extracted from electronic medical records, the initial epithelial thickness template that best matches the patient's ocular characteristics is automatically matched from the template library. For patients with a single medical history tag, such as those who have only undergone laser-assisted in situ keratomileusis (LASIK), the corresponding template is directly retrieved from the template library, characterized by central region thinning. For patients with multiple medical history tags, such as those with both LASIK and dry eye syndrome, a decision tree model is used to determine the influence of the primary medical history, prioritizing the LASIK template, and then weighting the local thickness according to the dry eye syndrome characteristics to generate a comprehensive template. The adjustment process is based on predefined weighting coefficients, comprehensively considering the cumulative effect of multiple medical histories.

[0026] In practice, templates are selected through automatic matching and combination, generating highly personalized initial templates for patients with complex medical histories. This method improves the fit between the template and the patient's ocular characteristics, ensuring that subsequent measurements accurately reflect individualized epithelial morphological changes.

[0027] Step 104, Corneal center positioning During corneal center localization, image data from the slit-lamp CCD imaging interface is used to determine the two-dimensional coordinates of the corneal center through image processing techniques. Specific steps include: first, applying edge detection algorithms, such as Canny edge detection, to extract the corneal boundary contour and generate a continuous set of edge points; then, using Hough transform to process the edge point set, and determining the best-fit circle for the corneal boundary by detecting circular features; the center of this circle is the corneal center coordinate. Localization accuracy requires multiple iterative optimizations to ensure the error is controlled at the sub-pixel level, i.e., less than one pixel.

[0028] In practice, precise corneal center localization provides a reliable spatial reference point for aligning the template with the actual cornea. A combination of edge detection and Hough transform is employed, leveraging the geometric properties of the image to ensure stable localization results unaffected by noise, thereby improving the accuracy of subsequent registration processes.

[0029] Step 105: Template Registration Preparation During the template registration preparation process, the selected initial epithelial thickness template is spatially aligned with the corneal center coordinates.

[0030] The specific steps include: first, translating the spatial coordinate system of the template so that the center point of the template coincides with the corneal center coordinates determined through image processing; then, adjusting the rotation angle of the template according to the anatomical axis of the cornea, such as the horizontal or vertical direction, so that the principal axis of the template is consistent with the natural axis of the cornea. After alignment, the spatial position of the template and the actual position of the cornea are consistent on the two-dimensional plane, providing a reference mapping surface for subsequent multi-angle point cloud acquisition.

[0031] Precise alignment between the template and the corneal center is achieved through translation and rotation adjustments, ensuring that the spatial distribution of the template remains consistent with the actual corneal morphology in the initial stage. This method improves the accuracy of mapping point cloud data onto the template surface, thereby enhancing the reliability and precision of the entire 3D reconstruction process.

[0032] When in use, the system extracts medical history tags from electronic medical records, constructs a diverse template library, selects a matching initial template, locates the corneal center, and completes template alignment, generating a personalized initial template for epithelial thickness that reflects the patient's ocular characteristics. This effectively overcomes the limitations of traditional universal templates that cannot adapt to individualized epithelial morphological changes, and improves the accuracy and clinical applicability of corneal epithelial thickness measurement in slit-lamp CCD three-dimensional reconstruction.

[0033] Step 2: Plan the scanning path using the geometric features of the personalized initial epithelial thickness template to ensure that the scanning angle and position match the actual morphology of the patient's cornea; perform multi-angle synchronous acquisition through time synchronization triggering, tear film stability monitoring, and reflectance intensity spectrum recording to obtain high-quality image data at the same time baseline; generate epithelial point cloud data using stereo vision algorithms and voxel filtering technology to ensure the accuracy and reliability of the data; align the epithelial point cloud data with the personalized initial epithelial thickness template in space through coarse and fine alignment to improve the accuracy of subsequent fitting and thickness measurement.

[0034] Step two includes the following: Step 201: Scanning path planning based on a personalized epithelial thickness initial template During the scanning path planning process, the corneal surface geometry provided by the personalized epithelial thickness initial template is first used to determine the scanning angle and position of the slit-lamp CCD device.

[0035] The specific method involves extracting the corneal surface normal vector and curvature distribution from a personalized initial epithelial thickness template. The normal vector represents the direction of each point on the corneal surface, while the curvature distribution reflects the local unevenness of the corneal surface. The incident angle of the scanning beam is calculated based on the normal vector, ensuring that the beam is as perpendicular as possible to the corneal surface, thereby enhancing the intensity and quality of the reflected signal. Simultaneously, the corneal surface is divided into multiple regions according to the curvature distribution, and the scanning path is determined as a series of continuous angle sequences, covering all areas from the center to the periphery of the cornea to achieve comprehensive data acquisition. The planned scanning path guides the subsequent multi-angle synchronous acquisition process.

[0036] By utilizing the geometric features of a personalized initial epithelial thickness template to plan the scanning path, it is possible to ensure that the scanning angle and position match the actual morphology of the patient's cornea, thereby improving the quality of the reflected signal and the representativeness of the acquired data, and thus enhancing the accuracy of point cloud generation and thickness measurement.

[0037] Step 202: Multi-angle synchronous acquisition During the multi-angle synchronous acquisition process, the modified slit lamp CCD equipment performs scanning according to the planned scanning path. The specific steps include the following.

[0038] First, the movements of the CCD camera and the light source are coordinated by a high-precision clock signal to ensure that multi-angle image acquisition is completed under the same time baseline, avoiding spatial deviations caused by eye movements. This process is called time synchronization triggering.

[0039] Secondly, during the scanning process, an infrared sensor is used to measure the tear film breakup time, and data is collected only during the tear film stable period. The tear film stable period is defined as a time window in which the tear film breakup time is greater than or equal to 5 seconds, in order to reduce the interference of tear film fluctuations on the reflected signal. This step is called tear film stability monitoring.

[0040] Finally, at each scanning angle, the CCD device captures the intensity of the light beam reflected from the corneal surface via a photodiode array and converts it into a digital signal, storing it as an intensity matrix. The rows and columns of the intensity matrix correspond to the azimuth and elevation angles of the scan, respectively, and the matrix element values ​​reflect the intensity of the reflected signal at the corresponding location. This process is called reflection intensity spectrum recording.

[0041] The collected intensity matrix provides raw data for the subsequent generation of epithelial point cloud data.

[0042] Time-synchronized triggering ensures spatial consistency of the acquired data, while tear film stability monitoring improves data quality and stability. Together, they prevent interference from eye movement and tear film fluctuations. Reflectance intensity spectrum recording provides detailed information on the corneal surface reflectance characteristics, which helps in the accurate generation of epithelial point cloud data.

[0043] Step 203: Generation of epithelial point cloud data In the process of generating epithelial point cloud data, based on image data acquired simultaneously from multiple angles, an epithelial point cloud is generated using a stereo vision algorithm. The specific steps are as follows: First, in images from different angles, the scale-invariant feature transform algorithm is used to extract key feature points on the corneal surface, such as edges or areas with significant texture. Then, the correspondence between feature points is established between images using the nearest neighbor matching method. This step is called feature point matching.

[0044] Secondly, based on the pixel coordinates of the matched feature points and the intrinsic and extrinsic parameters of the CCD camera, the three-dimensional spatial coordinates of the feature points are calculated using the principle of triangulation. Specifically, a coefficient matrix is ​​constructed using the camera projection matrix and the pixel coordinates of the feature points. The three-dimensional coordinates of the feature points are then obtained by solving a system of linear equations; this step is called three-dimensional coordinate calculation.

[0045] Finally, the calculated three-dimensional coordinate set is organized into a point cloud data structure, and noise points and outliers are removed by voxel filtering. Voxel filtering divides the point cloud into a three-dimensional grid, retaining only the representative points in each grid. The grid size is set to 0.1 mm to balance data density and quality. This step is called point cloud organization and filtering. The generated epithelial point cloud data contains three-dimensional spatial information of the corneal surface.

[0046] The application of feature point matching and triangulation principles ensures the accuracy and reliability of point cloud data, while voxel filtering technology effectively removes noise and outliers, improving the quality and usability of point cloud data.

[0047] Step 204: Spatial alignment of point cloud with the initial template of personalized epithelial thickness In the spatial alignment process between the point cloud and the personalized epithelial thickness initial template, coarse alignment is first performed. Using the corneal center located in step one as a reference, the geometric center of the epithelial point cloud is aligned with the center point of the personalized epithelial thickness initial template, completing the initial position matching. Subsequently, an iterative nearest-point algorithm is used to optimize the rotation and translation parameters of the point cloud by minimizing the Euclidean distance between the surface of the epithelial point cloud and the personalized epithelial thickness initial template. Specifically, the optimization process involves repeatedly calculating the distance between each point in the epithelial point cloud and the nearest point on the surface of the personalized epithelial thickness initial template, and adjusting the position and orientation of the epithelial point cloud until the distance error converges to below 0.01 mm. This step is called fine alignment. The aligned epithelial point cloud data is spatially consistent with the personalized epithelial thickness initial template.

[0048] The combination of coarse and fine alignment ensures accurate alignment between the epithelial point cloud data and the personalized initial template for epithelial thickness. The iterative nearest point algorithm improves the spatial consistency between the two by optimizing the spatial position of the epithelial point cloud, thereby enhancing the accuracy of subsequent fitting and thickness measurement.

[0049] In practice, high-quality epithelial point cloud data was successfully obtained by planning the scanning path based on a personalized initial template of epithelial thickness, synchronously acquiring data from multiple angles, generating epithelial point clouds using stereo vision algorithms, and employing spatial alignment techniques. Tear film stability monitoring and point cloud filtering improved the reliability of the data, while the iterative nearest-point algorithm ensured the spatial consistency between the epithelial point cloud data and the personalized initial template of epithelial thickness.

[0050] Step 3: Calculate the residual field between the epithelial point cloud data and the personalized initial epithelial thickness template. Determine the local deviation by quantifying the Euclidean distance between the two. Analyze the curvature gradient of the epithelial point cloud data and the personalized initial epithelial thickness template to identify abrupt changes in surface morphology. Calculate the residual elastic relaxation index and curvature gradient stability index based on the residual field and curvature gradient to dynamically control the fitting strength and stability. Use the Bayesian generalized linear deformation confidence model to evaluate the fitting reliability and output the individualized fitting confidence coefficient. Optimize the data term and smoothing term in the energy function through the dynamic coupling of elastic deformation and Kalman filtering, and achieve adaptive fitting through the state update formula until the mean of the residual field converges to the preset threshold, generating the individualized epithelial surface.

[0051] Step three includes the following: Step 301: Point Cloud-Template Residual Field Calculation In the point cloud-template residual field calculation process, the spatial deviation between the epithelial point cloud data and the personalized epithelial thickness initial template is first quantified.

[0052] The specific method is as follows: for each point in the epithelial point cloud data, determine its nearest point on the personalized epithelial thickness initial template surface, and calculate the straight-line distance between the two points, i.e., the Euclidean distance; by collecting the Euclidean distances of all points, construct a set that reflects the spatial distribution, called the residual field, which represents the local deviation between the epithelial point cloud data and the personalized epithelial thickness initial template surface.

[0053] The calculation of the residual field provides a data-driven basis for revealing the differences between the epithelial point cloud data and the personalized initial template for epithelial thickness. An accurate residual field can clearly reflect the areas that need adjustment and the degree of deviation, thus providing a clear direction and range for subsequent elastic deformation and ensuring more accurate correction of the personalized initial template for epithelial thickness.

[0054] Step 302, Curvature Gradient Analysis During curvature gradient analysis, curvature gradient calculations were performed on epithelial point cloud data and personalized epithelial thickness initial templates to extract local morphological features of the corneal epithelial surface.

[0055] The specific method involves first calculating the Gaussian curvature at each point on the surface. The Gaussian curvature is determined by the product of the curvatures in the two principal directions at that point, reflecting the local degree of bending. Then, by comparing the Gaussian curvatures of adjacent points, the spatial rate of change of the Gaussian curvature on the surface is calculated using the finite difference method, generating a curvature gradient. The intensity of the curvature gradient is represented by its norm, which is used to identify regions of abrupt changes in surface morphology, such as significant changes at the edges of the optical zone after surgery.

[0056] Curvature gradient analysis can accurately capture localized morphological abnormalities on the corneal epithelial surface, such as mutations caused by surgery or pathology. This analytical method provides crucial geometric feature information, enabling subsequent deformation adjustments to specifically address complex morphological changes, thereby improving the adaptability of the fitting results to actual corneal epithelial characteristics.

[0057] Step 303: Calculation of Residual Elastic Relaxation Index and Curvature Gradual Change Stability Index In the calculation of the residual elastic relaxation index and the curvature gradual stability index, two indices are introduced to achieve dynamic control for adaptive fitting.

[0058] The specific method involves calculating the residual elastic relaxation index based on the distribution characteristics of the residual field and curvature gradient using a logistic function. This function maps the values ​​of the residual field and curvature gradient to a range of 0 to 1, reflecting the intensity of deformation adjustment. The curvature gradient stability index is determined by calculating the reciprocal of the local rate of change, based on the spatial consistency of the curvature gradient distribution. A larger value indicates a smoother curvature distribution and thus greater stability of the morphological changes.

[0059] The residual elastic relaxation index and the curvature gradient stability index provide dynamically adjustable parameters for the fitting process. They can differentiate the fitting based on the magnitude of local deviations and the complexity of morphological features, ensuring finer adjustments in complex regions while maintaining stability in smooth regions, thereby improving the overall adaptability and accuracy of the fitting.

[0060] Step 304, Bayesian Generalized Linear Deformation Credibility Model In the process of developing a Bayesian generalized linear deformation confidence model, the reliability of the fitting results is evaluated, and an individualized fitting confidence coefficient is output. Specifically, the residual elastic relaxation index and the curvature gradual stability index are used as input features, and the posterior probability of the fitting reliability is calculated within the Bayesian framework. The expected value of the posterior probability is defined as the individualized fitting confidence coefficient, which ranges from 0 to 1, representing the degree of confidence of the fitting results.

[0061] When used, the introduction of individualized fitting confidence coefficients provides a means to quantitatively assess the reliability of fitting results. It can dynamically reflect the reliability of the fitting based on the complexity and degree of deviation of local features, ensuring that appropriate fitting strategies are adopted in different regions, thereby improving the reliability and accuracy of the overall results.

[0062] Step 305: Dynamic Coupling of Elastic Deformation and Kalman Filtering In the dynamic coupling process of elastic deformation and Kalman filtering, the parameters of elastic deformation and Kalman filtering are dynamically adjusted based on the residual elastic relaxation index, the curvature gradual stability index, and the individualized fitting confidence coefficient to achieve adaptive fitting. Specifically, elastic deformation is first applied to the personalized epithelial thickness initial template. The deformation field is determined by optimizing an energy function containing data and smoothing terms, where the weight of the smoothing term is jointly determined by the residual elastic relaxation index and the curvature gradual stability index.

[0063] Then, the deformed template is used as the prior state for Kalman filtering and optimized using a state update formula. The gain in the state update formula is calculated from the state covariance and the observation noise covariance, and the magnitude of the observation noise covariance is adjusted by the individualized fitting confidence coefficient. The iterative process of elastic deformation and Kalman filtering is repeated until the average value of the residual field converges to a preset threshold, ultimately generating an individualized epithelial surface.

[0064] In practice, the dynamic coupling method of elastic deformation and Kalman filtering achieves a balance between global smoothness and local fine adjustment. The data and smoothing terms in the energy function ensure that the deformation process fits the epithelial point cloud data while maintaining the continuity of the surface morphology. Kalman filtering further optimizes the fitting results through state updates, which can solve the problem of mismatch between complex morphological changes caused by surgical history or pathology and the initial template of personalized epithelial thickness, thus improving the accuracy of the fitting results.

[0065] By calculating the residual field and curvature gradient between epithelial point cloud data and a personalized initial epithelial thickness template, the residual elastic relaxation index and curvature gradient stability index are introduced. Combined with the individualized fitting confidence coefficient output by the Bayesian generalized linear deformation confidence model, elastic deformation and Kalman filtering are dynamically coupled to achieve adaptive adjustment of the personalized initial epithelial thickness template. This effectively handles the situation where complex morphological changes caused by surgical history or pathology do not match the template. The resulting personalized epithelial surface more accurately reflects the actual corneal epithelial characteristics, providing reliable support for subsequent thickness grid generation and clinical decision-making.

[0066] Step 4: Calculate the curvature distribution of the individualized epithelial surface to identify areas of abnormal curvature and apply the Laplacian smoothing algorithm to adjust the vertex positions of the surface to ensure smooth surface shape. Project the original epithelial point cloud data onto the image plane using camera intrinsic and extrinsic parameters and calculate the reprojection error to optimize the rotation matrix and translation vector of the individualized epithelial surface to improve spatial consistency. Generate a 2D thickness raster image by generating a base surface below the individualized epithelial surface and calculating the shortest distance from the surface points to the base surface. Calculate the fitting residuals and statistically analyze the mean and standard deviation of the residuals within the pixel coverage area of ​​the thickness raster image to generate an error distribution image to evaluate measurement reliability.

[0067] Step four includes the following: Step 401, Curvature Continuity Filtering In the curvature continuity filtering process, the curvature distribution of the individualized epithelial surface is first calculated.

[0068] The specific method involves calculating the Gaussian curvature of each point on the surface. The Gaussian curvature is determined by the product of the curvature values ​​in the two principal directions at that point, reflecting the local degree of curvature on the surface. Next, based on Gaussian curvature statistics from healthy individuals, a normal range, or 95% confidence interval, is determined. Regions with curvature values ​​exceeding this range are identified and marked as curvature aberration areas. Subsequently, a Laplace smoothing algorithm is applied to these curvature aberration areas to adjust the positions of the surface vertices. Each vertex is moved towards the average position of its neighboring vertices, with the movement step controlled by a smoothing coefficient, which is fixed to a preset value. This smoothing process is repeated until the rate of change of curvature falls below a preset threshold.

[0069] In practice, curvature continuity filtering effectively eliminates local discontinuities and noise on individualized epithelial surfaces by identifying regions of abnormal curvature and applying the Laplacian smoothing algorithm, ensuring that the surface morphology is physiologically reasonable and smooth. This method improves the continuity and stability of the surface, making it more consistent with actual anatomical features and providing high-quality surface data for subsequent processing.

[0070] Step 402, Optical Reprojection Correction In the optical reprojection correction process, the original epithelial point cloud data is first projected onto the image plane using the camera's intrinsic and extrinsic parameters from the slit-lamp CCD device, generating a reprojection point set. Specifically, based on the camera projection model, the 3D point cloud data is converted into 2D pixel coordinates. Next, the pixel distance error between the reprojection point set and the actual feature points of the image acquired in step two is calculated. Specifically, for each reprojection point, the Euclidean distance between it and the corresponding actual feature point is calculated, and the average of the Euclidean distances for all feature points is obtained to obtain the average error. Based on this average error, the spatial position of the individualized epithelial surface is optimized. Specifically, the rotation matrix and translation vector of the surface are adjusted to minimize the reprojection error, ensuring the consistency of the surface's projection with the original epithelial point cloud data on the image plane.

[0071] In practice, optical reprojection correction improves the spatial consistency between the individualized epithelial surface and the original epithelial point cloud data by minimizing reprojection error. This method utilizes projection relationships on the image plane to reduce registration bias, ensuring that the surface model visually matches the actual acquired data, thereby enhancing the accuracy of the surface model.

[0072] Step 403: Thickness grid generation In the thickness raster generation process, a fixed distance, such as 5 micrometers, is first offset along the normal direction below the individualized epithelial surface to generate a basal surface, representing the surface of the corneal stroma. Next, along the normal direction, the shortest distance from each point on the individualized epithelial surface to the basal surface is calculated as the thickness value at that point. Specifically, for each point on the individualized epithelial surface, the nearest point on the basal surface is searched, and the Euclidean distance between the two points is calculated. Subsequently, these thickness values ​​are mapped to a two-dimensional raster image with a resolution of 0.1 mm × 0.1 mm, generating a thickness raster image that reflects the thickness distribution of the corneal epithelium.

[0073] In practice, thickness raster generation provides a visually intuitive distribution image of corneal epithelial thickness by calculating the shortest distance from individualized epithelial surface points to the basal surface. This method ensures the accuracy of thickness measurements, while the two-dimensional raster image format facilitates analysis and understanding by clinicians, providing important reference data for surgical planning and pathological screening.

[0074] Step 404: Error Distribution Label Generation In the error distribution label generation process, the distance between points on the corrected individualized epithelial surface and the original epithelial point cloud data is first calculated and denoted as the fitting residual. Next, for each pixel-covered area of ​​the surface points in the thickness raster image, the corresponding fitting residuals are collected, and their mean and standard deviation are calculated. Specifically, the fitting residuals within each pixel region are statistically analyzed, and their mean and standard deviation are calculated and used as the error distribution label for that pixel, ultimately generating the error distribution image.

[0075] In practice, the generation of error distribution labels provides a reliability assessment of thickness measurement results by statistically fitting the mean and standard deviation of the residuals. This method quantifies the measurement error for each region, enabling clinicians to understand the accuracy distribution of the data, helping to identify areas where bias may exist, thereby improving data transparency and clinical applicability.

[0076] Local discontinuities are eliminated by smoothing the individualized epithelial surface, and then optical reprojection correction is used to ensure spatial consistency between the surface and the original epithelial point cloud data. Based on this, a thickness raster image is generated to reflect the corneal epithelial thickness distribution, and an error distribution labeling process is used to add a reliability assessment to the measurement results. The individualized epithelial surface is then progressively optimized, ultimately providing high-quality, reliable corneal epithelial thickness data.

[0077] Step 5: Encapsulate the thickness raster image, error distribution image, and anomaly alert information into objects in the digital imaging and communication medical standard format and the EyeCare extended data structure, a rapid medical interoperability resource. Push these objects to the surgical navigation platform via the Health Level 7 protocol and the Representational State Transition Architecture application interface. Calculate the proportion of pixels with a residual mean value lower than 0.05 mm based on the error distribution image as the confidence level. When the confidence level is lower than 0.9, display a warning message through the user interface prompting you to re-execute steps 2 to 4. Simultaneously, store the confidence level value, anomaly alert information, and feedback records in the audit log module.

[0078] Step five includes the following: Step 501, Data Encapsulation First, the thickness raster image and error distribution image generated in step four are encapsulated.

[0079] The thickness raster image is a two-dimensional pixel array, where each pixel value represents the thickness of the corneal epithelium in millimeters. The error distribution image is also a two-dimensional pixel array, where each pixel value represents the mean and standard deviation of the fitted residuals, reflecting the error in the measurement data. The data encapsulation process involves converting the thickness raster image and error distribution image into a format conforming to the Digital Imaging and Communication Medicine (DICOM) standard. During the conversion, key parameters from the measurement process are recorded in the metadata fields of the DICOM file, including the slit-lamp CCD scanning angle, patient identification, and image resolution, to ensure data readability and consistency across different medical systems.

[0080] Simultaneously, based on the residual mean and standard deviation in the error distribution image, abnormal regions are identified and abnormal alerts are generated. Specifically, the identification method is as follows: a threshold of 0.05 mm is set for the residual mean, and a threshold of 0.02 mm is set for the standard deviation; when the residual mean of a pixel exceeds 0.05 mm or the standard deviation exceeds 0.02 mm, the area containing that pixel is marked as an abnormal region. The abnormal alert is generated in text format, containing the coordinate range of the abnormal region and the specific residual mean or standard deviation value, and is stored in the private tag field of the DICOM file. Furthermore, the thickness raster image, error distribution image, and abnormal alert are further integrated into the EyeCare extended data structure of the rapid medical interoperability resource, stored in JSON format, to ensure semantic consistency and cross-platform traceability of the data.

[0081] When used, it is encapsulated using digital imaging and communication medical standard formats and the EyeCare extended data structure, a rapid medical interoperability resource, ensuring the standardization and interoperability of thickness raster images and error distribution images. This facilitates data transmission and use across different medical devices and systems. The generation and storage of anomaly alerts provide clinicians with intuitive feedback on abnormal areas, enabling them to quickly locate potential problems in the data, thereby improving the accuracy and efficiency of data analysis.

[0082] Step 502: Unified Interface Push After data encapsulation is completed, the encapsulated data is pushed to the surgical navigation platform through a unified interface.

[0083] The specific processing steps are as follows: First, the thickness raster image, error distribution image, and anomaly warning information in the standard format of digital imaging and communication medicine are converted into a standard radiotherapy object, namely a DICOM-RT object. This conversion enables the data to be recognized by the surgical navigation platform and used for the visualization, rendering, and analysis of thickness data. Next, using the Health Level 7 protocol, the DICOM-RT object is transmitted to the surgical navigation platform in encrypted form via a message queue mechanism, ensuring the security and real-time performance of data transmission. On the surgical navigation platform, an application programming interface (API) based on a representational state transition architecture, namely a RESTful API, is developed to receive and parse the DICOM-RT object. After parsing, the surgical navigation platform renders the data into a visualization interface, allowing clinicians to view and manipulate the thickness raster image and error distribution image in real time.

[0084] The Health Level 7 Protocol (HL7) is an international standard for the exchange, integration, sharing, and retrieval of electronic health information between healthcare information systems. Developed and maintained by Health Level Seven International (HL7 International), it aims to address the difficulties in data exchange between different systems in the healthcare field, promoting system interoperability and the efficiency of healthcare services.

[0085] In practice, by converting data into digital imaging and communication medical standard radiotherapy objects and transmitting it in conjunction with the Health Level 7 protocol, standardized data delivery and efficient integration are achieved, ensuring that the surgical navigation platform can acquire thickness raster images and error distribution images in real time. The development of the representational state transition architecture application programming interface simplifies the data reception and parsing process, supporting clinicians to quickly access and analyze data, thereby improving the efficiency and convenience of clinical workflows.

[0086] Step 503: Confidence Assessment and Feedback To ensure the reliability of the thickness raster image and the error distribution image, a confidence assessment mechanism is introduced here, and feedback is provided based on the assessment results. Specifically, the assessment process calculates the overall confidence level based on the error distribution image. The confidence level is defined as the proportion of pixels in the error distribution image with a residual mean value less than 0.05 mm to the total number of pixels in the thickness raster image.

[0087] The calculation process consists of the following steps: First, all pixels in the error distribution image are traversed, and the number of pixels with a residual mean value below 0.05 mm is counted. Then, the total number of pixels in the thickness raster image is obtained. Finally, the number of pixels with a residual mean value below 0.05 mm is divided by the total number of pixels in the thickness raster image to obtain the confidence score. To ensure data quality, a confidence score threshold of 0.9 is set. When the calculated confidence score is below 0.9, the surgical navigation platform's user interface displays a warning message, prompting clinicians that "the confidence score of the thickness data is lower than expected; it is recommended to repeat steps two through four to obtain more reliable data," and records the specific confidence score value. Simultaneously, the confidence score value, the anomaly warning message, and the feedback record are stored in the audit log module of the EyeCare extension of the rapid medical interoperability resource, marked with timestamps for subsequent quality review and process optimization.

[0088] In practice, the confidence assessment mechanism provides an intuitive and repeatable reliability metric by quantifying the proportion of pixels in the error distribution image with a residual mean below 0.05 mm, ensuring that the quality of the thickness raster images meets clinical application requirements. Setting a residual mean threshold of 0.05 mm reflects the high precision required for corneal epithelial thickness measurement, while a confidence threshold of 0.9 ensures that the overall data error is controlled within an acceptable range. An automatic feedback mechanism promptly prompts for data re-acquisition when confidence is insufficient, avoiding clinical misjudgments due to data quality issues. The storage of audit logs enhances data transparency and traceability, supporting quality control and process improvement, thereby improving the reliability and security of the entire measurement process.

[0089] In practice, thickness raster images, error distribution images, and anomaly alerts are encapsulated into objects based on digital imaging and communication medical standard formats and the EyeCare extended data structure, a rapid medical interoperability resource. Seamless integration with surgical navigation platforms is achieved using the Health Level 7 protocol and a representational state transition architecture application programming interface, ensuring standardized transmission and real-time visualization of corneal epithelial thickness data. A confidence assessment mechanism based on error distribution images effectively identifies low-reliability data by calculating the proportion of pixels with a residual mean below 0.05 mm and comparing it to a threshold of 0.9, prompting for re-acquisition when necessary. This comprehensive processing method not only improves the clinical usability of thickness raster images and error distribution images but also provides technical assurance for the accuracy of refractive surgery and early lesion screening through standardized interfaces and reliability assessment mechanisms.

[0090] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0091] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0092] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0093] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0094] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for measuring preoperative corneal epithelial thickness based on slit-lamp CCD three-dimensional reconstruction, characterized in that: include, Based on the extracted labels, an initial epithelial thickness template matching the patient's ocular features is selected from a pre-built template library. The corneal center is located on the slit-lamp CCD imaging interface through edge detection and Hough transform, and the selected template is aligned with the corneal center. By modifying a slit-lamp CCD device, multi-angle synchronous scanning was performed to monitor tear film stability and record the reflectance spectrum; After generating epithelial point cloud data using a stereo vision algorithm, the point cloud is spatially aligned with the template. After calculating the residual elastic relaxation index and curvature gradual stability index based on the residual field and curvature gradient, the individualized fitting confidence coefficient is output using the Bayesian generalized linear deformation confidence model. Adaptive fitting is completed through dynamic coupling of elastic deformation and Kalman filtering to generate an individualized epithelial surface. Curvature continuity filtering and optical reprojection correction are performed between the fitted surface and the original point cloud to generate a high-confidence thickness raster, and an error distribution label is attached to each pixel. After encapsulating the thickness grid, error labels, and anomaly alerts, the data is pushed to the surgical navigation platform, and a prompt to re-collect data is automatically generated when the confidence level is low.

2. The method for measuring preoperative corneal epithelial thickness according to claim 1, characterized in that: Medical history data, including refractive surgery records and pathology labels, is extracted from electronic medical records to identify patient-specific corneal features; based on the extracted medical history data, an initial epithelial thickness template matching the patient's corneal features is selected from a pre-built template library. The corneal center was located using image processing techniques on slit-lamp CCD imaging, and the selected initial epithelial thickness template was aligned with the located corneal center.

3. The method for measuring preoperative corneal epithelial thickness according to claim 2, characterized in that: The scanning path is planned using the geometric features of a personalized initial template of epithelial thickness to ensure that the scanning angle and position match the actual morphology of the patient's cornea; Multi-angle synchronous acquisition is performed by time synchronization triggering, tear film stability monitoring and reflectance intensity spectrum recording to obtain high-quality image data at the same time baseline.

4. The method for measuring preoperative corneal epithelial thickness according to claim 3, characterized in that: Epithelial point cloud data is generated using stereo vision algorithms and voxel filtering techniques to ensure the accuracy and reliability of the data; By using coarse and fine alignment, the epithelial point cloud data is spatially aligned with the personalized initial template for epithelial thickness, thereby improving the accuracy of subsequent fitting and thickness measurement.

5. The method for measuring preoperative corneal epithelial thickness according to claim 4, characterized in that: After calculating the residual fields of the epithelial point cloud data and the personalized initial epithelial thickness template, the local deviation is determined by quantifying the Euclidean distance between the two, and the curvature gradient of the epithelial point cloud data and the personalized initial epithelial thickness template is analyzed to identify areas of abrupt changes in surface morphology.

6. The method for measuring preoperative corneal epithelial thickness according to claim 5, characterized in that: The residual elastic relaxation index and curvature gradient stability index are calculated based on the residual field and curvature gradient to dynamically control the fitting strength and stability. The reliability of the fit is evaluated using a Bayesian generalized linear deformation confidence model, and individualized fit confidence coefficients are output. The data and smoothing terms in the energy function are optimized through the dynamic coupling of elastic deformation and Kalman filtering, and adaptive fitting is achieved through the state update formula until the residual mean converges to a preset threshold, generating an individualized epithelial surface.

7. The method for measuring preoperative corneal epithelial thickness according to claim 6, characterized in that: The curvature distribution of the individualized epithelial surface is calculated to identify regions of abnormal curvature. The Laplacian smoothing algorithm is applied to adjust the vertex positions of the surface to ensure a smooth surface shape. The original epithelial point cloud data is projected onto the image plane using camera intrinsic and extrinsic parameters.

8. The method for measuring preoperative corneal epithelial thickness according to claim 7, characterized in that: The rotation matrix and translation vector of the individualized epithelial surface are optimized by calculating the reprojection error to improve spatial consistency by generating a base surface below the individualized epithelial surface. The shortest distance from a point on the surface to the base surface is calculated to generate a two-dimensional thickness raster image. The fitting residuals are calculated, and the mean and standard deviation of the residuals within the pixel coverage area of ​​the thickness raster image are statistically analyzed to generate an error distribution image to evaluate the measurement reliability.

9. The method for measuring preoperative corneal epithelial thickness according to claim 8, characterized in that: Thick raster images, error distribution images, and anomaly alerts are encapsulated into objects of the EyeCare extended data structure, a digital imaging and communications medical standard format and a rapid medical interoperability resource, and pushed to the surgical navigation platform via the Health Level 7 protocol and the Representational State Transition Architecture application programming interface.

10. The method for measuring preoperative corneal epithelial thickness according to claim 9, characterized in that: The confidence level is calculated based on the proportion of pixels with a residual mean of less than 0.05 mm from the error distribution image. When the confidence level is below 0.9, a warning message is displayed through the user interface, and the confidence level value, abnormal message, and feedback record are stored in the audit log.

Images