A method and device for measuring corneal topography parameters from anterior segment OCT images

Abstract

A method and device for measuring corneal topography parameters of anterior segment OCT images, the method comprising: pre-processing the OCT image, then detecting the scleral protrusion feature points in the image, establishing the corneal central axis, and providing a reference for image registration. Next, the cornea and iris are segmented to generate a binary image, and the corneal contour is extracted and optimized to ensure high precision. Combining the scleral protrusion feature points and the optimized corneal contour, the corneal boundary is determined to provide spatial positioning for three-dimensional reconstruction and parameter calculation. By fitting the corneal surface, a mathematical model is established, and corneal topography parameters such as principal curvature and astigmatism are calculated. Finally, the parameters are integrated to generate a visual corneal topography map and report to assist in clinical diagnosis and surgical planning. This method ensures the objectivity, efficiency, accuracy, and repeatability of the measurement, providing strong data support for clinical diagnosis and surgical planning.

Description

Technical Field

[0001] The present invention relates to ophthalmic measurement, and in particular to a method and device for measuring corneal topography parameters of anterior segment OCT images. Background Art

[0002] Keratoconus (KC) is a blinding eye disease with significant risks. As keratoconus progresses, the cornea gradually thins and protrudes forward, forming an irregular cone shape. This leads to blurred vision, distorted vision, and multiple images, severely impacting patients' quality of life and visual function. Patients often experience symptoms such as light sensitivity, poor night vision, and visual fatigue. In severe cases, it can even lead to corneal rupture and acute blindness.

[0003] Corneal elasticity and stiffness play a crucial role in maintaining corneal morphological stability. Studies have found that the activity of matrix metalloproteinases (MMPs) is significantly increased in the corneas of patients with keratoconus. These enzymes degrade collagen fibrils and proteoglycans in the corneal stroma, leading to structural loosening and decreased strength of the corneal tissue. Extracellular matrix components, including collagen fibrils and proteoglycans, undergo abnormal reorganization in patients with keratoconus. This abnormal reorganization leads to decreased strength and elasticity of the corneal tissue, making the cornea more susceptible to deformation. Corneal stromal cells (corneal fibroblasts) exhibit increased apoptosis and dysfunction in patients with keratoconus. This loss of cell function further weakens the structural integrity and biomechanical properties of the corneal stroma. Due to the reduced corneal elasticity and stiffness, the cornea of ​​patients with keratoconus gradually convexes under the influence of intraocular pressure, forming a cone-shaped shape, leading to significant visual impairment. Patients often experience symptoms such as blurred vision, light sensitivity, poor night vision, and visual fatigue.

[0004] Keratoconus can be divided into early, middle, and late stages based on disease severity and progression. Early keratoconus presents with mild blurred vision and increased myopia; in the middle stage, corneal morphology changes significantly, leading to a significant decrease in vision; and in the late stage, vision is severely impaired, potentially requiring a corneal transplant. The pathogenesis of keratoconus is not fully understood, but it is generally believed to be a combination of genetic and environmental factors.

[0005] KC has a long course and insidious onset, making it difficult to detect in the early stages of the disease. When vision decreases significantly, it usually indicates that the disease has entered the middle or late stages. In these stages, the thinning and deformation of the cornea become significant, leading to severe blurred vision, distortion and other vision problems. If KC can be diagnosed in its early stages and surgical medical intervention is performed in a timely manner, the progression of the disease can be effectively prevented. The main characteristics of KC are the gradual thinning of the central or lower central part of the cornea, forming a cone shape, and the gradual aggravation of myopia and irregular astigmatism. Clinical evaluation of corneal thickness (Pachymetry) and corneal refractive power (Keratometric) is crucial for the early assessment and diagnosis of KC.

[0006] Currently, the gold standard for assessing keratoconus is primarily corneal topography. Corneal topography is categorized based on its imaging principle into corneal reflectance topography, anterior corneal scanning topography, and posterior corneal scanning topography. Corneal reflectance topography measures the anterior corneal curvature using concentric circular reflective patterns, such as with the Pentacam and Orbscan devices. Anterior corneal scanning topography utilizes a Scheimpflug camera to scan the anterior and posterior corneal surfaces at multiple angles, providing detailed three-dimensional images, such as with the Pentacam and Galilei devices. Posterior corneal scanning topography utilizes optical coherence tomography (OCT) technology to obtain precise images of the anterior and posterior corneal surfaces. Commonly used devices include the Visante OCT and RTVue OCT. Commonly used clinical assessment and classification methods include the Amsler-Krumeich classification and the Belin / Ambrosio enhanced display (BAD) system, which categorize corneal thickness and curvature based on parameters such as corneal thickness and curvature. Although corneal topography is the gold standard for keratoconus assessment, the accuracy of its results depends on the precision of the equipment and the professional skills of the operator. There may be differences between different equipment and operators, and a comprehensive analysis combining multiple methods and clinical experience is required to obtain accurate diagnostic results.

[0007] The Placido ring method is a corneal topography technique commonly used in ophthalmology. Its imaging principle involves projecting concentric ring patterns onto the corneal surface using a Placido ring. The cornea reflects these patterns, which are captured by a camera and converted into electrical signals. These signals are then converted into a three-dimensional image of the anterior corneal surface based on a specific algorithm. Unlike other low-frequency techniques, the Placido ring method utilizes optical reflection principles, offering high resolution and precision. The Pentacam device utilizes this technology, acquiring multi-angle corneal images using a rotating Scheimpflug camera. This method not only provides detailed images of the anterior corneal curvature but also measures corneal thickness and anterior chamber depth. This non-contact imaging method requires no direct contact with the eyeball, allowing the patient to complete the examination while seated. Due to its ease of use and high image quality, the Placido ring method is widely used in the diagnosis and preoperative evaluation of keratoconus. However, the Placido ring method cannot provide accurate topography of the posterior corneal surface and cannot accurately capture subtle changes in the posterior corneal surface. The accuracy of the imaging results depends on the calibration of the equipment and the professional skills of the operator. Improper operation or equipment failure may affect the accuracy and repeatability of the measurement. Therefore, when performing a Placido ring examination, the examiner needs to have certain professional knowledge and operating experience to ensure the reliability of the examination results.

[0008] OCT corneal topography processes the diffusely reflected light that returns along the same path as the incident light, and visualizes it based on the intensity of its light signal and the transmission time, thereby achieving non-contact, non-invasive tomographic imaging of living corneal tissue and assisting in the diagnosis of corneal diseases. Compared with the traditional Placido ring method, OCT corneal topography can perform a clearer and more accurate quantitative analysis of the anterior and posterior surfaces of the cornea. OCT corneal topography can not only provide a curvature map of the anterior corneal surface, but also obtain detailed images of the corneal thickness and posterior surface. Using OCT corneal topography, the corneal thickness (Central Corneal Thickness, CCT), the curvature of the anterior and posterior corneal surfaces, and the corneal morphology can be quantitatively measured, enabling a more comprehensive and objective evaluation of corneal-related diseases.

[0009] OCT corneal topography enables fast, non-contact imaging without causing discomfort to the patient and making the operation easier for the examiner. Furthermore, the three-dimensional tomographic images and precise quantitative data provided by OCT corneal topography give it significant advantages in diagnosis and preoperative evaluation. However, due to the wide variety of parameters in OCT corneal topography, manual measurement alone cannot guarantee detection efficiency and is significantly affected by subjective factors such as the operator's clinical experience.

[0010] The diagnosis of early keratoconus requires a wide variety of anterior segment parameters and corneal morphological parameters. Manual measurement is not only inefficient, but the evaluation results are also affected by subjective factors such as the operator's clinical experience. In addition, the changes in corneal morphology caused by early keratoconus are almost impossible to observe with the naked eye.

[0011] It should be noted that the information disclosed in the above background technology section is only used to understand the background of this application, and therefore may include information that does not constitute prior art known to ordinary technicians in this field. Summary of the Invention

[0012] The main purpose of the present invention is to overcome the defects of the above-mentioned background technology and provide a method and device for measuring corneal topography parameters of anterior segment OCT images.

[0013] To achieve the above object, the present invention adopts the following technical solutions:

[0014] A method for measuring corneal topography parameters from anterior segment OCT images comprises the following steps:

[0015] S1. Preprocessing the collected anterior segment OCT images;

[0016] S2. Use deep learning algorithms to detect scleral protrusion feature points in OCT images and determine the corneal central axis to provide a reference for image registration;

[0017] S3. Use the trained semantic segmentation model to process the OCT image, segment the cornea and iris, and generate a binary image.

[0018] S4. extracting the corneal contour based on the segmented image;

[0019] S5. Determine the corneal boundary by combining the scleral protrusion feature points and the corneal contour;

[0020] S6. Fitting the corneal surface to establish a mathematical model of the corneal morphology; wherein the corneal central axis determined by the scleral protrusion feature points is used as a symmetry reference in the fitting process, and the corneal contour data is used as the corneal boundary of the fitting model to perform an overall fitting of the corneal surface;

[0021] S7. calculating corneal topographic parameters based on the fitted mathematical model of corneal morphology;

[0022] S8. Integrate the calculated corneal topography parameters to generate a visual corneal topography map and parameter report.

[0023] Furthermore, in step S2, detecting the scleral protrusion feature points in the OCT image specifically includes the following sub-steps:

[0024] S21. Using a direct regression algorithm and computer vision technology, locate the angle region and crop a rectangular region containing the key points of the scleral protrusion.

[0025] S22. Apply the deep learning ResNet50 model to perform key point detection on the cropped area and output the predicted coordinate positions of the scleral protrusion key points; the ResNet50 model is trained using the scleral protrusion key point dataset, and the model's loss function uses the mean square error (MSE) loss function.

[0026] Furthermore, in step S2, through scleral protrusion key point detection, the key points identified on both sides of the chamber angle are connected with a straight line and the image is rotated to a straight horizontal level, and then the image is translated along the y-axis until the corneal vertex reaches a preset uniform height. After batch processing a group of OCT images, the alignment is completed.

[0027] Furthermore, in step S3, the trained U2net is used as a semantic segmentation model to perform semantic segmentation on the preprocessed OCT image, separate different tissues of the anterior segment such as the cornea and iris, and generate a binary image; wherein BCELoss is used as the training loss function;

[0028] In step S4, based on the binary image, an edge detection operator is applied to extract the precise contours of the cornea and iris; the extracted contours are optimized to remove noise and discontinuous edges, thereby achieving smoothness and precision of the contours.

[0029] Furthermore, step S6 specifically includes:

[0030] Based on steps S3 and S4, the corneal image is further semantically segmented to capture the corneal section morphology and record the image position;

[0031] Based on the corneal contour extracted in step S4, the coordinates of the corneal closure boundary points are recorded using an edge detection function;

[0032] Using the corneal boundary determined in step S5 and the corneal contour optimized in step S4, the two-dimensional coordinates of the corneal edge are converted into three-dimensional coordinates of the entire cornea through a two-dimensional-three-dimensional coordinate conversion matrix to construct a more accurate three-dimensional corneal boundary point cloud model;

[0033] The obtained corneal three-dimensional point cloud model was fitted with Zernike polynomials to obtain the edge equations of the upper and lower surfaces of the cornea, providing smooth surface data for subsequent topographic analysis.

[0034] Furthermore, in step S7, the key parameters for calculating corneal topography specifically include one or more of the following parameters:

[0035] Calculate the average corneal curvature by calculating the gradient of the corneal surface height data in the x and y directions, obtain the curvature field, and take the average value to obtain the average curvature;

[0036] Calculate the main curvature and direction of the cornea, and obtain the maximum and minimum curvatures and their directions of the corneal surface through the second-order derivative analysis of the curvature field;

[0037] Measure the central corneal thickness by analyzing the value of the corneal surface height data in the central area to obtain the central thickness;

[0038] Calculate corneal astigmatism, which is obtained by calculating the inverse of the curvature field, reflecting the curvature of the corneal surface;

[0039] The corneal dysmorphism index was calculated to assess the overall corneal morphology and variability by the ratio of the anterior to posterior surface curvatures.

[0040] Furthermore, in step S8, based on the extracted corneal distribution information, a specific colorbar mapping method is used to convert this information into a visual corneal topography map, so as to intuitively display the ups and downs of the corneal topography and the refractive state.

[0041] A device for measuring corneal topography parameters of anterior segment OCT images for executing the method, comprising:

[0042] An image preprocessing unit, used for preprocessing the collected OCT image of the anterior segment of the eye;

[0043] a deep learning processing unit, configured to run a deep learning algorithm to detect scleral protrusion feature points in the OCT image and determine the corneal central axis;

[0044] The semantic segmentation unit is used to process the OCT image using the trained semantic segmentation model to segment the cornea and iris and generate a binary image;

[0045] A contour extraction unit, configured to extract a corneal contour based on the segmented image;

[0046] Corneal boundary determination unit, used to determine the corneal boundary by combining the scleral protrusion feature points and the optimized corneal contour, and provide spatial positioning for three-dimensional reconstruction and parameter calculation;

[0047] Corneal morphology modeling unit, used to fit the corneal surface and establish a mathematical model of corneal morphology;

[0048] A parameter calculation unit, used for calculating corneal topography parameters according to the fitted corneal morphology mathematical model;

[0049] The result integration and visualization unit is used to integrate the calculated corneal topography parameters and generate a visual corneal topography map and parameter report.

[0050] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the method for measuring corneal topography parameters of anterior segment OCT images.

[0051] A computer program product includes a computer program, which implements the method for measuring corneal topography parameters of anterior segment OCT images when executed by a processor.

[0052] The present invention has the following beneficial effects:

[0053] The present invention proposes an automated method and device for measuring corneal topography parameters in anterior segment OCT images, achieving efficient and accurate analysis of corneal topography. The present invention utilizes a deep learning algorithm to identify scleral protrusion feature points in OCT images, and accordingly determines the corneal central axis, providing a benchmark for image registration. Furthermore, a well-trained semantic segmentation model is used to segment the cornea and iris to produce a clear binary image. On this basis, the corneal boundary is established by extracting and optimizing the corneal contour, providing accurate spatial positioning for three-dimensional reconstruction and parameter calculation. The present invention combines scleral protrusion feature points with precise corneal contour data to determine the corneal boundary, and establishes a mathematical model by fitting the corneal surface. By finely fitting the corneal surface, smooth curved surface data is generated. By calculating and integrating key topographic distribution information parameters of the cornea, a visual corneal topography map and a detailed parameter report are generated, providing strong data support for clinical diagnosis and surgical planning.

[0054] This invention, through the construction of an automated measurement algorithm, ensures the objectivity, efficiency, accuracy, and repeatability of corneal topography parameter measurements, thereby assisting doctors in diagnosing corneal diseases more quickly and accurately. This method significantly improves detection speed and system response efficiency, ensuring real-time and efficient image processing and parameter detection, and has significant application value.

[0055] Other beneficial effects of the embodiments of the present invention will be further described below. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 4 is a flow chart of an automatic measurement method according to an embodiment of the present invention.

[0057] Figure 2 It is a schematic diagram of the hardware module construction of an embodiment of the present invention.

[0058] Figure 3 This is a structural diagram of the Resnet50 model according to an embodiment of the present invention.

[0059] Figure 4AThis is a flowchart of the implementation of the scleral protrusion key point detection and image registration module in an embodiment of the present invention.

[0060] Figure 4B 2. It is a diagram of the recognition and application of scleral protrusion feature points in an OCT image according to an embodiment of the present invention.

[0061] Figure 5 4 is a structural diagram of the U2net model of the contour recognition module according to an embodiment of the present invention.

[0062] Figure 6 It is a logic diagram of the automatic measurement method executed in an embodiment of the present invention.

[0063] Figure 7 This is a logic diagram of the interaction between an embodiment of the present invention and an OCT device.

[0064] Figure 8 This is an example of an OCT image input in an embodiment of the present invention.

[0065] Figure 9 This is the result returned after processing by the embodiment of the present invention. The corneal topography height map is the visualization effect measured by the embodiment of the present invention, wherein parameters such as Ks and Kf are not visualized but directly calculated and the calculation results are returned. DETAILED DESCRIPTION

[0066] The following is a detailed description of the embodiments of the present invention. It should be emphasized that the following description is only exemplary and is not intended to limit the scope of the present invention and its application.

[0067] See Figure 1 The present invention provides a method for measuring corneal topography parameters of anterior segment OCT images, comprising the following steps:

[0068] S1. Preprocess the collected anterior segment OCT images, including normalizing them, to improve the robustness and classification accuracy of the model.

[0069] S2. Use deep learning algorithms to detect scleral protrusion feature points in OCT images and determine the corneal central axis to provide a reference for image registration;

[0070] S3. Use the trained semantic segmentation model to process the OCT image, segment the cornea and iris, and generate a binary image.

[0071] S4. Based on the segmented image, extract the corneal contour and perform optimization processing to remove noise and discontinuous edges to ensure high contour accuracy;

[0072] S5. Combine the scleral protrusion feature points and the optimized corneal contour to determine the corneal boundary, providing accurate spatial positioning for 3D reconstruction and parameter calculation;

[0073] S6. Fitting the corneal surface to establish a mathematical model of the corneal morphology; wherein the corneal central axis determined by the scleral protrusion feature points is used as a symmetry reference in the fitting process; the optimized corneal contour data is used as a boundary condition of the fitting model; and the corneal surface is fitted as a whole based on the determined corneal boundary to achieve a comprehensive description of the corneal morphology;

[0074] S7. Calculate corneal topographic parameters, such as corneal principal curvature, astigmatism, mean curvature, corneal vertex angle, and mean corneal curvature power, based on the fitted mathematical model of corneal morphology;

[0075] S8. Integrate the calculated corneal topography parameters to generate a visual corneal topography map and parameter report for clinical diagnosis and / or surgical planning.

[0076] The present invention provides an automated method for measuring corneal topography parameters of anterior segment OCT images, which realizes efficient and accurate analysis of corneal topography. In the present invention, a deep learning algorithm is used to detect the scleral protrusion feature points in the image, establish the corneal central axis, and provide an accurate benchmark for image registration. Then, a semantic segmentation model is applied to effectively segment the cornea and iris to generate a high-quality binary image. On this basis, the corneal contour is further extracted and optimized, laying a solid foundation for determining the corneal boundary and performing three-dimensional reconstruction. The present invention combines the scleral protrusion feature points and accurate corneal contour data to determine the corneal boundary, providing accurate spatial positioning for three-dimensional reconstruction and parameter calculation of corneal topography. A mathematical model is established by fitting the corneal surface, and preferably Zernike polynomials are used to finely fit the upper and lower surfaces of the cornea to generate smooth curved surface data, and then the key topographic distribution information such as corneal height, refractive power, and curvature are calculated. Finally, the parameters obtained by these calculations are integrated to not only generate a visual corneal topography map, but also output a detailed parameter report, providing strong data support for clinical diagnosis and surgical planning.

[0077] By constructing an automatic measurement algorithm for OCT corneal topography, the present invention can achieve objective quantitative analysis of corneal topography parameters in anterior segment OCT images, ensure the efficiency, accuracy and repeatability of the measurement, and assist doctors in diagnosing corneal diseases more quickly and accurately.

[0078] Specific embodiments of the present invention are further described below.

[0079] The automated method for measuring corneal topographic parameters from OCT images consists of several components: a signal acquisition module, a corneal image semantic segmentation module, a scleral protrusion keypoint detection module, and a corneal topography visualization and parameter calculation module. The device focuses on the entire 360° circumference of the eye and repeatedly captures fixed-wavelength light signals diffusely reflected from the eye. These signals are processed by the signal acquisition module and recorded by a computer as a set of 128 binary grayscale OCT images. The input OCT images undergo image registration preprocessing. The registration process involves detecting scleral protrusion keypoints. Keypoints identified on both sides of the angle are connected by straight lines and the image is rotated to a horizontal position. The image is then translated along the y-axis until the corneal vertex reaches a preset uniform height. Registration is then completed after batch processing of the OCT images. The registered corneal OCT images undergo corneal image semantic segmentation, capturing the corneal cross-section morphology and recording the image positions. An edge detection function then records the coordinates of the corneal closure boundary points. After successfully extracting the corneal edge points, we leveraged the image-encoded position information and converted the 2D corneal edge into 3D coordinates for the entire cornea using a 2D-3D coordinate conversion matrix, resulting in a point cloud model of the 3D corneal boundary. Eight Zernike polynomials were fitted to the upper and lower corneal surface point clouds to infer the equations for the upper and lower corneal edges, providing smooth surface data for subsequent topographic analysis. Processing the fitted surface data revealed distribution information for corneal height, refractive power, and curvature, encompassing both subtle variations and overall morphological features of the corneal surface. A specific colorbar mapping method was then used to generate a visual corneal topography map. The corneal height distribution polynomial, corneal curvature distribution polynomial, corneal refractive power distribution polynomial, and corneal contour information served as basic parameters for input into a corneal topography parameter calculation program. Different corneal topography parameter calculation results were then output based on different parameter measurement modes.

[0080] OCT signal acquisition

[0081] The hardware solution of the OCT signal acquisition module includes:

[0082] (1) Light source:

[0083] Santec's HSL-1 tunable VCSEL light source was selected, which features a 1060nm wavelength, high-speed scanning, long coherence length, and low mode noise, making it suitable for imaging ophthalmic and high-water-content samples.

[0084] (2) Detector:

[0085] Based on the characteristics of the light source, THORLABS's PDB482C-AC photoelectric balanced detector was selected. It is specifically designed for SS-OCT. It has an operating wavelength range of 900-1400nm, a bandwidth of 1MHz-2.5GHz, good common-mode rejection, and a transconductance gain of up to 28x 10^3V / A. It is suitable for high-speed signal acquisition and has stability in complex environments.

[0086] (3) Signal acquisition card:

[0087] The Acqiris SA220P high-speed data acquisition card features 14-bit sampling accuracy, a 2GS / s sampling rate, 1.2GHz analog bandwidth, 50Ω input impedance, and DC coupling. It comes with 4GB or 8GB of storage, a PCIe interface, onboard FPGA real-time signal processing, hardware accumulation, and synchronized data output.

[0088] (4) Optical path design:

[0089] The detection light emitted by the light source is divided into detection light and sample light in the reference arm and sample arm. The light path of each arm passes through components such as a fiber optic circulator, a polarizer, and a collimator to ensure the collimation of the light path and the accuracy of the reflection. The photodetector realizes high-quality collection of the interference light signal.

[0090] (5) Hardware optimization:

[0091] Wavelength Optimization: The wavelength of the optimal scanning light is screened to perform best at 1060nm (within the range of 900-1400nm) to maximize the quality and resolution of the anterior segment image.

[0092] Detector type: InGaAs / PIN type detector is used to ensure high sensitivity near the wavelength of 1060nm.

[0093] Input interface selection: Optical input uses FC / APC interface, standard 50Ω input impedance and DC coupling, which is easy to connect with common devices.

[0094] Passband width and sampling rate: Ensure that the passband width of the device is within the range of 1MHz to 2.5GHz, and the maximum sampling rate reaches 2GS / s to ensure accurate acquisition of high-speed signals.

[0095] (6) Collection parameter configuration:

[0096] Scan parameters: Select an acquisition size of 4096 to obtain high-resolution details of the eye structure. Use a Decimation Ratio of 1 to retain all sampling points to ensure that signal details are not lost.

[0097] K-clock parameters: The delays of the main scan and the secondary scan were set to -1500 and 0 to ensure the synchronization of the scanning process, and the Hilbert Gain was set to 0.

[0098] Image parameters: A-scan Size was set to 4096, consistent with the A-scan acquisition size, and B-scan Size was set to 100 to ensure that each B-scan consisted of enough A-scans to obtain complete information about the eye structure.

[0099] The design of this swept-frequency OCT hardware system enables efficient and accurate imaging of ocular structures. The optimized selection of hardware components ensures the efficiency and stability of optical signal transmission, reflection, interference, and conversion, providing reliable data support for in-depth research on anterior segment structures. This system design not only ensures high-quality signal acquisition but also lays a solid foundation for subsequent image processing.

[0100] Scleral protrusion key point detection

[0101] The key points of the sclera image are special anchor points that can be observed in all-peripheral cross-sectional images of the eyeball. Their positions are highly stable. There are two sclera protrusion key points with axisymmetric characteristics in an anterior segment OCT image. Therefore, we can restore the central axis of the eyeball by locating the sclera protrusion feature points in each anterior segment OCT image. The sclera protrusion serves as the reference point for determining the central axis of the cornea. The accuracy of its positioning will determine the accuracy of subsequent measurements. Sclera protrusion localization is a key point detection task, mainly used in scenarios such as facial key point detection and human key point detection. Compared with classification and object detection tasks, key point detection has higher requirements for spatial accuracy.

[0102] There are currently two implementation methods for the mainstream algorithms of key point detection: one is to use a fully connected layer to directly regress the coordinates of the feature points; the other is to generate a Gaussian Heatmap through the network, and then use the argmax function or the maximum likelihood (MLE) function to calculate the coordinates of the key points. The advantage of the first method is that it can achieve end-to-end training and a faster inference speed. The second method constructs a Gaussian Heatmap as an intermediate state, which can achieve high-precision key point prediction while ensuring the stability of model training. The direct regression method provides less supervision information than the Gaussian Heatmap method and is more difficult to train. The network needs to convert the spatial position into coordinates by itself, and its spatial generalization ability is weak. The preferred embodiment uses an algorithm based on the direct regression method in the detection of the scleral protrusion key point, locates the chamber angle area through the CV method, and cuts out a small part of the rectangular chamber angle area containing the scleral protrusion key point for direct regression. By reducing the target area, the generalization of the model is improved and the training difficulty is reduced, which solves the shortcomings of the direct regression method.

[0103] The preferred embodiment uses ResNet50 as the keypoint detection model. ResNet50 is a deep convolutional neural network with excellent feature extraction capabilities and high accuracy. The ResNet50 network structure includes multiple residual blocks, which use skip connections to achieve efficient information transfer, avoid the vanishing gradient problem, and enable the model to better learn high-level features in the image.

[0104] The ResNet50 network is trained using the scleral protrusion key point dataset. The training process of the model is as follows Figure 4A As shown in Figure 2, during the training process, the coordinate labels of the scleral protrusions are directly input into the network, and the loss function of the model is the mean squared error (MSE), which is defined as:

[0105]

[0106] Among them, xi represents the model predicted value and yi represents the true value.

[0107] The model's inference process is similar to the training process. After the input OCT image is processed by the ResNet50 network, it is no longer compared with the true label. Instead, the predicted coordinate position is directly output to achieve the localization of the scleral protrusion key points.

[0108] The Euclidean distance represents the straight-line distance between the predicted coordinate point and the true label coordinate point. The average Euclidean distance error of the trained scleral protrusion key point detection model on the validation set is 5.6 pixels, which is about 41.98 μm.

[0109] Corneal contour recognition

[0110] High-quality binary images of the cornea and iris are crucial for accurately capturing the corneal contour. To achieve this, the preferred embodiment utilizes deep learning techniques, specifically semantic segmentation models such as FCN, U-Net, and Attention U-Net. These models have been widely used in the medical imaging field and have achieved remarkable results on small datasets. Semantic segmentation allows for classification of each pixel in the OCT image, achieving fine-grained pixel-level segmentation.

[0111] In the preferred embodiment of the present invention, U2net is selected as the semantic segmentation model. U2net is a saliency detection model proposed on the basis of Unet, and its segmentation effect on small-sized feature maps is very good. The U2net network adopts a two-level nested U structure, in which RSU (Residual U-blocks) replaces the convolutional blocks in Unet, combines receptive fields of different sizes, and can capture more semantic information from different scales. U2net is trained using the OCT segmentation dataset. The network training process is similar to Figure 4A Similarly, the input is replaced by the global OCT image, and the true label is replaced by the segmentation label of the anterior segment. BCELoss is selected as the loss function for training, which is defined as follows:

[0112]

[0113] The model was trained for 300 epochs, using Intersection-over-Union (IoU) as the evaluation metric for OCT image segmentation results. The segmentation model achieved an average IoU of 94.7% for cornea and 99.0% for background segmentation on the validation set.

[0114] Implementation steps:

[0115] (1) Data preprocessing: OCT images are standardized to improve the robustness and classification accuracy of the model.

[0116] (2) Application of semantic segmentation model: The pre-processed OCT images are semantically segmented using a trained U-Net model. This model can effectively separate different tissues in the anterior segment of the eye, such as the cornea and iris, and generate high-quality binary images.

[0117] (3) Edge detection: The generated binary image is processed using an edge detection operator to extract the precise contours of tissues such as the cornea and iris. This step ensures high accuracy of contour recognition and provides a reliable data basis for further image analysis.

[0118] (4) Contour optimization: Optimize the initially extracted contour to remove noise and discontinuous edges, making the contour smoother and more accurate. The accuracy of contour recognition can be further improved through the optimization algorithm.

[0119] (5) Result output: The optimized contour data is output to provide support for subsequent 3D reconstruction and parameter calculation. This data can be used for various applications such as corneal morphology analysis, surgical planning, and disease diagnosis.

[0120] Automatic measurement method of corneal topography parameters

[0121] The trained scleral protrusion keypoint detection model and corneal contour recognition model are embedded in the automatic corneal topography parameter measurement algorithm framework. After image processing through steps such as correction, scleral protrusion keypoint recognition, corneal tissue image segmentation, and corneal boundary detection, we can obtain depth information for each pixel, providing high-quality data for subsequent topographic analysis. Ultimately, the information output by each module is fed into the parameter measurement program for parameter calculation and output.

[0122] When performing medical corneal topography analysis, key indicators are considered to help assess the shape, curvature, and other morphological features of the cornea. These indicators provide a scientific basis for the diagnosis and treatment of corneal diseases. The following are some important corneal topography indicators and their corresponding solutions:

[0123] Mean curvature

[0124] The mean curvature is an assessment of the overall curvature of the corneal surface. The mean curvature can be obtained by calculating the gradient of the curvature field. In digital signal processing, a differential calculation method can be used to calculate the gradient of the corneal surface height data in the x and y directions. The curvature field is obtained by taking the square root of the sum of the squares of these two gradients, and the mean curvature is then averaged. A larger mean curvature value indicates a more dramatic change in the corneal curvature.

[0125] Principal curvatures and principal curvature directions

[0126] The principal curvatures describe the maximum and minimum curvatures of the corneal surface at a specific point. By calculating the second-order derivative of the curvature field, the principal curvatures and directions of the corneal surface can be determined. The magnitude of the principal curvature reflects the degree of curvature of the cornea at that point, while the directions of the principal curvature indicate the directions of maximum and minimum curvature. This information is crucial for assessing corneal shape and stability, particularly in the diagnosis and treatment of corneal diseases.

[0127] Central corneal thickness

[0128] Central corneal thickness refers to the thickness of the cornea in the central region. It is determined by analyzing the corneal surface height data in the central region. Central corneal thickness is a key parameter for assessing corneal health and surgical suitability, and plays an important role in diagnosing corneal diseases and selecting surgical procedures.

[0129] Corneal astigmatism

[0130] Corneal astigmatism describes the curvature of the corneal surface, typically expressed as the radius of curvature at a specific point. In digital signal processing, this value can be obtained by calculating the inverse of the curvature field. A higher corneal astigmatism value indicates a smaller radius of curvature, meaning a steeper cornea.

[0131] Corneal dysmorphia index

[0132] The corneal dysmorphology index (KDI) is a comprehensive measure of the curvature of the anterior and posterior corneal surfaces, typically calculated as the ratio of the anterior to posterior curvatures. This index is important for assessing the overall morphology and variability of the cornea, providing a reference for planning and evaluating corneal surgery.

[0133] Comprehensive analysis of these indicators helps provide a comprehensive understanding of corneal morphology. Through efficient automated measurement methods, these corneal topographic parameters can be accurately calculated, providing a scientific basis for the diagnosis and treatment of corneal diseases and assisting physicians in developing more precise surgical plans.

[0134] Implementation steps:

[0135] Data preprocessing: Collect and standardize corneal OCT image data.

[0136] Gradient calculation: Use the differential calculation method to calculate the gradient of the corneal surface height data in the x and y directions to obtain the curvature field.

[0137] Second-order derivative calculation: Calculate the second-order derivative of the curvature field to obtain the principal curvature and principal curvature direction.

[0138] Thickness analysis: Analyze corneal surface height data and calculate central corneal thickness.

[0139] Astigmatism calculation: Calculate the inverse of the curvature field to obtain corneal astigmatism.

[0140] Corneal dysmorphic index calculation: The corneal dysmorphic index is calculated by the ratio of the anterior surface curvature to the posterior surface curvature.

[0141] Result output: Output various corneal topography parameters and generate reports for clinical diagnosis and surgical planning.

[0142] Through this measurement method, corneal topography parameters can be measured automatically, efficiently and accurately, providing a solid data foundation for corneal health assessment and disease treatment.

[0143] To achieve seamless interaction between the automatic parameter detection algorithm and the OCT device, the Django framework was selected to encapsulate the detection algorithm. To ensure rapid response during the detection process, the system design avoids the traditional method of storing images captured by the OCT device on disk and then retrieving them. Instead, Redis caching technology is used to cache images directly in memory. The automatic detection algorithm then directly accesses these cached images from memory using a specific key for measurement. This design significantly improves detection speed and system response efficiency, ensuring real-time, efficient image processing and parameter detection.

[0144] Examples

[0145] Figure 1 This is a flowchart of the visualization method and automatic measurement method implemented in an embodiment of the present invention. The left side shows the hardware equipment selection and module construction of the method, which mainly includes the selection of the anterior segment OCT laser light source, signal acquisition card, photodetector, and the construction of the optical path and signal transmission path of the signal path, so as to collect high-definition corneal grayscale images for subsequent measurement methods. The upper right side of the technology roadmap is the overall framework for the construction of the anterior segment OCT data set and the training of the deep learning model, which consists of two basic tasks: the scleral protrusion key point detection model and the corneal contour recognition model. The lower right side of the technology roadmap is the data visualization and parameter automatic measurement module, which realizes the visualization of the corneal topography information heat map and the automatic measurement of corneal parameters, and automatically diagnoses the incidence of keratoconus based on multi-dimensional parameter information. Figure 2 It is a schematic diagram of the hardware basis of the present invention, including the transmission paths of the optical signal path and the digital signal path.

[0146] Figure 6 This is a logic diagram for executing the automatic measurement method according to an embodiment of the present invention. A set of 128 OCT images input to the automatic parameter measurement algorithm is first processed by the deep learning algorithm module to obtain a 2D coordinate cloud of the corneal boundary section. This is then converted into a 3D coordinate cloud of the upper and lower corneal surfaces through image sequence processing and a 2D-to-3D coordinate conversion matrix. Zernike polynomials are then fitted to the upper and lower surfaces to obtain Zernike polynomials for the corneal topography. Multiple key parameters are then calculated for the polynomials to output the patient's corneal topography parameters.

[0147] Figure 7This is a logic diagram for the interaction between an embodiment of the present invention and an OCT device. To implement interaction between the automatic parameter detection algorithm and the OCT device, the present invention uses the Django framework to encapsulate the algorithm. To ensure rapid response of the automatic detection method, when interacting with the OCT device host, the present invention does not first store the images captured by the OCT device on the hard disk before inputting the OCT images stored on the hard disk into the automatic detection algorithm. Instead, the images captured by the OCT device are cached in memory via Redis. The automatic detection algorithm then directly retrieves the OCT images from memory using the corresponding key for measurement.

[0148] Figure 8 is an example of an OCT image input in the present invention, Figure 9 The corneal topography height map is the visualization effect of the measurement of the present invention, wherein parameters such as Ks and Kf are not visualized but directly calculated and the calculation results are returned.

[0149] The parameter calculation combines the contour information of the chamber angle area and the coordinate position information of the scleral protrusion. The specific method is as follows Figure 6 As shown in the figure, after obtaining the formula of the corneal surface through Zernike polynomial fitting, the parameters of the anterior surface such as Ks, Kf, CYL, AvgK, AA, and ACCP can be further calculated. The following are the specific steps and calculation methods:

[0150] Corneal curvature:

[0151] Corneal curvature is an important feature of corneal shape and can be obtained by calculating the local curvature of the corneal surface.

[0152] Ks (steep curvature): The curvature of the corneal surface in the direction of maximum curvature.

[0153] Kf (flat curvature): The curvature of the corneal surface in the direction of minimum curvature.

[0154]

[0155] (1) Calculate the second-order derivative:

[0156] K1=Z xx +Z yy

[0157] (2) Calculate the principal curvature:

[0158] The corneal surface can be reconstructed using Zernike polynomials to calculate the local curvature at each point and find the direction with the maximum curvature (Ks) and the direction with the minimum curvature (Kf).

[0159]

[0160] Astigmatism CYL (Cylinder):

[0161] Astigmatism is the difference between the steep curvature (Ks) and the flat curvature (Kf).

[0162] CYL=K s -K f

[0163] Average Keratometric Value (AvgK):

[0164] The mean curvature is the average of the steep curvature (Ks) and the flat curvature (Kf).

[0165]

[0166] AA (Apex Angle): The angle of the corneal apex

[0167] It can be calculated by analyzing the normal vector of the corneal surface.

[0168]

[0169] ACCP (Average Corneal Curvature Power):

[0170] The average corneal curvature within a specific area (such as 9mm).

[0171]

[0172] An embodiment of the present invention further provides a storage medium for storing a computer program, which at least performs the above method when executed.

[0173] An embodiment of the present invention further provides a control device, comprising a processor and a storage medium for storing a computer program; wherein the processor is configured to execute at least the method described above when executing the computer program.

[0174] An embodiment of the present invention further provides a processor, which executes a computer program and at least performs the method described above.

[0175] The storage medium can be implemented by any type of non-volatile storage device, or a combination thereof. Among them, the non-volatile memory can be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), a magnetic random access memory (FRAM), a flash memory, a magnetic surface memory, an optical disc, or a compact disc read-only memory (CD-ROM); the magnetic surface memory can be a magnetic disk memory or a magnetic tape memory. The storage medium described in the embodiments of the present invention is intended to include, but is not limited to, these and any other suitable types of memory.

[0176] In the several embodiments provided by the present invention, it should be understood that the disclosed systems and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.

[0177] The units described above as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units; some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0178] In addition, all functional units in the embodiments of the present invention may be integrated into one processing unit, or each unit may be separately used as a unit, or two or more units may be integrated into one unit; the above-mentioned integrated units may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0179] Those skilled in the art will appreciate that all or part of the steps of the above-mentioned method embodiments may be implemented by hardware associated with program instructions, and the aforementioned program may be stored in a computer-readable storage medium. When the program is executed, the program executes the steps of the above-mentioned method embodiments. The aforementioned storage medium includes various media that can store program codes, such as mobile storage devices, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical disks.

[0180] Alternatively, if the above-mentioned integrated unit of the present invention is implemented in the form of a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a number of instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as mobile storage devices, ROM, RAM, magnetic disks or optical disks.

[0181] The methods disclosed in the several method embodiments provided by the present invention can be arbitrarily combined without conflict to obtain new method embodiments.

[0182] The features disclosed in several product embodiments provided by the present invention can be arbitrarily combined without conflict to obtain new product embodiments.

[0183] The features disclosed in several method or device embodiments provided by the present invention can be arbitrarily combined without conflict to obtain new method embodiments or device embodiments.

[0184] The above is a further detailed description of the present invention in conjunction with specific preferred embodiments, and the specific implementation of the present invention should not be considered to be limited to these descriptions. Those skilled in the art will recognize that, without departing from the scope of the present invention, several equivalent substitutions or obvious variations can be made, and the performance or use of the same should be considered to fall within the scope of protection of the present invention.

Claims

1. A method for measuring corneal topography parameters from anterior segment OCT images, characterized in that: The following steps are involved: S1. Preprocessing the collected anterior segment OCT images; S2. Use deep learning algorithms to detect scleral protrusion feature points in OCT images and determine the corneal central axis to provide a reference for image registration; S3. Use the trained semantic segmentation model to segment the cornea and iris in the OCT image and generate a binary image. S4. extracting the corneal contour based on the segmented image; S5. Determine the corneal boundary by combining the scleral protrusion feature points and the corneal contour; S6. Fitting the corneal surface to establish a mathematical model of the corneal morphology; wherein the corneal central axis determined by the scleral protrusion feature points is used as a symmetry reference in the fitting process, and the corneal contour data is used as the corneal boundary of the fitting model to perform an overall fitting of the corneal surface; S7. calculating corneal topographic parameters based on the fitted mathematical model of corneal morphology; S8. integrating the calculated corneal topography parameters to generate a visual corneal topography map and parameter report; Step S6 specifically includes: Based on steps S3 and S4, the corneal image is further semantically segmented to capture the corneal section morphology and record the image position; Based on the corneal contour extracted in step S4, the coordinates of the corneal closure boundary points are recorded using an edge detection function; Using the corneal boundary determined in step S5 and the corneal contour obtained in step S4, the two-dimensional coordinates of the corneal edge are converted into three-dimensional coordinates of the entire cornea through a two-dimensional-three-dimensional coordinate conversion matrix to construct a more accurate three-dimensional corneal boundary point cloud model; The obtained corneal three-dimensional point cloud model was fitted with Zernike polynomials to obtain the edge equations of the upper and lower surfaces of the cornea, providing smooth surface data for subsequent topographic analysis.

2. The method for measuring corneal topography parameters from anterior segment OCT images according to claim 1, wherein: In step S2, the detection of scleral protrusion feature points in the OCT image specifically includes the following sub-steps: S21. Using a direct regression algorithm and computer vision techniques, locate the anterior chamber angle and crop a rectangular region containing the key points of the scleral protrusion. S22. Apply the deep learning ResNet50 model to detect key points in the cropped area and output the predicted coordinates of the scleral protrusion key points. The ResNet50 model is trained using the scleral protrusion key point dataset, and the mean squared error (MSE) loss function is used as the model loss function.

3. The method for measuring corneal topography parameters from anterior segment OCT images according to claim 1 or 2, wherein: In step S2, through scleral protrusion key point detection, the key points identified on both sides of the chamber angle are connected by straight lines and the image is rotated to a straight horizontal level. Then, the image is translated along the y-axis until the corneal vertex reaches a preset uniform height. After batch processing a group of OCT images, the alignment is completed.

4. The method for measuring corneal topography parameters from anterior segment OCT images according to any one of claims 1 to 2, wherein: In step S3, the trained U2net is used as a semantic segmentation model to perform semantic segmentation on the preprocessed OCT image, separating different tissues of the anterior segment including the cornea and iris, and generating a binary image; wherein BCELoss is used as the training loss function; In step S4, based on the binary image, an edge detection operator is applied to extract the precise contours of the cornea and iris; the extracted contours are optimized to remove noise and discontinuous edges, thereby achieving smoothness and precision of the contours.

5. The method for measuring corneal topography parameters from anterior segment OCT images according to any one of claims 1 to 2, characterized in that: In step S7, the key parameters for calculating corneal topography specifically include one or more of the following parameters: Calculate the average corneal curvature by calculating the gradient of the corneal surface height data in the x and y directions, obtain the curvature field, and take the average value to obtain the average curvature; Calculate the main curvature and direction of the cornea, and obtain the maximum and minimum curvatures and their directions of the corneal surface through the second-order derivative analysis of the curvature field; Measure the central corneal thickness by analyzing the value of the corneal surface height data in the central area to obtain the central thickness; Calculate corneal astigmatism, which is obtained by calculating the inverse of the curvature field, reflecting the curvature of the corneal surface; The corneal dysmorphism index was calculated to assess the overall corneal morphology and variability by the ratio of the anterior to posterior surface curvatures.

6. The method for measuring corneal topography parameters from anterior segment OCT images according to any one of claims 1 to 2, characterized in that: In step S8, based on the extracted corneal distribution information, a specific colorbar mapping method is used to convert this information into a visual corneal topography map, so as to intuitively display the ups and downs of the corneal topography and the refractive state.

7. A device for measuring corneal topography parameters of anterior segment OCT images for executing the method according to any one of claims 1 to 6, characterized in that: include: An image preprocessing unit, used for preprocessing the collected OCT image of the anterior segment of the eye; a deep learning processing unit, configured to run a deep learning algorithm to detect scleral protrusion feature points in the OCT image and determine the corneal central axis; The semantic segmentation unit is used to process the OCT image using the trained semantic segmentation model to segment the cornea and iris and generate a binary image; A contour extraction unit, configured to extract a corneal contour based on the segmented image; Corneal boundary determination unit, used to determine the corneal boundary by combining the scleral protrusion feature points and the optimized corneal contour, and provide spatial positioning for three-dimensional reconstruction and parameter calculation; Corneal morphology modeling unit, used to fit the corneal surface and establish a mathematical model of corneal morphology; A parameter calculation unit, used for calculating corneal topography parameters according to the fitted corneal morphology mathematical model; The result integration and visualization unit is used to integrate the calculated corneal topography parameters and generate a visual corneal topography map and parameter report.

8. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method for measuring corneal topography parameters of anterior segment OCT images according to any one of claims 1 to 6 is implemented.

9. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the method for measuring corneal topography parameters of anterior segment OCT images according to any one of claims 1 to 6 is implemented.

Images