Method and system for calculating cumulative area of anterior macular membrane of eye

Abstract

The invention relates to the technical field of medical image processing, in particular to an eye anterior macular membrane cumulative area calculation method and system. The method comprises the following steps: acquiring an optical coherence tomography cube scanning image of an eyeball macular region, wherein the cube scanning image comprises a plurality of parallel cross section images; constructing an artificial intelligence model to automatically segment the anterior macular membrane area of each cross section image in the cubic scanning image; stacking each cross section image in the cube scanning image according to a spatial sequence and projecting to obtain a macular anterior membrane projection drawing; calculating the area of the anterior macular film projection drawing by using a pixel point area calculation method to obtain an anterior macular film accumulated area; according to the method, the anterior macular membrane accumulated area can be automatically calculated by reading the scanning image of the optical coherence tomography cube in the macular region.

Description

Technical Field

[0001] This invention relates to the field of medical image processing and pattern recognition technology, specifically to a method and system for calculating the area of ​​the epiretinal membrane in the eye. Background Technology

[0002] The epiretinal membrane (ERM) is a fibroblast membrane formed by the proliferation of anterior retinal muscle fibroblasts in the macular region. On optical coherence tomography (OCT) images, it appears as a highly reflective band-like structure closely adhering to the surface of the internal limiting membrane of the retina and separated from the retinal neuroepithelial layer. ELM is a common vitreoretinal interface disease characterized by the formation of an anterior retinal fibroblast membrane. Its prevalence ranges from 4% to 28.9%, increasing with age. Although some patients are asymptomatic, the progression of ELM can cause pathological conditions such as retinal vascular tortuosity, macular cystoid edema, foveal heterotopia, and macular holes, leading to severe visual impairment. Typical clinical manifestations include decreased visual acuity, metamorphopsia, and anisotropy, significantly impacting patients' quality of life. The retinal deformities and decreased visual acuity associated with ELM are related to the traction exerted on the retina by the ELM; a larger area of ​​ELM can cause greater traction on the retina, resulting in a worse prognosis. Therefore, the extent of ELM involvement is an important factor in prognostic assessment.

[0003] In related technologies, the detection of the epiretinal membrane mainly relies on imaging examinations such as optical coherence tomography (OCT) and fundus photography. OCT, due to its high sensitivity and superior ability to visualize epiretinal membrane features, has become the gold standard for epiretinal membrane diagnosis. However, there is currently no method to automatically assess the size of the epiretinal membrane involvement based on OCT images. The analysis primarily depends on the physician's clinical experience and subjective judgment, which is not only time-consuming and labor-intensive but also prone to inaccurate judgments due to physician fatigue or differences in experience. With the development of computer vision and artificial intelligence technologies, automated image analysis methods have made it possible to automatically calculate the area of ​​epiretinal membrane involvement. Therefore, developing a method and system that can accurately and quickly identify epiretinal membrane regions and calculate the size of the involved area based on OCT images has significant clinical significance and application value. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides a method and system for calculating the area of ​​the epiretinal membrane in the eye, which aims to automatically calculate the area of ​​the epiretinal membrane by reading the optical coherence tomography cubic scan image of the macular region.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] On one hand, embodiments of the present invention provide a method for calculating the area of ​​epiretinal membrane in the eye, the method comprising the following steps:

[0007] S100, acquire an optical coherence tomography cube scan image of the macular region of the eyeball, the cube scan image containing multiple parallel cross-sectional images;

[0008] S200, Construct an artificial intelligence model to automatically segment the premacula region of each cross-sectional image in the cube scan image;

[0009] S300, stack the cross-sectional images of the cube scan image in spatial order and project them to obtain a macular anterior membrane projection image;

[0010] S400, the area of ​​the projection map of the premacula is calculated using the pixel area calculation method to obtain the accumulated area of ​​the premacula.

[0011] Optionally, in S100, acquiring an optical coherence tomography cube scan image of the macular region of the eye includes:

[0012] S110 acquires cubic scan images of the macular region of the eye using an optical coherence tomography scanner;

[0013] The S110 instructs the subject to fixate on a preset target and automatically completes the localization and focusing of the scanning area. It drives the scanning mirror to execute a preset raster scanning sequence, acquires A-scan signals across the entire area, integrates real-time eye tracking to compensate for motion artifacts, automatically reconstructs and evaluates the quality of the three-dimensional volume data, and outputs a cube scan image that meets diagnostic requirements.

[0014] Optionally, in S200, the construction of the artificial intelligence model to automatically segment the premacula region of each cross-sectional image in the cube scan image includes:

[0015] S210, Data annotation is performed on the cross-sectional images in the cube scan image. 16 cross-sectional images are extracted from each cube at equal intervals. For a cube containing 128 cross-sectional images, one image is extracted every 8 images. For a cube containing 512 cross-sectional images, one image is extracted every 32 images. The region where the premacula is located in each extracted cross-sectional image is marked using annotation software.

[0016] S220, perform image preprocessing on the labeled cross-sectional image, including grayscale normalization, linear normalization to the [0,1] interval for 8-bit images, contrast stretching and cropping to the same interval for 16-bit images based on 1%-99% quantiles, bilinear interpolation scaling of the image according to the preset size, nearest neighbor interpolation scaling of the mask, threshold binarization of the labeled mask, and conversion to a {0,1} binary image;

[0017] S230 is designed as an artificial intelligence model based on the U-Net core network structure of encoder-decoder architecture. The encoder uses a ResNet34 backbone network pre-trained on the ImageNet dataset. The decoder performs multi-level feature fusion with the jump connection in the feature map of the encoder through upsampling operation. The predicted logits are output through a single-channel segmentation head.

[0018] S240, train the artificial intelligence model, load ImageNet pre-trained weights for transfer learning, perform end-to-end fine-tuning of the encoder, adopt a hybrid loss function combining binary cross-entropy and Dice coefficients, select the AdamW optimizer, set weight decay, initial learning rate, batch size and training period, and save the best model weights based on the Dice coefficients of the validation set.

[0019] S250 uses a trained artificial intelligence model to predict the anterior macular membrane region. It selects the corresponding dataset for data reading based on whether the input data contains a ground truth mask. It performs sigmoid normalization on the logits output by the model, binarizes them with a threshold of 0.5, generates the initial segmentation result, and restores the predicted mask to the original image size through nearest neighbor interpolation.

[0020] Optionally, in S300, the step of stacking the cross-sectional images of the cubic scan image in spatial order and projecting them to obtain a projection image of the anterior macula includes:

[0021] S310: Using the binary profiles of each cross section of the segmented cube, stack the profiles of each cross section in the original spatial order to generate the initial three-dimensional binary volume of the anterior macular membrane.

[0022] S320, filters out isolated noise of less than 25 voxels in the three-dimensional binary volume through maximum connected component analysis, and performs morphological opening operation along the Z-axis to smooth the three-dimensional boundary;

[0023] S330 performs maximum intensity projection on the processed three-dimensional volume along the Z-axis to generate a projection map of the macular anterior membrane.

[0024] Optionally, in S400, the step of calculating the area of ​​the anterior macula projection map using a pixel area calculation method to obtain the accumulated area of ​​the anterior macula includes:

[0025] S410 calculates the actual area corresponding to a single pixel based on the axial resolution and transverse scan length calibration parameters of the optical coherence tomography (OCT) equipment.

[0026] S420, count the total number of pixels classified as anterior macular membrane in the projection image of the macula, and multiply the total number of pixels by the actual area of ​​each pixel to obtain the accumulated area of ​​the anterior macular membrane.

[0027] Optionally, in S410, calculating the actual area corresponding to a single pixel based on the axial resolution and transverse scan length calibration parameters of the optical coherence tomography (OCT) device includes:

[0028] S411, acquire the lateral scan length and axial scan depth of the optical coherence tomography (OCT) device, as well as the pixel size of the digitized image in the lateral direction and the pixel size in the axial direction;

[0029] S412, divide the horizontal scan length by the pixel size of the digitized image in the horizontal direction to obtain the actual physical size represented by a single pixel in the horizontal direction; divide the axial scan depth by the pixel size of the digitized image in the axial direction to obtain the actual physical size represented by a single pixel in the axial direction.

[0030] S413 multiplies the actual physical size of a single pixel in the horizontal direction with the actual physical size of a single pixel in the axial direction to obtain the actual area of ​​a single pixel.

[0031] Optionally, in S110, acquiring a cubic scan image of the macular region of the eye using an optical coherence tomography scanner includes:

[0032] S111, Initialize the optical coherence tomography scanner, perform equipment self-test and calibration, including laser source power stability test, reference arm optical path alignment verification and detector sensitivity calibration;

[0033] S112, instruct the examinee to place their chin on the jaw rest and their forehead against the forehead rest, adjust the lifting platform to align the corneal apex with the optical axis, and turn on the real-time anterior segment imaging module to confirm the eyelid position and pupil center;

[0034] S113, Select the preset macular region scanning protocol. This protocol includes scanning range parameters, covering a 6mm×6mm area laterally, with an axial depth of 2mm. The A-scan frequency is set to 100kHz, the B-scan rate is 80 frames / second, and each cube scan image is composed of stacked B-scan images consisting of 512×512 A-scan points, containing a total of 128 or 512 continuous cross-sectional images.

[0035] S114, start the scanning process, control the scanning probe to perform three-dimensional scanning according to the preset trajectory, synchronously acquire the original interference signal, convert the interference signal into an optical signal in the depth direction through fast Fourier transform, and reconstruct the two-dimensional cross-sectional image in real time and form three-dimensional cube data.

[0036] S115 After scanning is completed, the system automatically performs quality assessment on the cube data, including signal strength analysis, motion artifact detection, and inter-layer registration accuracy verification. For unqualified data, an automatic rescan mechanism is triggered until a cube scan image that meets the quality standards is obtained.

[0037] On the other hand, embodiments of the present invention provide a system for calculating the area of ​​epiretinal membrane involvement in the eye, comprising:

[0038] At least one processor;

[0039] At least one memory for storing at least one program;

[0040] When the at least one program is executed by the at least one processor, the at least one processor performs the method described above.

[0041] On the other hand, embodiments of the present invention provide a computer-readable storage medium storing a processor-executable program, which, when executed by a processor, is used to perform the above-described method.

[0042] The embodiments of the present invention have the following beneficial effects:

[0043] This invention provides a method and system for calculating the area of ​​epiretinal membrane involvement in the eye. By acquiring a cube scan image of the macular region using optical coherence tomography (OCT), an artificial intelligence model is constructed to automatically segment the epiretinal membrane region in each cross-section of the cube scan image. The cross-section images are then stacked in spatial order and projected to obtain a projection image of the epiretinal membrane. Finally, the area of ​​this projection image is calculated using a pixel area calculation method, thus obtaining the area of ​​the epiretinal membrane involvement. This method eliminates reliance on doctors' clinical experience and subjective judgment, achieving automated and objective calculation of the epiretinal membrane involvement area based on OCT images. It not only significantly improves analysis efficiency and saves doctors' time and effort but also effectively avoids inaccurate judgments that may be caused by doctor fatigue or experience differences. It provides a more accurate and reliable quantitative indicator for the prognostic assessment of epiretinal membrane diseases, contributing to improved scientific rigor in clinical diagnosis and treatment decisions. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0045] Figure 1This is a flowchart illustrating the method for calculating the area of ​​the epiretinal membrane in the eye according to an embodiment of the present invention.

[0046] Figure 2 This is a structural diagram of the method for calculating the area of ​​the epiretinal membrane in the eye according to an embodiment of the present invention. Detailed Implementation

[0047] The following will provide a clear and complete description of the concept, specific structure, and technical effects of the present invention in conjunction with embodiments and accompanying drawings, so as to fully understand the purpose, solution, and effects of the present invention. It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other.

[0048] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. In the following description, when referring to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the embodiments of this invention; they are merely examples of apparatuses and methods consistent with some aspects of the embodiments of this invention as detailed in the appended claims.

[0049] It is understood that the terms "first," "second," etc., used in this invention may be used to describe various concepts, but unless specifically stated otherwise, these concepts are not limited by these terms. These terms are only used to distinguish one concept from another. For example, without departing from the scope of embodiments of this invention, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the words "if" or "when" as used herein may be interpreted as "when," "in response to determination," or "in the event of a determination."

[0050] The terms “at least one,” “multiple,” “each,” “any,” etc., used in this invention, “at least one” includes one, two, or more than two; “multiple” includes two or more than two; “each” refers to each of the corresponding multiple; and “any” refers to any one of the multiple.

[0051] Unless otherwise defined, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used in this invention is for descriptive purposes only and is not intended to limit the invention.

[0052] refer to Figure 1 ,like Figure 1The figure shown is a method for calculating the area of ​​epiretinal membrane in the eye according to an embodiment of the present invention. The method includes the following steps:

[0053] S100, acquire an optical coherence tomography cube scan image of the macular region of the eyeball, the cube scan image containing multiple parallel cross-sectional images;

[0054] S200, Construct an artificial intelligence model to automatically segment the premacula region of each cross-sectional image in the cube scan image;

[0055] S300, stack the cross-sectional images of the cube scan image in spatial order and project them to obtain a macular anterior membrane projection image;

[0056] S400, the area of ​​the projection map of the premacula is calculated using the pixel area calculation method to obtain the accumulated area of ​​the premacula.

[0057] In steps S100 to S400 of this embodiment, the accurate quantification of the area affected by the anterior macular membrane is achieved through an automated processing flow of the cube scan image of the macular region obtained by optical coherence tomography (OCT). In S100, acquiring high-quality cube scan images through an OCT scanner is the foundation of the entire calculation process. This step not only includes the subject fixing their gaze on the target to assist in scanning positioning, but also integrates real-time eye-tracking technology to effectively compensate for motion artifacts, ensuring the stability and accuracy of the scan data. Simultaneously, the system automatically reconstructs the three-dimensional volume data and evaluates its quality; only cube scan images that meet diagnostic requirements are output, providing reliable data assurance for subsequent accurate segmentation. The construction and application of the artificial intelligence model in S200 is one of the core innovations of this method. The data annotation stage employs different extraction strategies based on the number of cross-sectional images in the cube, ensuring both the representativeness of the annotated samples and annotation efficiency. Image preprocessing steps, through operations such as grayscale normalization, contrast stretching, interpolation scaling, and threshold binarization, unify the image and mask to a standard format suitable for model input, effectively improving the model's generalization ability. Based on the U-Net architecture and combined with the ResNet34 backbone network, the AI ​​model achieves multi-level feature fusion through encoder feature extraction and decoder upsampling and skip connections, accurately capturing the subtle structural features of the anterior macula. During model training, the application of transfer learning accelerates model convergence, while the use of a hybrid loss function balances the class imbalance problem, ensuring the accuracy of model segmentation. Finally, the trained model can automatically predict the input cross-sectional image and generate accurate binary masks of the anterior macula region. The S300 stacks the segmented anterior macula regions in spatial order to generate a three-dimensional binary volume. Maximum connected component analysis filters out isolated noise, and morphological opening operations are performed along the Z-axis to further optimize the boundaries of the three-dimensional structure. Subsequently, a maximum intensity projection is performed along the Z-axis to convert the three-dimensional epiretinal membrane structure into a two-dimensional projection image, visually demonstrating the extent of epiretinal membrane involvement in the macula. The S400 then converts the area of ​​the projection image into the actual area of ​​epiretinal membrane involvement using a pixel area calculation method. This step first accurately calculates the actual physical area corresponding to a single pixel based on the axial resolution and lateral scan length calibration parameters of the optical coherence tomography (OCT) equipment. Then, the total number of pixels representing the epiretinal membrane in the projection image is counted, and multiplied by the actual area of ​​a single pixel to obtain the accurate value of the epiretinal membrane involvement area. This process achieves a precise conversion from image pixels to actual physical area, providing objective and quantitative evaluation indicators for clinical practice.

[0058] In some embodiments, S100, acquiring an optical coherence tomography cube scan image of the macular region of the eye includes:

[0059] S110 acquires cubic scan images of the macular region of the eye using an optical coherence tomography scanner;

[0060] The S110 instructs the subject to fixate on a preset target and automatically completes the localization and focusing of the scanning area. It drives the scanning mirror to execute a preset raster scanning sequence, acquires A-scan signals across the entire area, integrates real-time eye tracking to compensate for motion artifacts, automatically reconstructs and evaluates the quality of the three-dimensional volume data, and outputs a cube scan image that meets diagnostic requirements.

[0061] This embodiment involves comprehensive performance tuning of the optical coherence tomography (OCT) scanner before scanning. This includes calibrating the laser wavelength to 840nm ± 5nm to ensure a balance between tissue penetration and resolution, and adjusting the light source bandwidth to 100nm to obtain high-resolution images with an axial resolution ≤ 5μm. During scanning, dynamic focusing technology is used to track changes in the macula's position in real time. When eye movement displacement exceeds 30μm, the scan is automatically paused and repositioned to ensure that the acquisition position accuracy of each cross-sectional image is controlled within ±10μm. The acquired raw A-scan signal is preprocessed with dispersion compensation and noise reduction filtering, and then converted into a depth-direction optical signal using a Fast Fourier Transform (FFT). The number of sampling points for the FFT is set to 4096 to improve depth resolution. The reconstructed 3D volume data is stored in the DICOM standard format, including metadata such as patient ID, scan time, and device model, facilitating subsequent data exchange and traceability between systems. In addition to routine signal strength analysis (average signal strength > 60dB) and motion artifact detection (motion blur pixel percentage < 5%), the quality assessment process also includes inter-layer registration accuracy verification. By calculating the mutual information entropy value of adjacent cross-sectional images, it ensures that the inter-layer displacement error is < 2 pixels, thus comprehensively guaranteeing that the quality of the cube scan image meets the stringent requirements of subsequent artificial intelligence model segmentation.

[0062] In some embodiments, S110, acquiring a cubic scan image of the macular region of the eye using an optical coherence tomography scanner includes:

[0063] S111, Initialize the optical coherence tomography scanner, perform equipment self-test and calibration, including laser source power stability test, reference arm optical path alignment verification and detector sensitivity calibration;

[0064] S112, instruct the examinee to place their chin on the jaw rest and their forehead against the forehead rest, adjust the lifting platform to align the corneal apex with the optical axis, and turn on the real-time anterior segment imaging module to confirm the eyelid position and pupil center;

[0065] S113, Select the preset macular region scanning protocol. This protocol includes scanning range parameters, covering a 6mm×6mm area laterally, with an axial depth of 2mm. The A-scan frequency is set to 100kHz, the B-scan rate is 80 frames / second, and each cube scan image is composed of stacked B-scan images consisting of 512×512 A-scan points, containing a total of 128 or 512 continuous cross-sectional images.

[0066] S114, start the scanning process, control the scanning probe to perform three-dimensional scanning according to the preset trajectory, synchronously acquire the original interference signal, convert the interference signal into an optical signal in the depth direction through fast Fourier transform, and reconstruct the two-dimensional cross-sectional image in real time and form three-dimensional cube data.

[0067] S115 After scanning is completed, the system automatically performs quality assessment on the cube data, including signal strength analysis, motion artifact detection, and inter-layer registration accuracy verification. For unqualified data, an automatic rescan mechanism is triggered until a cube scan image that meets the quality standards is obtained.

[0068] This embodiment ensures that the optical coherence tomography scanner operates in optimal condition through a standardized equipment initialization and calibration process. The final output cube scan image must simultaneously meet all quality control indicators to provide high-quality raw data input for subsequent artificial intelligence model segmentation.

[0069] In some embodiments, in S200, the construction of the artificial intelligence model to automatically segment the premacula region of each cross-sectional image in the cube scan image includes:

[0070] S210, Data annotation is performed on the cross-sectional images in the cube scan image. 16 cross-sectional images are extracted from each cube at equal intervals. For a cube containing 128 cross-sectional images, one image is extracted every 8 images. For a cube containing 512 cross-sectional images, one image is extracted every 32 images. The region where the premacula is located in each extracted cross-sectional image is marked using annotation software.

[0071] S220, perform image preprocessing on the labeled cross-sectional image, including grayscale normalization, linear normalization to the [0,1] interval for 8-bit images, contrast stretching and cropping to the same interval for 16-bit images based on 1%-99% quantiles, bilinear interpolation scaling of the image according to the preset size, nearest neighbor interpolation scaling of the mask, threshold binarization of the labeled mask, and conversion to a {0,1} binary image;

[0072] S230 is designed as an artificial intelligence model based on the U-Net core network structure of encoder-decoder architecture. The encoder uses a ResNet34 backbone network pre-trained on the ImageNet dataset. The decoder performs multi-level feature fusion with the jump connection in the feature map of the encoder through upsampling operation. The predicted logits are output through a single-channel segmentation head.

[0073] S240, train the artificial intelligence model, load ImageNet pre-trained weights for transfer learning, perform end-to-end fine-tuning of the encoder, adopt a hybrid loss function combining binary cross-entropy and Dice coefficients, select the AdamW optimizer, set weight decay, initial learning rate, batch size and training period, and save the best model weights based on the Dice coefficients of the validation set.

[0074] S250 uses a trained artificial intelligence model to predict the anterior macular membrane region. It selects the corresponding dataset for data reading based on whether the input data contains a ground truth mask. It performs sigmoid normalization on the logits output by the model, binarizes them with a threshold of 0.5, generates the initial segmentation result, and restores the predicted mask to the original image size through nearest neighbor interpolation.

[0075] This embodiment ensures a balance between the representativeness of training samples and annotation efficiency through a scientifically sound data annotation strategy. Differential extraction ratios are used for cube images with varying cross-sectional numbers, avoiding increased labor costs due to over-annotation while ensuring that the anterior macula's distribution characteristics in three-dimensional space are covered by the anterior macula. Image preprocessing effectively eliminates interference from grayscale differences caused by varying equipment and scanning conditions through grayscale normalization and contrast stretching. Interpolation scaling and threshold binarization unify the size and format of images and masks, providing standardized data for model input. The combination of the U-Net core network and the ResNet34 backbone network fully leverages ResNet34's advantages in image feature extraction. Its residual connections effectively alleviate the gradient vanishing problem during deep network training, while U-Net's encoder-decoder architecture and jumper connection design achieve the fusion of low-level to high-level features, enabling the model to not only capture the overall contour of the anterior macula but also accurately identify its edge details. The application of transfer learning significantly reduces the model's dependence on large-scale labeled data, accelerates model convergence, and improves initial performance. In the hybrid loss function, binary cross-entropy focuses on the classification accuracy of individual pixels, while the Dice coefficient emphasizes the cross-union ratio of the foreground region. The combination of the two effectively solves the class imbalance problem caused by the low pixel proportion in the macular anterior membrane region, allowing the model to focus more on the segmentation accuracy of the anterior membrane region during training. In the model prediction stage, sigmoid normalization and 0.5 threshold binarization convert the network output into a clear binary mask, while the interpolation restoration step ensures that the segmentation result is consistent with the original image size, laying an accurate foundation for subsequent 3D stacking and area calculation.

[0076] In some embodiments, S300, the step of stacking the cross-sectional images of the cube scan image in spatial order and projecting them to obtain a projection image of the anterior macula includes:

[0077] S310: Using the binary profiles of each cross section of the segmented cube, stack the profiles of each cross section in the original spatial order to generate the initial three-dimensional binary volume of the anterior macular membrane.

[0078] S320, filters out isolated noise of less than 25 voxels in the three-dimensional binary volume through maximum connected component analysis, and performs morphological opening operation along the Z-axis to smooth the three-dimensional boundary;

[0079] S330 performs maximum intensity projection on the processed three-dimensional volume along the Z-axis to generate a projection map of the macular anterior membrane.

[0080] This embodiment precisely stacks the segmented results of each cross section according to the spatial relationship during the original scan, ensuring that the interlayer spacing between adjacent cross sections is consistent with the actual scanning parameters, thereby constructing a complete binary volume reflecting the three-dimensional distribution of the anterior macular membrane. During the stacking process, the system automatically reads metadata such as layer thickness and interlayer spacing from the DICOM file and calibrates the Z-axis coordinates of each cross section to avoid distortion of the three-dimensional structure caused by interlayer displacement that may occur during the scanning process. The maximum connected component analysis step of S320 calculates the number of voxels of all connected components in the three-dimensional binary volume and retains the largest connected component as the true anterior macular membrane structure, effectively eliminating small volume artifacts caused by image noise, vitreous opacity, or model segmentation errors. The morphological opening operation performed along the Z-axis uses spherical structural elements and smooths the burrs and depressions on the three-dimensional boundary through an operation of erosion followed by expansion, while preserving the main morphological features of the anterior membrane. The S330's Maximum Intensity Projection (MIP) algorithm iterates through every (X,Y) coordinate point along the Z-axis in the 3D volume, and uses the maximum pixel value of that point across all cross-sections (i.e., the highest probability value of the presence of the epiretinal membrane) as the pixel value of the corresponding location in the projection image. This ultimately generates a projection image that visually shows the overall extent of the epiretinal membrane on the 2D plane of the macular region. The resolution of the projection image remains consistent with the original cross-sectional image, ensuring the accuracy of subsequent area calculations.

[0081] In some embodiments, S400, calculating the area of ​​the anterior macula projection map using a pixel area calculation method to obtain the accumulated area of ​​the anterior macula includes:

[0082] S410 calculates the actual area corresponding to a single pixel based on the axial resolution and transverse scan length calibration parameters of the optical coherence tomography (OCT) equipment.

[0083] S420, count the total number of pixels classified as anterior macular membrane in the projection image of the macula, and multiply the total number of pixels by the actual area of ​​each pixel to obtain the accumulated area of ​​the anterior macular membrane.

[0084] This embodiment ensures the accuracy of single-pixel actual area calculation through precise acquisition of calibration parameters. In S410, the axial resolution parameter is directly read from the DICOM metadata of the optical coherence tomography (OCT) device. This parameter represents the physical distance corresponding to each unit pixel in the Z-axis direction. For different models or OCT devices with adjusted parameters, the system automatically adapts to the calibration parameters in their DICOM files without manual intervention, ensuring the universality and accuracy of area calculation. When counting the total number of pixels, all pixels in the epiretinal membrane projection image are traversed, and pixels with a value of 1 (representing the epiretinal membrane region) in the binarized mask are counted. Multiplying the total number of pixels obtained by the calculation of the actual area of ​​a single pixel yields the accumulated area of ​​the epiretinal membrane in square micrometers. This process is fully automated, providing clinicians with objective and quantitative key indicators for assessing the severity of epiretinal membrane, monitoring disease progression, and evaluating treatment effectiveness.

[0085] In some embodiments, S410, calculating the actual area corresponding to a single pixel based on the axial resolution and transverse scan length calibration parameters of the optical coherence tomography (OCT) device includes:

[0086] S411, acquire the lateral scan length and axial scan depth of the optical coherence tomography (OCT) device, as well as the pixel size of the digitized image in the lateral direction and the pixel size in the axial direction;

[0087] S412, divide the horizontal scan length by the pixel size of the digitized image in the horizontal direction to obtain the actual physical size represented by a single pixel in the horizontal direction; divide the axial scan depth by the pixel size of the digitized image in the axial direction to obtain the actual physical size represented by a single pixel in the axial direction.

[0088] S413 multiplies the actual physical size of a single pixel in the horizontal direction with the actual physical size of a single pixel in the axial direction to obtain the actual area of ​​a single pixel.

[0089] This embodiment ensures the accuracy of single-pixel area by analyzing the calculation logic of the lateral and axial physical dimensions step by step. The calculation process strictly follows the physical mapping relationship of optical coherence tomography images, ensuring accurate conversion from digital image pixels to actual anatomical structure areas, providing a solid data foundation for clinical quantitative assessment.

[0090] refer to Figure 2 The following is a specific embodiment provided by the present invention:

[0091] (1) Optical coherence tomography cube scan image acquisition module: acquires cube scan images of the macular region of the eyeball through an optical coherence tomography scanner. The cube contains multiple parallel cross-sectional images.

[0092] (2) Macular anterior membrane region segmentation module: Construct an artificial intelligence model for automatically segmenting the macular anterior membrane region in each cross-sectional image;

[0093] (3) Macular anterior membrane projection image construction module: Stack the cross-sectional images of the cube in spatial order and project them to obtain the macular anterior membrane projection image;

[0094] (4) Calculation module for the area of ​​the epiretinal membrane: The area of ​​the epiretinal membrane projection is calculated using the pixel area calculation method;

[0095] In step (1) above, the image acquisition module, based on the optical coherence tomography (OCT) scanner, instructs the subject to fixate on a preset target and automatically completes the localization and focusing of the scanning area. Subsequently, the scanning mirror is driven to execute a preset raster scanning sequence, systematically acquiring A-scan signals across the entire area. During this process, real-time eye tracking is integrated to compensate for motion artifacts. Finally, the quality of the three-dimensional volume data is automatically reconstructed and evaluated to ensure that the output cube scan image meets diagnostic requirements. This system collected a total of 32 OCT cube scan images, of which 16 were from the Zeiss OCT scanner, each cube including 128 cross-sectional images; and 16 were from the Vision Micro OCT scanner, each cube including 512 cross-sectional images; a total of 10,240 cross-sectional images were collected.

[0096] In step (2) of the above-mentioned epiretinal membrane region segmentation module, it is necessary to segment the epiretinal membrane region on each cross-sectional image. The implementation process includes modules such as data annotation, image preprocessing, algorithm design, training process, and prediction process.

[0097] Data annotation: 16 cross-sectional images were extracted from each cube at equal intervals. For a cube with 128 cross-sectional images, one image was extracted every 8 images. For a cube with 512 cross-sectional images, one image was extracted every 32 images. Labelme software was used to annotate the region where the epiretinal membrane of the macula is located in each extracted cross-sectional image.

[0098] Image preprocessing: First, grayscale normalization is performed. For 8-bit images, linear normalization is used to the [0,1] interval. For 16-bit images, contrast stretching based on the 1%~99th percentile is used, followed by cropping to the same interval to enhance the model's adaptability to different imaging devices. Then, bilinear interpolation scaling is applied to the image according to a preset size, and nearest-neighbor interpolation scaling is used for the mask to ensure uniform input size. The labeled mask is thresholded and binarized to convert it into a {0,1} binary image for supervised training. Data augmentation strategies include horizontal flipping and copying to avoid tensor transformation errors. To protect the morphological integrity of the thin film structure, geometric transformations such as rotation and elastic deformation are excluded by default.

[0099] Algorithm Design: Based on an encoder-decoder architecture, U-Net is used as the core network structure. The encoder uses a ResNet34 backbone network pre-trained on the ImageNet dataset for multi-level feature extraction; the decoder performs multi-level feature fusion with jump connections in the encoder feature maps through upsampling operations to enhance the recovery of edges and fine structures. Finally, a single-channel segmentation head outputs predicted logits, which are then used in conjunction with a binary cross-entropy loss function to complete model optimization.

[0100] Training Process: First, transfer learning is performed using ImageNet pre-trained weights, followed by end-to-end fine-tuning of the encoder. A hybrid loss function combining binary cross-entropy and the Dice coefficient is used. Binary cross-entropy adjusts the imbalance between positive and negative samples using positive sample weights, while the Dice loss improves the prediction stability for thin-walled structures. The optimizer used is AdamW, with weight decay set to 1e-5, an initial learning rate of 1e-3, a batch size of 4, and a training cycle of 30 epochs. During training, a progress bar is used to monitor the loss and evaluation metrics in real time. TensorBoard is used to record the training curve, and prediction results are periodically visualized. Finally, the optimal model weights are saved based on the Dice coefficients from the validation set.

[0101] Prediction Process: First, based on whether the input data contains a ground truth mask, the system automatically selects either Seg OCTInferDataset or Seg OCT Dataset to read the data. Then, the model's output logits are sigmoid-normalized and binarized with a threshold of 0.5 to generate initial segmentation results. In the dedicated inference path, the predicted mask is restored to the original image size using nearest neighbor interpolation and saved to the same directory as the original image (file naming format: "original_pred.png"). In the full evaluation mode containing ground truth, the system generates a triptych preview image containing the original image, ground truth, and prediction results, and saves it to the specified results directory.

[0102] In step (3) of the above-mentioned module for constructing the projection map of the anterior macular membrane, the method for constructing the projection map of each optical coherence tomography cube is as follows: First, using the binary contours of each cross section of the cube obtained from the segmentation model in step (2), the contours of each cross section are stacked in the original spatial order to generate the initial three-dimensional binary volume of the anterior macular membrane. Subsequently, isolated noise of less than 25 voxels in the volume is filtered out by maximum connected component analysis, and morphological opening operation is performed along the Z-axis to smooth the three-dimensional boundary. Finally, the processed three-dimensional volume is projected with maximum intensity along the Z-axis to generate the final two-dimensional projection map of the anterior macular membrane for quantitative analysis.

[0103] In step (4) above, the module for calculating the area of ​​the macular anterior membrane first converts the image pixel units into actual physical dimensions based on the axial resolution and transverse scan length calibration parameters of the optical coherence tomography (OCT) equipment. The actual area (unit: mm²) corresponding to a single pixel is then calculated accordingly. The formula for calculating the actual area of ​​a single pixel is as follows:

[0104] ;

[0105] in, The actual physical area represented by a single pixel (unit: mm²). The scanning length (in mm) of the optical coherence tomography (OCT) equipment in the transverse (X-axis) direction. The scanning depth of the optical coherence tomography (OCT) device in the axial direction (Z-axis) is expressed in mm. W represents the pixel size (width) of the digitized image in the horizontal direction (X-axis), and H represents the pixel size (height) of the digitized image in the axial direction (Z-axis). This refers to the actual physical size (i.e., the width of a single pixel) represented by a single pixel in the horizontal (X-axis) direction. What is calculated is the actual physical size (i.e., single pixel height) represented by a single pixel in the axial (Z-axis) direction.

[0106] Subsequently, the system automatically counts the total number of pixels in all anterior macula regions of the maximum intensity projection image, and accurately outputs the cumulative area of ​​the anterior macula (unit: mm²) using the pixel area calculation formula. The pixel area calculation formula is as follows:

[0107] ;

[0108] in, denoted as the affected area of ​​the epiretinal membrane (unit: mm²), and N is the total number of pixels classified as epiretinal membrane in the two-dimensional projection image. This represents the actual area of ​​a single pixel (unit: mm²).

[0109] Compared with the prior art, the present invention has the following significant advantages:

[0110] This invention, based on optical coherence tomography (OCT) imaging technology and deep learning algorithms, can automatically and rapidly calculate the area of ​​epiretinal membrane involvement in the eye, assisting in clinical disease assessment and prognosis prediction, and has significant clinical significance and application value. This invention also has the advantages of being non-contact and non-invasive.

[0111] This invention also provides a system for calculating the area of ​​epiretinal membrane involvement in the eye, comprising:

[0112] At least one processor;

[0113] At least one memory for storing at least one program;

[0114] When the at least one program is executed by the at least one processor, the at least one processor performs the method described above.

[0115] The content of the above method embodiments is applicable to this embodiment. The specific functions implemented in this embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments. Therefore, they will not be repeated here.

[0116] This invention also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the method described above. This electronic device can be any smart terminal, including tablet computers, in-vehicle computers, etc.

[0117] It is understood that the content of the above method embodiments is applicable to this device embodiment. The specific functions implemented by this device embodiment are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0118] This invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method.

[0119] It is understood that the content of the above method embodiments is applicable to this storage medium embodiment. The specific functions implemented in this storage medium embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0120] This invention also provides a computer program product, including a computer program or computer instructions, which are stored in a memory. A processor of a computer device reads the computer program or computer instructions from the memory and executes the computer program or computer instructions, causing the computer device to perform the above-described method.

[0121] It is understood that the content of the above method embodiments is applicable to the embodiments of this program product. The specific functions implemented by the embodiments of this program product are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0122] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0123] It will be understood by those skilled in the art that all or some of the steps and systems in the methods disclosed above can be implemented as software, firmware, hardware, and suitable combinations thereof. Some or all of the physical components can be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit. Such software can be distributed on a computer-readable medium, which can include computer storage media (or non-transitory media) and communication media (or transient media). As is known to those skilled in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and is accessible to a computer. Furthermore, as is known to those skilled in the art, communication media typically include computer-readable instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.

[0124] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

Claims

1. A method for calculating the area of ​​epiretinal membrane involvement in the eye, characterized in that, The method includes the following steps: S100, acquire an optical coherence tomography cube scan image of the macular region of the eyeball, the cube scan image containing multiple parallel cross-sectional images; S200, Construct an artificial intelligence model to automatically segment the premacula region of each cross-sectional image in the cube scan image; S300, stack the cross-sectional images of the cube scan image in spatial order and project them to obtain a macular anterior membrane projection image; S400, the area of ​​the projection map of the premacula is calculated using the pixel area calculation method to obtain the accumulated area of ​​the premacula.

2. The method according to claim 1, characterized in that, In S100, acquiring the optical coherence tomography cube scan image of the macular region of the eye includes: S110 acquires cubic scan images of the macular region of the eye using an optical coherence tomography scanner; The S110 instructs the subject to fixate on a preset target and automatically completes the localization and focusing of the scanning area. It drives the scanning mirror to execute a preset raster scanning sequence, acquires A-scan signals across the entire area, integrates real-time eye tracking to compensate for motion artifacts, automatically reconstructs and evaluates the quality of the three-dimensional volume data, and outputs a cube scan image that meets diagnostic requirements.

3. The method according to claim 1, characterized in that, In S200, the construction of the artificial intelligence model to automatically segment the premacula region of each cross-sectional image in the cube scan image includes: S210, Data annotation is performed on the cross-sectional images in the cube scan image. 16 cross-sectional images are extracted from each cube at equal intervals. For a cube containing 128 cross-sectional images, one image is extracted every 8 images. For a cube containing 512 cross-sectional images, one image is extracted every 32 images. The region where the premacula is located in each extracted cross-sectional image is marked using annotation software. S220, perform image preprocessing on the labeled cross-sectional image, including grayscale normalization, linear normalization to the [0,1] interval for 8-bit images, contrast stretching and cropping to the same interval for 16-bit images based on 1%-99% quantiles, bilinear interpolation scaling of the image according to the preset size, nearest neighbor interpolation scaling of the mask, threshold binarization of the labeled mask, and conversion to a {0,1} binary image; S230 is designed as an artificial intelligence model based on the U-Net core network structure of encoder-decoder architecture. The encoder uses a ResNet34 backbone network pre-trained on the ImageNet dataset. The decoder performs multi-level feature fusion with the jump connection in the feature map of the encoder through upsampling operation. The predicted logits are output through a single-channel segmentation head. S240, train the artificial intelligence model, load ImageNet pre-trained weights for transfer learning, perform end-to-end fine-tuning of the encoder, adopt a hybrid loss function combining binary cross-entropy and Dice coefficients, select the AdamW optimizer, set weight decay, initial learning rate, batch size and training period, and save the best model weights based on the Dice coefficients of the validation set. S250 uses a trained artificial intelligence model to predict the anterior macular membrane region. It selects the corresponding dataset for data reading based on whether the input data contains a ground truth mask. It performs sigmoid normalization on the logits output by the model, binarizes them with a threshold of 0.5, generates the initial segmentation result, and restores the predicted mask to the original image size through nearest neighbor interpolation.

4. The method according to claim 1, characterized in that, In S300, the step of stacking the cross-sectional images of the cubic scan image in spatial order and projecting them to obtain a projection image of the anterior macula includes: S310: Using the binary profiles of each cross section of the segmented cube, stack the profiles of each cross section in the original spatial order to generate the initial three-dimensional binary volume of the anterior macular membrane. S320, filters out isolated noise of less than 25 voxels in the three-dimensional binary volume through maximum connected component analysis, and performs morphological opening operation along the Z-axis to smooth the three-dimensional boundary; S330 performs maximum intensity projection on the processed three-dimensional volume along the Z-axis to generate a projection map of the macular anterior membrane.

5. The method according to claim 1, characterized in that, In S400, the step of calculating the area of ​​the anterior macula projection map using a pixel area calculation method to obtain the accumulated area of ​​the anterior macula includes: S410 calculates the actual area corresponding to a single pixel based on the axial resolution and transverse scan length calibration parameters of the optical coherence tomography (OCT) equipment. S420, count the total number of pixels classified as anterior macular membrane in the projection image of the macula, and multiply the total number of pixels by the actual area of ​​each pixel to obtain the accumulated area of ​​the anterior macular membrane.

6. The method according to claim 5, characterized in that, In S410, the step of calculating the actual area corresponding to a single pixel based on the axial resolution and transverse scan length calibration parameters of the optical coherence tomography (OCT) device includes: S411, acquire the lateral scan length and axial scan depth of the optical coherence tomography (OCT) device, as well as the pixel size of the digitized image in the lateral direction and the pixel size in the axial direction; S412, divide the horizontal scan length by the pixel size of the digitized image in the horizontal direction to obtain the actual physical size represented by a single pixel in the horizontal direction; divide the axial scan depth by the pixel size of the digitized image in the axial direction to obtain the actual physical size represented by a single pixel in the axial direction. S413 multiplies the actual physical size of a single pixel in the horizontal direction with the actual physical size of a single pixel in the axial direction to obtain the actual area of ​​a single pixel.

7. The method according to claim 2, characterized in that, In S110, acquiring a cubic scan image of the macular region of the eye using an optical coherence tomography scanner includes: S111, Initialize the optical coherence tomography scanner, perform equipment self-test and calibration, including laser source power stability test, reference arm optical path alignment verification and detector sensitivity calibration; S112, instruct the examinee to place their chin on the jaw rest and their forehead against the forehead rest, adjust the lifting platform to align the corneal apex with the optical axis, and turn on the real-time anterior segment imaging module to confirm the eyelid position and pupil center; S113, Select the preset macular region scanning protocol. This protocol includes scanning range parameters, covering a 6mm×6mm area laterally, with an axial depth of 2mm. The A-scan frequency is set to 100kHz, the B-scan rate is 80 frames / second, and each cube scan image is composed of stacked B-scan images consisting of 512×512 A-scan points, containing a total of 128 or 512 continuous cross-sectional images. S114, start the scanning process, control the scanning probe to perform three-dimensional scanning according to the preset trajectory, synchronously acquire the original interference signal, convert the interference signal into an optical signal in the depth direction through fast Fourier transform, and reconstruct the two-dimensional cross-sectional image in real time and form three-dimensional cube data. S115 After scanning is completed, the system automatically performs quality assessment on the cube data, including signal strength analysis, motion artifact detection, and inter-layer registration accuracy verification. For unqualified data, an automatic rescan mechanism is triggered until a cube scan image that meets the quality standards is obtained.

8. A system for calculating the area of ​​epiretinal membrane involvement in the eye, characterized in that, include: At least one processor; At least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor performs the method as described in any one of claims 1 to 7.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 7.

Images