Ellipse band image feature analysis method for human eye retina OCT image
By employing ETDRS partitioning and ELM registration techniques in retinal OCT images, combined with gray-scale gradient algorithms to segment the EZ boundary, the accuracy and efficiency issues of EZ image feature quantization were resolved, enabling early diagnosis and functional assessment of retinal diseases.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- THE EYE HOSPITAL OF WENZHOU MEDICAL UNIVERSITY
- Filing Date
- 2026-05-26
- Publication Date
- 2026-06-23
AI Technical Summary
Existing technologies for EZ image feature quantification analysis suffer from high subjectivity and large errors, making it difficult to accurately quantify reflectivity and thickness changes, and thus failing to meet the needs of large-scale screening analysis, especially when EZ boundaries are blurred or diseases are complex.
Retinal OCT images were acquired using a radial scanning mode, segmented based on the ETDRS partitioning standard, and the retinal sublayers were segmented using a two-dimensional dynamic programming algorithm. The EZ boundary was segmented using ELM registration and gray-level gradient algorithm, and the thickness, reflectance intensity, and gray-level morphology ratio of the EZ were quantified to generate an ETDRS spatial distribution map.
It achieves efficient and accurate quantification of EZ image features, provides non-contact quantitative assessment of mitochondrial structure of retinal photoreceptor cells, and supports the diagnosis of early ophthalmic diseases and neurodegenerative diseases.
Smart Images
Figure CN122265276A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of ophthalmic optical imaging technology, specifically to a method for analyzing the elliptic band image features of human retinal OCT images. Background Technology
[0002] OCT (Optical Computed Tomography) is a non-invasive, high-resolution tomographic imaging technique, representing a major breakthrough in modern medical imaging. It plays a crucial role in clinical diagnosis and is an indispensable core tool for the diagnosis and research of retinal diseases. Specifically, this technique is based on low-coherence optical interferometry, acquiring three-dimensional images of the internal microstructure of biological tissues by measuring the intensity of backscattered light. OCT's axial resolution reaches the micrometer level, enabling it to clearly distinguish the various layers of the retina. OCT excels particularly in displaying the fine anatomical features of the outer retina, such as providing clear imaging of the elliptic zone (EZ), offering key evidence for studying the pathogenesis and pathological changes of retinal diseases.
[0003] The retinal zone (EZ) plays a crucial role in the structure and function of the retina. Located between the external limiting membrane (ELM) and the pigment epithelium (RPE), it is primarily composed of photoreceptor cell segments and contains numerous mitochondria. Mitochondria are the centers of cellular metabolism and are highly sensitive to pathological changes in cells. In OCT images, the EZ appears as a highly reflective band, possessing unique imaging characteristics that facilitate identification by analytical tools. Functionally, the integrity, reflectivity, and spatial distribution characteristics of the EZ are correlated with retinal diseases. For example, in OCT images of patients with age-related macular degeneration (AMD), diabetic retinopathy (DR), and Parkinson's disease (PD), the EZ exhibits different morphological changes, thickness variations, and reflectivity indices compared to normal patients. Therefore, quantitative analysis of the imaging characteristics of the EZ has significant clinical value for the early diagnosis and disease analysis of retinal diseases.
[0004] Currently, OCT technology is widely used in the diagnosis of retinal diseases, but further breakthroughs are needed in the quantitative analysis of EZ (extracorporeal membrane oxygenation) image features. Early EZ analysis relied on subjective human assessment, typically involving doctors observing and judging OCT images as "complete / interrupted / missing." This method is highly subjective and prone to error, with different doctors using different criteria, resulting in a Kappa coefficient of only 0.52 between groups, lower than the clinically acceptable 0.8. In the diagnosis of some complex diseases, such as AMD patients with localized lesions or diabetic macular edema, the EZ boundary becomes blurred, increasing the difficulty of judgment. The three-level grading method is too simplistic, only describing the continuity of the EZ and failing to quantify changes in reflectivity and thickness, affecting the accuracy of disease assessment. Furthermore, manual assessment is time-consuming and labor-intensive; analyzing a single OCT image takes 3-5 minutes, which cannot meet the needs of large-scale screening analysis and follow-up examinations. In recent years, various automated recognition and analysis methods based on image processing have been proposed, such as algorithms based on threshold segmentation, region growing, and active contour models. However, these methods often struggle to obtain accurate results when processing complex retinal OCT images, particularly in cases of blurred EZ boundaries or complex diseases. Existing research lacks a comprehensive quantification of the spatial distribution characteristics of the EZ and related studies on the EZ itself. Therefore, developing an efficient, accurate, and universally applicable method for quantifying EZ image features is of significant practical importance. Summary of the Invention
[0005] To overcome the shortcomings of the prior art, this invention provides a method for analyzing the elliptic zone image features of human retinal OCT images, mainly addressing the current lack of efficient, accurate, and universally applicable EZ image feature quantification methods.
[0006] The technical solution of the present invention is as follows: A method for analyzing the elliptic band image features of human retinal OCT images includes the following steps. Step 1: Using a retinal OCT device, with the fovea of the macula as the origin, samples are taken in a radial scanning mode to obtain a high-resolution OCT image sequence of the retina; Step 2, Retinal OCT Partitioning: Based on the ETDRS partitioning standard, with the fovea as a reference, concentric rings with diameters of 1mm, 3mm and 6mm are drawn to divide the retina into the foveal region, inner region and outer region. Step 3: Retinal boundary segmentation: A two-dimensional dynamic programming algorithm is used to segment each sublayer of the retina, including at least the boundary between the external retinal membrane (ELM) and the retinal pigment epithelium (RPE). Step 4: External membrane registration: Extract the boundary line of the external membrane and normalize it to a horizontal straight line through coordinate transformation; Step 5: Elliptic zone boundary segmentation: Based on the gray-level gradient, a dynamic programming algorithm is constructed to locate the upper and lower boundaries of the elliptic zone within the range from the outer membrane to the epithelial layer on each A-line scan line; Step 6: Elliptic Zone Image Feature Quantization: Based on ETDRS partitioning, spatial partition quantization is performed on the thickness, reflectance intensity, and grayscale morphology ratio of the EZ.
[0007] Step four, external membrane registration, specifically includes: extracting the lowest point in the depth direction of the external membrane boundary line as a reference line, calculating the coordinate transformation parameters of each A-line signal, and moving the external membrane layer to a horizontal position.
[0008] In the EZ boundary segmentation described in step five, the highest peak of the gray-level gradient is positioned as the upper boundary of EZ, the lowest valley is positioned as the lower boundary of EZ, and a search window is set between continuous A-lines to constrain the boundary search range.
[0009] The specific implementation process of step four is as follows: Phase 1: Automatically extract the lowest point in the depth direction of the ELM boundary line. Assume the point set of the ELM boundary is... Since the depth direction of an OCT image is encoded in ascending order from top to bottom, the maximum value corresponding to the ELM boundary is the minimum point of that boundary in the depth direction.
[0010] Using this as a baseline, calculate the set of coordinate transformation parameters Shift required to flatten the ELM layer.
[0011] Phase Two: Each A-line signal in the OCT image is moved downwards along the depth direction by a distance equal to the coordinate transformation parameter Shift at each position on the ELM boundary line. This spatial transformation operation positions the ELM horizontally within the retinal OCT image. This significantly improves the accuracy of EZ (elliptic zone) image feature measurement and allows for comparison of OCT images from different individuals based on the same standard, laying a solid foundation for subsequent statistical analysis and lesion feature extraction. Figure 4 The OCT images before and after ELM registration are shown.
[0012] The formula for calculating the thickness mentioned in step six is:
[0013] Where EZThickness is the EZ layer thickness, EZLow is the lower boundary pixel position of the EZ layer, EZUp is the upper boundary pixel position of the EZ layer, and D is the physical length of a unit pixel.
[0014] The formula for calculating the reflection intensity in step six is:
[0015] in This is the set of pixel coordinates where the EZ layer is located. Represents a set The total number of pixels in the middle. This represents the grayscale value of the corresponding pixel.
[0016] The calculation of the grayscale ratio in step six includes the following sub-steps: Sub-step 1: Baseline Correction: Connect the two endpoints of the curve to construct a linear baseline; then, calculate the vertical distance from each point on the original curve to this baseline, and define this distance as the point-corrected grayscale value of that pixel. ; Sub-step two: Calculate the zeroth moment, the formula is as follows: ; Sub-step 3: Calculate the first moment to determine the centroid ( )
[0017]
[0018] in:
[0019] ; Sub-step four: Calculate the second-order central moments: The second-order central moments describe the distribution shape of the profile.
[0020]
[0021] ; Sub-step 5: Calculate the eigenvalues of the covariance matrix: eigenvalues This represents the variance of the profile in the two principal directions.
[0022] ; Sub-step 6: Calculate the grayscale ratio: The grayscale ratio (GAR) is: .
[0023] It also includes generating ETDRS spatial distribution maps of EZ thickness, reflection intensity, and grayscale morphology ratio.
[0024] The beneficial effects of this invention are: This invention provides a method for analyzing the elliptic band image features of human retinal OCT images, which can realize non-contact quantitative assessment of mitochondrial clusters in human retinal photoreceptor cells, providing a new approach for assessing the structure and function of mitochondria in in vivo human retinal photoreceptor cells, and providing a new tool for early warning research on ophthalmic diseases such as AMD and neurodegenerative diseases such as PD. Attached Figure Description
[0025] Figure 1 This is a schematic diagram of retinal OCT acquisition based on a radial scanning pattern.
[0026] Figure 2 An ETDRS partition map is superimposed on a typical retinal OCT image. Figure 3 The image shows a typical retinal OCT image layering result (A) and an OCT image within 6 mm of the fovea of the macular region superimposed with the upper boundaries of the ELM and RPE (B). The blue solid line represents the boundary of the ELM and the red solid line represents the upper boundary of the RPE.
[0027] Figure 4 Images of typical retinal OCT images before (A) and after (B) ELM registration.
[0028] Figure 5 This is an OCT image with the EZ layer boundary superimposed.
[0029] Figure 6 The original grayscale distribution curve (A) and the schematic diagram of image moment analysis for the EZ region (B) are shown. Figure 7 Spatial distribution results of retinal EZ image features in normal individuals, AMD, and PD. (A, D, G) represent retinal EZ thickness in normal individuals, AMD, and PD, respectively; (B, E, H) represent retinal EZ reflectance intensity in normal individuals, AMD, and PD, respectively; (C, F, I) represent retinal EZ grayscale morphology ratio in normal individuals, AMD, and PD, respectively. Detailed Implementation
[0030] The invention will be further described below with reference to the accompanying drawings. This invention provides a method for analyzing EZ image features in human retinal OCT images, which includes key steps such as image acquisition, retinal OCT partitioning, retinal boundary segmentation, ELM registration, EZ boundary segmentation, and EZ image feature quantization, which are described in detail below: (1) Image acquisition: Using a commercial retinal OCT device, a radial scanning pattern was adopted with the fovea of the macula as the origin (see Figure 1 Sampling was performed to obtain high-resolution OCT image sequences of the retina for subsequent image analysis.
[0031] (2) Retinal OCT Partitioning: Retinal ETDRS partitioning is one of the most representative and influential assessment standards in modern optometry. It uses the fovea as the origin and divides the retina into three ring-shaped regions (fovea, inner ring, and outer ring) by three concentric circles (diameters 1 mm, 3 mm, and 6 mm). It also combines four quadrants (nasal, temporal, superior, and inferior) to locate lesions, aiming for early detection of small fundus lesions. This invention uses an interactive method to select the fovea in retinal OCT sequence images. Subsequently, using three-dimensional reconstruction, concentric rings with diameters of 1 mm, 3 mm, and 6 mm are drawn with the fovea as a reference to obtain the ETDRS partitioning (foveal region 1 mm, inner region 1-3 mm, outer region 3-6 mm). Figure 2 The ETDRS partitioning results are shown in a typical retinal OCT image.
[0032] (3) Retinal boundary segmentation: In the boundary detection process, this invention uses the previously independently developed two-dimensional dynamic programming retinal OCT sublayer segmentation algorithm (Liu X, Shen M, Huang S, Leng L, Zhu D, Lu F. Repeatability and reproducibility of eight macular intra-retinal layer thicknesses determined by an automated segmentation algorithm using two SD-OCT instruments. PLoS One. 2014;9(2):e87996.) to achieve automatic segmentation of the eight sublayers of the retina, including the retinal nerve fiber layer (RNFL), ganglion cell-inner plexiform layer (GCIPL), inner nuclear layer (INL), outer plexiform layer (OPL), outer nuclear layer (ONL), myoid and ellipsoid zones (MEZ), and photoreceptor outer ganglion (OPL). The EZ layer (Segment, OS) and RPE layer are also present. The EZ layer, being relatively thin, was not extracted separately but is instead embedded within the MEZ layer. Figure 3(A) shows a typical retinal OCT layering result, which can accurately determine the boundaries of the ELM layer and the RPE layer. By combining the retinal OCT layering result with the ETDRS partitioning result, an OCT image with the ELM and RPE boundaries superimposed within a 6mm range of the fovea can be obtained.
[0033] (4) ELM Registration: Following the previous step, the ELM boundary line in the retinal OCT image can be obtained. The ELM layer is then normalized to a horizontal straight line through coordinate transformation. The specific implementation process is as follows: ① Automatically extract the lowest point in the depth direction of the ELM boundary line. Assume the click value of the ELM boundary is... Since the depth direction of an OCT image is encoded in ascending order from top to bottom, the maximum value corresponding to the ELM boundary is the minimum point of that boundary in the depth direction.
[0034] Using this as a baseline, calculate the set of coordinate transformation parameters Shift required to flatten the ELM layer.
[0035] ② Move each A-line signal in the OCT image downwards along the depth direction, the distance of which is equal to the coordinate transformation parameter Shift at each position on the ELM boundary line. Through this image spatial transformation operation, the ELM can be positioned horizontally in the retinal OCT image. This not only significantly improves the accuracy of EZ image feature measurement but also allows for comparison of OCT images from different individuals based on the same standard, laying a solid foundation for subsequent statistical analysis and lesion feature extraction. Figure 4 The OCT images before and after ELM registration are shown.
[0036] 5) EZ Boundary Segmentation: The EZ layer appears as a bright band in OCT images, clearly distinguishable from the adjacent myoid band and OS layer (dark band). This invention uses a dynamic programming algorithm based on gray-level gradient construction, starting from the fovea of the macula and dynamically expanding point by point to both sides. Specifically, on each A-line scan line, the gray-level gradient (i.e., the gray value of the lower pixel minus the gray value of the upper pixel) from the ELM to the RPE is calculated. The highest gradient peak is located as the upper boundary of the EZ layer, and the lowest gradient valley is located as its lower boundary. Between consecutive A-lines, the EZ boundary identified at the previous point is used as the initial value to set a search window to constrain the boundary search range of the current site, thereby effectively suppressing noise interference. Finally, the identified EZ boundaries are smoothed and filtered to eliminate abnormal fluctuations. Figure 5 The OCT image shows the results of overlaying EZ layering.
[0037] (6) EZ Image Feature Quantization: Based on the ETDRS zoning standard, the EZ image features are spatially quantized by zoning, and the main radiometric indicators are as follows: 1) Thickness: The distance between the upper and lower boundaries of EZ is calculated using the following formula.
[0038] Where EZThickness is the thickness of the EZ layer. This represents the pixel position at the lower boundary of the EZ layer. denoted as the upper boundary pixel position of the EZ layer, and D as the physical length per unit pixel.
[0039] 2) Reflection Intensity: The average gray value of the EZ region is calculated using the formula shown below.
[0040] in This is the set of pixel coordinates where the EZ layer is located. Represents a set The total number of pixels in the middle. This represents the grayscale value of the corresponding pixel.
[0041] 3) Gray-scale shape ratio: Based on the gray-scale distribution curve of the EZ region, the gray-scale shape ratio is calculated using image moment analysis to quantify the light reflection intensity characteristics of the EZ region (see...). Figure 6 The process is as follows: ① Baseline Correction: To eliminate the effects of background noise or intensity drift, the extracted elliptic band (EZ) grayscale distribution curve is de-trending. Specifically, a linear baseline is constructed by connecting the two endpoints of the curve; then, the vertical distance from each point on the original curve to this baseline is calculated, and this distance is defined as the corrected grayscale value of that pixel. .
[0042] ② Calculation of the zeroth moment: The zeroth moment The total intensity of the profile is expressed by the following formula:
[0043] ③ Calculate the first moment: The first moment is used to determine the centroid of the profile. ):
[0044]
[0045] in:
[0046]
[0047] ④ Calculate the second central moment: The second central moment describes the distribution shape of the profile.
[0048]
[0049] ⑤ Calculate the eigenvalues of the covariance matrix: eigenvalues This represents the variance of the profile in the two principal directions.
[0050]
[0051] ⑥ Calculate the grayscale ratio: The grayscale ratio (GAR) is:
[0052] In summary, this invention will obtain the ETDRS spatial distribution map of the above three key image indicators. Figure 7 It shows the distribution of imaging indicators in normal individuals and under different diseases.
[0053] The embodiments described with reference to the accompanying drawings are exemplary and intended to explain the invention, and should not be construed as limiting the invention. The embodiments should not be considered as limiting the invention, but any improvements made based on the spirit of the invention should be within the scope of protection of the invention.
Claims
1. A method for analyzing the elliptic band image features of human retinal OCT images, characterized by: Includes the following steps, Step 1: Using a retinal OCT device, with the fovea of the macula as the origin, samples are taken in a radial scanning mode to obtain a high-resolution OCT image sequence of the retina; Step 2, Retinal OCT Partitioning: Based on the ETDRS partitioning standard, with the fovea as a reference, concentric rings with diameters of 1mm, 3mm and 6mm are drawn to divide the retina into the foveal region, inner region and outer region. Step 3: Retinal boundary segmentation: A two-dimensional dynamic programming algorithm is used to segment each sublayer of the retina, including at least the boundary between the outer membrane and the retinal pigment epithelium. Step 4: External membrane registration: Extract the boundary line of the external membrane and normalize it to a horizontal straight line through coordinate transformation; Step 5: Elliptic zone boundary segmentation: Based on the gray-level gradient, a dynamic programming algorithm is constructed to locate the upper and lower boundaries of the elliptic zone within the range from the outer membrane to the epithelial layer on each A-line scan line; Step 6: Elliptic zone image feature quantization: Based on ETDRS partitioning, spatial partition quantization is performed on the thickness, reflectance intensity, and grayscale morphology ratio of the elliptic zone.
2. The method for analyzing the elliptic band image features of human retinal OCT images according to claim 1, characterized in that, Step four, external membrane registration, specifically includes: extracting the lowest point in the depth direction of the external membrane boundary line as a reference line, calculating the coordinate transformation parameters of each A-line signal, and moving the external membrane layer to a horizontal position.
3. The method for analyzing the elliptic band image features of human retinal OCT images according to claim 1, characterized in that, In step five, the highest peak of the grayscale gradient is located as the upper boundary of the elliptic band, the lowest valley is located as the lower boundary of the elliptic band, and a search window is set between consecutive A-lines to constrain the boundary search range.
4. The method for analyzing the elliptic band image features of human retinal OCT images according to claim 1, characterized in that, The specific implementation process of step four is as follows: Phase 1: Automatically extract the lowest point in the depth direction of the ELM boundary line, assuming the point set of the outer membrane boundary is... Since the depth direction of an OCT image is encoded in ascending order from top to bottom, the maximum value corresponding to the ELM boundary is the minimum point of that boundary in the depth direction. Using this as a baseline, calculate the set of coordinate transformation parameters Shift required to flatten the outer membrane layer. Phase 2: Move each A-line signal in the OCT image downwards along the depth direction, the distance of which is equal to the coordinate transformation parameter Shift at each position on the ELM boundary line.
5. The method for analyzing the elliptic band image features of human retinal OCT images according to claim 1, characterized in that, The formula for calculating the thickness mentioned in step six is: Where EZThickness is the thickness of the EZ layer. This represents the pixel position at the lower boundary of the EZ layer. denoted as the upper boundary pixel position of the EZ layer, and D as the physical length per unit pixel.
6. The method for analyzing the elliptic band image features of human retinal OCT images according to claim 1, characterized in that, The formula for calculating the reflection intensity in step six is: in This is the set of pixel coordinates where the EZ layer is located. Represents a set The total number of pixels in the middle. This represents the grayscale value of the corresponding pixel.
7. The method for analyzing the elliptic band image features of human retinal OCT images according to claim 6, characterized in that, The calculation of the grayscale ratio in step six includes the following sub-steps: Sub-step 1: Baseline Correction: Connect the two endpoints of the curve to construct a linear baseline; then, calculate the vertical distance from each point on the original curve to this baseline, and define this distance as the point-corrected grayscale value of that pixel. ; Sub-step two: Calculate the zeroth moment, the formula is as follows: ； Sub-step 3: Calculate the first moment to determine the centroid ( ) in: ； Sub-step four: Calculate the second-order central moments: The second-order central moments describe the distribution shape of the profile. ； Sub-step 5: Calculate the eigenvalues of the covariance matrix: eigenvalues This represents the variance of the profile in the two principal directions. ； Sub-step 6: Calculate the grayscale ratio: The grayscale ratio (GAR) is: 。 8. The method for analyzing the elliptic band image features of human retinal OCT images according to any one of claims 1-7, characterized in that, It also includes ETDRS spatial distribution maps of generated ellipsoidal band thickness, reflection intensity, and grayscale morphology ratio.