Method for extracting edge features in optical coherence tomography image
By dividing the region of interest in OCT tomographic images and using graph search methods combined with vertical gradient weighting coefficient optimization, the problem of low edge detection accuracy in existing technologies is solved, and efficient and accurate edge feature extraction is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-18
- Publication Date
- 2026-03-27
AI Technical Summary
Existing OCT tomographic image segmentation techniques require a large amount of training data and computational resources when detecting the edges of layered structures. Graph search methods have low segmentation accuracy for complex structures, especially when the edges are complex.
The region of interest in OCT tomographic images is delineated using the peak search method, and the minimum cumulative cost path is found within a limited range using the graph search method. Combined with the differentiated setting of vertical gradient weighting coefficients, the upper and lower surface boundaries are accurately segmented.
It improves the accuracy and computational efficiency of edge feature detection, reduces the computational load, and enhances the reliability of edge segmentation, especially significantly improving the segmentation accuracy of complex structures.
Smart Images
Figure CN121746728A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of electronic information and image processing, and in particular to a method for extracting edge features from optical coherence tomography images. Background Technology
[0002] Optical coherence tomography (OCT) is a tomographic imaging technique with resolution down to the micrometer level. It has been widely used in ophthalmology for clinical imaging and diagnosis of the cornea and retina, as well as in industrial non-destructive testing for the detection and quality control of layered structures such as tablet coatings and automotive coatings. In these applications, image segmentation algorithms are required to detect the boundary features of layered structures in OCT tomographic images, thereby accurately calculating the key characteristics of the sample and identifying defects.
[0003] Current OCT tomographic image segmentation techniques can be broadly categorized into two types: deep learning-based semantic segmentation and graph search methods. Deep learning-based edge detection algorithms can accurately detect boundaries in OCT tomographic images. However, for different types of samples, it typically requires pre-measuring thousands of OCT tomographic images and expending significant manpower to create training sets. Simultaneously, model training consumes substantial computational resources and time. On the other hand, graph search methods usually first utilize edge detection operators (such as gradient operators and Sobel operators) to enhance the edges of layered structures, and then search for the minimum cumulative cost in the OCT tomographic image to extract boundary features. Compared to deep learning-based edge detection, graph search methods are simpler and faster to deploy, but their edge segmentation accuracy depends heavily on the quality of the OCT tomographic image and the complexity of the sample structure. For samples with gently changing layered edges, the segmentation accuracy is higher; conversely, for structures with complex edge variations, the segmentation accuracy drops significantly. Summary of the Invention
[0004] To address the shortcomings of existing technologies, this invention provides a method for extracting edge features from optical coherence tomography (OCT) images. By dividing the regions of interest (ROIs) of each edge of the layered structure in the OCT image into regions of interest (ROIs), the approximate ROIs of different boundaries can be identified in advance, reducing the complexity of boundary changes, decreasing the axial search range of the cumulative cost path of adjacent A-scans in the graph search method, and improving the accuracy of sample boundary feature detection in OCT images.
[0005] To achieve the above objectives, the present invention provides the following technical solution: A method for extracting edge features from optical coherence tomography images includes the following steps: S1: The peak position of the surface peak of the layered structure in each A scan of the optical coherence tomography (OCT) image is estimated by using the peak search method; S2: The estimated surface peak position is shifted to obtain the region of interest of the layered structure surface boundary in the OCT tomographic image; S3: Use graph search to find the path with the minimum cumulative cost within the region of interest of the surface boundary as the surface boundary for precise segmentation; S4: Based on the estimated thickness of the layered structure, the precisely segmented surface boundary is translated to obtain the region of interest of the lower surface boundary in the OCT tomographic image; S5: Use a graph search method to find the path with the minimum cumulative cost within the region of interest of the lower surface boundary as the precise segmentation of the lower surface boundary.
[0006] The present invention further defines the technical solution as follows: Preferably, step S1 specifically includes the following sub-steps: S11: Perform a backward difference on each A scan in the OCT tomographic image to obtain Diff (n) = A (n+1) - A (n), where A (n) is the value of the nth data point in the A scan; S12: In each A scan, select all data point positions that satisfy Diff (n)>0 and Diff (n+1)<0 as peak values, and sort them to obtain the data point positions of the maximum and minimum peak values in the A scan and the corresponding A scan data values; S13: Calculate the arithmetic mean of the maximum and minimum peak values and use it as the screening threshold for the true reflection peak value; S14: In each A scan, peak values greater than the screening threshold are recorded as true reflectance peak values, and peak values less than the screening threshold are recorded as noise. S15: Among all the true reflection peaks in each A scan, select the location of the true reflection peak with the smallest axial position as the estimated surface position.
[0007] Preferably, step S2 specifically includes the following sub-steps: S21: In each A-scan, the estimated surface position is shifted forward by 1 / 3 of the estimated thickness to obtain the upper limit position of the region of interest of the upper surface boundary in that A-scan; S22: In each A-scan, the estimated surface position is shifted backward by 1 / 3 of the estimated thickness to obtain the lower limit position of the region of interest of the upper surface boundary in that A-scan; S23: In each A scan, the data points between the upper limit position and the lower limit position are extracted as the region of interest of the upper surface boundary in that A scan. The regions of interest of the upper surface boundary of all A scans are arranged in parallel to obtain the region of interest of the upper surface boundary in the OCT tomographic image.
[0008] Preferably, step S3 specifically includes the following sub-steps: S31: For all pixels in the region of interest on the upper surface boundary, calculate the cost function C for each pixel; S32: Calculate the cumulative cost of the tomographic image from left to right for the region of interest on the upper surface boundary based on the cost function C for each pixel. S33: The left-to-right pixel path that the minimum cumulative search cost traverses within the region of interest on the upper surface boundary is used as the upper surface boundary for precise segmentation.
[0009] Preferably, step S4 specifically includes the following sub-steps: S41: In each A-scan, the precise upper surface boundary position is shifted backward by 2 / 3 of the estimated thickness to obtain the upper limit position of the region of interest of the lower surface boundary in that A-scan; S42: In each A-scan, the precise upper surface boundary position is shifted backward by 4 / 3 of the estimated thickness to obtain the lower limit position of the region of interest of the lower surface boundary in that A-scan; S43: In each A scan, the data points between the upper limit position and the lower limit position of the region of interest of the lower surface boundary are extracted as the region of interest of the lower surface boundary in that A scan; S44: Arrange the regions of interest (ROIs) of the lower surface boundary of all A scans in parallel to obtain the ROI of the lower surface boundary in the OCT tomographic image.
[0010] Preferably, step S5 specifically includes the following sub-steps: S51: For all pixels in the region of interest at the lower surface boundary, calculate the cost function C for each pixel; S52: Calculate the cumulative cost of the tomographic image from left to right for the region of interest at the lower surface boundary based on the cost function C for each pixel. S53: The minimum cumulative search cost is obtained by finding the left-to-right pixel path traversed by the region of interest on the lower surface boundary, and this path is used as the lower surface boundary for precise segmentation.
[0011] Preferably, the formula for calculating the cost function C is: ; Where I(x,z) represents the pixel gray value of the x-th row and z-th column in the two-dimensional region of interest, ω1 is the weighting coefficient of the gradient in the horizontal direction, and ω2 is the weighting coefficient of the gradient in the vertical direction; when calculating the cost function of the region of interest at the upper surface boundary, ω2 is a positive number; when calculating the cost function of the region of interest at the lower surface boundary, ω2 is a negative value; and |ω2|>5|ω1|.
[0012] Preferably, the cumulative cost calculation method is as follows: ; Row max This represents the maximum number of rows in the two-dimensional region of interest.
[0013] Beneficial effects Compared with existing technologies, it has the following advantages: This invention finds the RoIs of each boundary in OCT tomographic images, making the boundary changes in the layered structure of the sample smoother. The cumulative cost of the graph search method in adjacent A scans is reduced, and the accuracy of boundary detection is improved. This invention determines the true reflection peak through peak search and threshold screening, accurately divides the region of interest (RoI) of each boundary, and limits the edge detection range of complex layered structures to a preset effective area, avoiding interference from irrelevant areas. This makes the originally complex boundary tend to be smooth within the RoI, greatly improving the accuracy of edge feature extraction and solving the pain point of low accuracy of existing graph search methods for segmenting complex edge structures. The precise RoI segmentation of this invention narrows the axial search range of the cumulative cost path of adjacent A scans during the graph search process, eliminating the need for a full-domain search of the entire image. This reduces computational load while maintaining detection accuracy. Furthermore, this invention differentiates the positive and negative attributes of the vertical gradient weighting coefficients based on the grayscale change characteristics of the upper and lower surface boundaries and strengthens the weight ratio of the vertical gradient, making the cost function more closely match the physical characteristics of the edges. This further improves the reliability of accurate segmentation and effectively distinguishes different boundary types of transitions between bright and dark regions. Attached Figure Description
[0014] Figure 1 This is a flowchart illustrating a method for extracting edge features of layered structures from optical coherence tomography images, as provided in the example. Figure 2 Two-dimensional tablet coating tomographic images obtained by optical coherence tomography (OCT). Figure 3 A scan is an A-scan in a two-dimensional tablet coating tomographic image. Figure 4 Images showing the region of interest (RoI) segmentation of the upper and lower boundaries of the tablet coating; Figure 5 Images of the upper and lower boundaries of the tablet coating extracted after implementing the method of the present invention. Detailed Implementation
[0015] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Example 1
[0016] This embodiment provides a method for extracting edge features from optical coherence tomography images, including the following steps: S1: The estimated location of the peak on the surface of the layered structure in each A scan of the optical coherence tomography (OCT) image is obtained by using the peak search method. S11: Perform a backward difference on each A scan in the OCT tomographic image to obtain Diff (n) = A (n+1) - A (n), where A (n) is the value of the nth data point in the A scan; S12: In each A scan, select all data point positions that satisfy Diff (n)>0 and Diff (n+1)<0 as peak values, and sort them to obtain the data point positions of the maximum and minimum peak values in the A scan and the corresponding A scan data values; S13: Calculate the arithmetic mean of the maximum and minimum peak values and use it as the screening threshold for the true reflection peak value; S14: In each A scan, peak values greater than the screening threshold are recorded as true reflectance peak values, and peak values less than the screening threshold are recorded as noise. S15: Among all the true reflection peaks in each A scan, select the location of the true reflection peak with the smallest axial position as the estimated surface location. S2: The estimated surface peak position is shifted to obtain the region of interest of the layered structure surface boundary in the OCT tomographic image, specifically: S21: In each A-scan, the estimated surface position is shifted forward by 1 / 3 of the estimated thickness to obtain the upper limit position of the region of interest of the upper surface boundary in that A-scan; S22: In each A-scan, the estimated surface position is shifted backward by 1 / 3 of the estimated thickness to obtain the lower limit position of the region of interest of the upper surface boundary in that A-scan; S23: In each A scan, the data points between the upper limit position and the lower limit position are extracted as the region of interest of the upper surface boundary in the A scan. The regions of interest of the upper surface boundary of all A scans are arranged in parallel to obtain the region of interest of the upper surface boundary in the OCT tomographic image. S3: The path traversed by the minimum cumulative cost within the region of interest of the surface boundary is found using a graph search method, which serves as the precise surface boundary segmentation. Specifically: S31: For all pixels in the region of interest on the upper surface boundary, calculate the cost function C for each pixel; S32: Calculate the cumulative cost of the tomographic image from left to right for the region of interest on the upper surface boundary based on the cost function C for each pixel. S33: The left-to-right pixel path that the minimum cumulative search cost is traversed in the region of interest on the upper surface boundary, and this path is used as the upper surface boundary for precise segmentation. S4: Based on the estimated thickness of the layered structure, the precisely segmented surface boundary is translated to obtain the region of interest of the lower surface boundary in the OCT tomographic image, specifically: S41: In each A-scan, the precise upper surface boundary position is shifted backward by 2 / 3 of the estimated thickness to obtain the upper limit position of the region of interest of the lower surface boundary in that A-scan; S42: In each A-scan, the precise upper surface boundary position is shifted backward by 4 / 3 of the estimated thickness to obtain the lower limit position of the region of interest of the lower surface boundary in that A-scan; S43: In each A scan, the data points between the upper limit position and the lower limit position of the region of interest of the lower surface boundary are extracted as the region of interest of the lower surface boundary in that A scan; S44: Arrange the regions of interest (ROIs) of the lower surface boundary of all A scans in parallel to obtain the ROIs of the lower surface boundary in the OCT tomographic image; S5: The path traversed by the minimum cumulative cost within the region of interest of the lower surface boundary is found using a graph search method, which serves as the precise segmentation boundary of the lower surface. Specifically: S51: For all pixels in the region of interest at the lower surface boundary, calculate the cost function C for each pixel; S52: Calculate the cumulative cost of the tomographic image from left to right for the region of interest at the lower surface boundary based on the cost function C for each pixel. S53: The minimum cumulative search cost is obtained by finding the left-to-right pixel path traversed by the region of interest on the lower surface boundary, and this path is used as the lower surface boundary for precise segmentation.
[0017] The formula for calculating the cost function C above is: ; Where I(x,z) represents the pixel gray value of the x-th row and z-th column in the two-dimensional region of interest, ω1 is the weighting coefficient of the gradient in the horizontal direction, and ω2 is the weighting coefficient of the gradient in the vertical direction; when calculating the cost function of the region of interest at the upper surface boundary, ω2 is a positive number; when calculating the cost function of the region of interest at the lower surface boundary, ω2 is a negative value; and |ω2|>5|ω1|.
[0018] The above cumulative cost calculation method is as follows: ; Row max This represents the maximum number of rows in the two-dimensional region of interest.
[0019] Example 2 This embodiment determines the surface RoI by analyzing the peak values in a single A-scan, such as... Figure 3 The positions marked with red asterisks represent all peaks located in each A-scan by the peak search algorithm. Then, all marked peaks in the A-scan are compared with a pre-defined threshold. If a peak is greater than the threshold, it is considered a genuine scattering signal originating from the upper or lower boundaries of the coating, or from particles and bubbles within the coating layer; otherwise, it is considered noise. This threshold is actually an empirical value; after several trials, we set it as the average of the maximum and minimum peaks in the A-scan during processing. After determining all genuine scattering peaks, as shown... Figure 3 As shown by the solid blue circle, we take the position of the foremost true scattering peak as the location of the tablet surface, and define the range before and after this estimated surface location as the RoI of the upper coating boundary. When determining the lower coating boundary RoI, we first set a maximum and minimum expected coating thickness D. max and D min Then add D to the estimated surface position. max and D min The range is used as the RoI of the lower boundary of the coating. Figure 4 The red and blue areas shown are the RoIs of the upper and lower boundaries of the tablet coating.
[0020] To accurately segment the contours of the upper and lower boundaries of the coating layer, we applied a graph search method to the two RoIs. This method finds the shortest path from the left to the right side of the tomographic image by minimizing the cumulative cost T(x,z) of pixel (x,z) in the image. The cumulative cost function T(x,z) is defined as follows: For each pixel, we typically consider the seven neighboring pixels in the previous column to find the minimum cumulative cost. The cost C(x, z) of each pixel (x, z) is defined as a weighted sum of the vertical and horizontal gradients, where w1 and w2 are the weighting coefficients for the vertical and horizontal gradients. Considering that the upper and lower boundaries of the coating in RoI are almost horizontally straightened, the contribution of the horizontal gradient to the cumulative cost evaluation is much smaller than that of the vertical gradient. Therefore, the horizontal weighting coefficient w2 should be much smaller than the vertical weighting coefficient w1. Furthermore, it is obvious that... Figure 2 The upper boundary is the edge from the bright area to the dark area, and the lower boundary is the edge from the dark area to the bright area. Therefore, the lower boundary segmentation should set the horizontal gradient weight w1 to a negative value. Figure 5 The upper boundary contour of the coating layer is shown when w1=1 and w2=0.1, and the lower boundary contour of the coating layer is shown when w1=-1 and w2=0.1.
[0021] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.
[0022] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.
Claims
1. A method for extracting edge features from optical coherence tomography images, characterized in that, Includes the following steps: S1: The peak position of the surface peak of the layered structure in each A scan of the optical coherence tomography (OCT) image is estimated by using the peak search method; S2: The estimated surface peak position is shifted to obtain the region of interest of the layered structure surface boundary in the OCT tomographic image; S3: Use graph search to find the path with the minimum cumulative cost within the region of interest of the surface boundary as the surface boundary for precise segmentation; S4: Based on the estimated thickness of the layered structure, the precisely segmented surface boundary is translated to obtain the region of interest of the lower surface boundary in the OCT tomographic image; S5: Use a graph search method to find the path with the minimum cumulative cost within the region of interest of the lower surface boundary as the precise segmentation of the lower surface boundary.
2. The method for assessing and predicting the top-pass safety of unmanned vehicles according to claim 1, characterized in that, Step S1 specifically includes the following sub-steps: S11: Perform a backward difference on each A scan in the OCT tomographic image to obtain Diff (n) = A (n+1) - A(n), where A (n) is the value of the nth data point in the A scan; S12: In each A scan, select all data point positions that satisfy Diff (n)>0 and Diff (n+1)<0 as peak values, and sort them to obtain the data point positions of the maximum and minimum peak values in the A scan and the corresponding A scan data values; S13: Calculate the arithmetic mean of the maximum and minimum peak values and use it as the screening threshold for the true reflection peak value; S14: In each A scan, peak values greater than the screening threshold are recorded as true reflectance peak values, and peak values less than the screening threshold are recorded as noise. S15: Among all the true reflection peaks in each A scan, select the location of the true reflection peak with the smallest axial position as the estimated surface position.
3. The method for extracting edge features from optical coherence tomography images according to claim 2, characterized in that, Step S2 specifically includes the following sub-steps: S21: In each A-scan, the estimated surface position is shifted forward by 1 / 3 of the estimated thickness to obtain the upper limit position of the region of interest of the upper surface boundary in that A-scan; S22: In each A-scan, the estimated surface position is shifted backward by 1 / 3 of the estimated thickness to obtain the lower limit position of the region of interest of the upper surface boundary in that A-scan; S23: In each A scan, the data points between the upper limit position and the lower limit position are extracted as the region of interest of the upper surface boundary in that A scan. The regions of interest of the upper surface boundary of all A scans are arranged in parallel to obtain the region of interest of the upper surface boundary in the OCT tomographic image.
4. The method for extracting edge features from optical coherence tomography images according to claim 3, characterized in that, Step S3 specifically includes the following sub-steps: S31: For all pixels in the region of interest on the upper surface boundary, calculate the cost function C for each pixel; S32: Calculate the cumulative cost of the tomographic image from left to right for the region of interest on the upper surface boundary based on the cost function C for each pixel. S33: The left-to-right pixel path that the minimum cumulative search cost traverses within the region of interest on the upper surface boundary is used as the upper surface boundary for precise segmentation.
5. The method for extracting edge features from optical coherence tomography images according to claim 4, characterized in that, Step S4 specifically includes the following sub-steps: S41: In each A-scan, the precise upper surface boundary position is shifted backward by 2 / 3 of the estimated thickness to obtain the upper limit position of the region of interest of the lower surface boundary in that A-scan; S42: In each A-scan, the precise upper surface boundary position is shifted backward by 4 / 3 of the estimated thickness to obtain the lower limit position of the region of interest of the lower surface boundary in that A-scan; S43: In each A scan, the data points between the upper limit position and the lower limit position of the region of interest of the lower surface boundary are extracted as the region of interest of the lower surface boundary in that A scan; S44: Arrange the regions of interest (ROIs) of the lower surface boundary of all A scans in parallel to obtain the ROI of the lower surface boundary in the OCT tomographic image.
6. The method for extracting edge features from optical coherence tomography images according to claim 5, characterized in that, Step S5 specifically includes the following sub-steps: S51: For all pixels in the region of interest at the lower surface boundary, calculate the cost function C for each pixel; S52: Calculate the cumulative cost of the tomographic image from left to right for the region of interest at the lower surface boundary based on the cost function C for each pixel. S53: The minimum cumulative search cost is obtained by finding the left-to-right pixel path traversed by the region of interest on the lower surface boundary, and this path is used as the lower surface boundary for precise segmentation.
7. A method for extracting edge features from optical coherence tomography images according to claim 4 or 6, characterized in that, The formula for calculating the cost function C is as follows: ; Where I(x,z) represents the pixel gray value of the x-th row and z-th column in the two-dimensional region of interest, ω1 is the weighting coefficient of the gradient in the horizontal direction, and ω2 is the weighting coefficient of the gradient in the vertical direction; when calculating the cost function of the region of interest at the upper surface boundary, ω2 is a positive number; when calculating the cost function of the region of interest at the lower surface boundary, ω2 is a negative value; and |ω2|>5|ω1|.
8. A method for extracting edge features from an optical coherence tomography image according to claim 4 or 6, characterized in that, The method for calculating the cumulative cost is as follows: ; Row max This represents the maximum number of rows in the two-dimensional region of interest.