Schlemm's canal segmentation and quantification method based on ultrasound bio-microscopy images

Abstract

The application discloses a kind of based on ultrasound biomicroscopy image's schlemm canal segmentation and quantification method, based on the improved fuzzy C means clustering algorithm (Fuzzy C-Means, FCM) in ultrasound biomicroscopy (Ultrasound Bio-microscopy, for short UBM) image schlemm canal segmentation and quantification method, specially for linear scanning mode obtained ultrasound biomicroscopy image, by extracting anterior chamber angle region in UBM image, then by an improved FCM segmentation method schlemm canal in anterior chamber angle region is segmented, finally output binary image and quantification result.The application embodiment can effectively segment schlemm canal in ultrasound biomicroscopy image, the application improves the identification precision and segmentation efficiency of schlemm canal in UBM image, is not influenced by the refraction interval transparency state of examinee, and the schlemm canal quantification result provided is favorable to clinical diagnosis.

Description

Technical Field

[0001] This invention relates to a segmentation and quantization technique for ultrasound biomicroscopy (UBM) images, and more particularly to a Schlem tube segmentation and quantization method based on ultrasound biomicroscopy images. Background Technology

[0002] Glaucoma is the leading cause of irreversible blindness worldwide, with an estimated 110 million glaucoma patients globally by 2040. Elevated intraocular pressure (IOP) is a significant factor in the progression of glaucoma. Research shows that elevated IOP leads to a reduction in the Schlem's canal region of the eye, indicating that structural changes in the Schlem's canal play a crucial role in IOP regulation and glaucoma pathogenesis. Therefore, the Schlem's canal has become an important research subject in the field of glaucoma. Accurate identification and quantification of the Schlem's canal region in medical images are essential for improving computer-aided diagnosis and reducing the probability of misdiagnosis, and also represent a current challenge in clinical practice.

[0003] Researchers can manually delineate the boundaries of Schlemm's canal region in the anterior chamber angle of optical coherence tomography (OCT) images and UBM images using open-source image processing software, and perform quantitative analysis. This has made a positive contribution to elucidating the pathogenesis and treatment effects of glaucoma. Although manual segmentation has the advantage of high precision, its segmentation efficiency is low and it is affected by factors such as the doctor's energy and emotional fluctuations, resulting in poor repeatability and low efficiency. Existing Schlemm's canal segmentation methods based on improved level sets, such as those described in the literature (Wang X, Zhai Y, Liu X, et al. Level-set method for image analysis of Schlemm's canal and trabecular meshwork[J]. Translational Vision Science & Technology, 2020, 9(10):7-7), can effectively extract the Schlemm's canal region and has high segmentation efficiency and strong repeatability. However, this method only marks the boundary of the Schlemm's canal region and does not directly quantify the Schlemm's canal region, which is not conducive to further analysis and clinical diagnosis by doctors. In addition, some research teams have used deep learning technology to segment Schlemm's canals in mouse OCT images, as described in the literature (Choy KC, Li G, Stamer WD, et al. Open-source deep learning-based automatic segmentation of mouse Schlemm's canal in optical coherence tomography images[J]. Experimental eyeresearch,2022,214:108844). However, deep learning technology requires a large number of training samples and a long pre-training time, resulting in low efficiency. Therefore, more efficient and intuitive methods are needed to automatically segment and quantize Schlemm's canal regions.

[0004] To accurately identify and quantify Schlem's canal regions, improve computer-aided diagnosis, and reduce the probability of misdiagnosis, it is necessary to explore more efficient image segmentation algorithms for segmenting Schlem's canal regions in UBM images, and ensure that the segmentation results can be intuitively quantified. This is of great significance for large-scale studies on the occurrence and development of glaucoma patients. Summary of the Invention

[0005] (a) Technical problems to be solved

[0006] To achieve computer-aided segmentation and intuitive quantization of Schlem tubes in UBM images, this invention proposes a method for Schlem tube segmentation and quantization based on ultrasonic biological microscope images. This method uses an improved FCM algorithm to achieve computer-aided segmentation of Schlem tubes in UBM images. By extracting the boundary curves of the Schlem tubes in the UBM images and outputting the number of pixels in the Schlem tube region, the quantization of the Schlem tubes is achieved.

[0007] (II) Technical Solution

[0008] This invention proposes a Schlemma tube segmentation and quantization method based on ultrasound biological microscope images. This method is a segmentation and quantization method based on an improved FCM algorithm, and includes the following steps:

[0009] Step 1: Acquire ultrasound biomicroscopy images from multiple orientations, select appropriate ultrasound biomicroscopy images and extract the original UBM images containing Schlem tubes, perform histogram clustering on them, and initialize the membership matrix of the original UBM images according to the number of clusters.

[0010] Step 2: Perform morphological closure reconstruction, i.e., MCR operation, on the original UBM image to obtain statistical information of the reconstructed UBM image as prior knowledge;

[0011] Step 3: Add the prior knowledge from step (2) to the objective function of the FCM algorithm, and iteratively solve the objective function to obtain the membership partition matrix of the MCR-reconstructed UBM image;

[0012] Step 4: Use the membership partition matrix from step (3) as prior information to fill the membership partition matrix of the original UBM image, obtain the filled membership partition matrix, perform median filtering on the filled membership partition matrix, obtain the filtered membership partition matrix, which is the updated membership partition matrix of the original UBM image, and use this membership partition matrix to segment the original UBM image.

[0013] Step 5: Using the gradient edge detection operator, the boundary curves of each sub-region of the original UBM image segmentation result are binarized to visually display the Schlem tube in the anterior chamber angle region and output the quantization result.

[0014] As a further improvement to this technical solution:

[0015] In step 1, the selected UBM image is converted into a grayscale image, the anterior chamber angle region containing the Schlem tube is extracted, and based on the histogram information of the original UBM image X, the number of peaks c in the histogram is counted, and the grayscale values ​​V = {v1, v2, ..., v} corresponding to all peaks are calculated. cLet} be the initial cluster center, and initialize the membership matrix of the original UBM image as U′=zeros(N,c), where N is the total number of pixels in the original UBM image.

[0016] In step 2, structural elements are used. The formula used for MCR reconstruction of UBM image X is:

[0017]

[0018] Where ε represents the erosion operation, δ represents the dilation operation, and ε K (X) indicates that K is used to perform an erosion operation on X; Indicates X relative to ε K Morphological dilatation reconstruction results of (X); Indicates the use of K pairs Perform an expansion operation; express Compared to The morphological corrosion reconstruction results.

[0019] In step 2, obtaining the statistical information of the reconstructed UBM image includes gray levels {z1, z2, ..., z...} i ,...,z q}, and the number of pixels corresponding to each gray level {a1, a2, ..., a i ,...,a q},satisfy q represents the number of gray levels in the reconstructed UBM image, and N represents the total number of pixels in the original UBM image. q is much smaller than N.

[0020] In step 3, the iterative solution of the objective function is expressed by the following formula:

[0021]

[0022] Among them, u ij This indicates the grayscale value z of the MCR-reconstructed UBM image. i Clustering centers v of the original UBM image j The fuzzy membership degree, called the fuzzy index, determines the degree of fuzziness in the classification result, and ||·|| represents the Euclidean distance. The membership degree partitioning matrix is ​​obtained by solving the objective function (1).

[0023] In step 4, the membership matrix of the UBM image is reconstructed based on MCR, and the membership matrix of the original UBM image is filled with the gray values ​​x corresponding to the original UBM image X. k For cluster center v j The fuzzy membership degree is obtained by the following formula:

[0024]

[0025] The membership partitioning matrix U′ = [u] corresponding to the original UBM image X is obtained. kj ] N×c .

[0026] In step 4, a 5×5 median filter is used to filter the membership partition matrix U′ of the padded original UBM image to obtain a membership partition matrix U″ suitable for the original UBM image. The median filter used is denoted as:

[0027] U″(k,j)=med{U′(s,t)|k-2≤s≤k+2,j-2≤t≤j+2} (3)

[0028] In step 4, the original UBM image X is divided into several sub-regions according to the obtained membership degree partitioning matrix U″.

[0029] In step 5, the boundary curves of each sub-region of the original UBM image segmentation result are binarized using the gradient edge detection operator, calculated as follows: and The edge detection operator performs a convolution operation on the segmentation results of the UBM image, through... Calculate the gradient of each pixel. If the gradient G is greater than a set threshold, the pixel is considered a boundary point.

[0030] In step 5, the regionprops function in Matlab is used to find the area of ​​each region and output the quantified area result of the Schlem tube region.

[0031] (III) Beneficial Effects

[0032] The proposed method for segmentation and quantization of Schlem's tubes in ultrasound microscope images based on improved FCM has the following characteristics: First, in step 2, MCR operation on the UBM image effectively reduces the impact of noise on the segmentation result while enhancing the image's detail preservation ability. Second, in step 3, the gray-level statistical information of the MCR-reconstructed UBM image is introduced into the objective function of the FCM algorithm as a prior constraint, thereby reducing the algorithm's time complexity and improving segmentation efficiency. Third, in step 4, median filtering is used to optimize the membership partition matrix to improve the accuracy and robustness of UBM image segmentation. Finally, in step 5, the quantization results of the boundary curve and area of ​​the Schlem's tube are visually displayed, achieving accurate quantization of the Schlem's tube and providing strong support for the diagnosis and efficacy evaluation of glaucoma patients. The application of this method will help improve the early diagnosis and treatment of glaucoma. Attached Figure Description

[0033] Figure 1 This is a flowchart of the Schlem tube segmentation and quantization method based on ultrasonic biological microscope images of the present invention.

[0034] Figure 2 This is the segmentation result of Schlem's tubes in the ultrasonic biological microscope image of the present invention.

[0035] Figure 3 The image and quantization result of the Schlem tube of the present invention are shown. Detailed Implementation

[0036] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical means adopted by the present invention will be further described below in conjunction with specific embodiments and with reference to the accompanying drawings.

[0037] Based on the Schlem tube segmentation and quantization method of ultrasonic biological microscope images, this invention is specifically for images acquired by ultrasonic biological microscope imaging systems in linear scanning mode.

[0038] like Figure 1 As shown, the method includes:

[0039] Step 1: Acquire ultrasound biomicroscopy images from multiple orientations, select suitable ultrasound biomicroscopy images and extract the raw UBM image containing the Schlem's tube, perform histogram clustering on them, and initialize the membership matrix of the raw UBM image according to the number of clusters; that is, acquire ultrasound biomicroscopy images from different orientations, select suitable images, extract the anterior chamber angle region, and then convert this region into a grayscale image, finally obtaining a raw UBM image containing the Schlem's tube X={x1,x2,…,x k ,...,x N}, where N is the number of pixels in the UBM image; the UBM image is clustered based on the number of peaks c in the gray value histogram of X, and the gray values ​​V = {v1, v2, ..., v} corresponding to all peaks are clustered. c Let} be the initial cluster center, and initialize the membership matrix of the original UBM image as U′=zeros(N,c).

[0040] Step 2: Perform morphological closure reconstruction (MCR) on the original UBM image from Step 1 to obtain statistical information of the reconstructed UBM image as prior knowledge; that is, perform morphological closure reconstruction (MCR) on the UBM image X to suppress different types and intensities of noise in the UBM image X, and obtain the MCR reconstructed image Z, where the gray levels of Z are denoted as {z1, z2, ..., z...}.i ,...,z q}, where q is the number of gray levels in Z, and the number of pixels corresponding to each gray level in Z is denoted as {a1, a2, ..., a}. i ,...,a q},satisfy

[0041] Grayscale image morphology includes two basic operations: dilation (δ) and erosion (ε). Using structuring elements... Scan each pixel of the UBM image X, calculate the maximum value of the pixels in the UBM image region covered by the structuring element K, and assign the maximum value to the pixel in the UBM image X corresponding to the center element of K, thus obtaining the dilated image Y. d =δ K (X), Y d The gray value at position (s,t) is expressed by the following formula.

[0042]

[0043] Calculate the minimum value of the pixels in the UBM image region covered by the structuring element K, and assign the minimum value to the pixel of the UBM image X corresponding to the center element of K, to obtain the image Y after the erosion operation. e =ε K (X), Y e The gray value at position (s,t) is expressed by the following formula.

[0044]

[0045] According to Y e and Y d The definition of Y is obtained e ≤X,Y d ≥X.

[0046] Morphological dilatation reconstruction is defined as satisfy ∧ represents a pixel-by-pixel comparison between two images, taking the smaller value. Morphological erosion reconstruction is defined as... satisfy The symbol ∨ indicates that the pixel-by-pixel values ​​of the two images are compared and the larger value is selected.

[0047] Morphological reconstruction of MCR is defined in four steps:

[0048] First, an erosion operation is performed on the input UBM image X to obtain Y. e =ε K (X),

[0049] Secondly, using X and erosion image Y e Morphological dilation reconstruction operation was performed to obtain

[0050] Secondly, regarding the morphological dilation reconstruction results Perform dilation operation to obtain

[0051] Finally, the second step generates morphological dilation reconstruction. The third step generates an expansion operation. Morphological erosion reconstruction is performed to obtain the morphological closure reconstruction result of the original image X.

[0052] By performing MCR on the input UBM image, a morphologically reconstructed UBM image Z is obtained, which is abbreviated as:

[0053] Z = R C (X) (3)

[0054] And let the gray levels of Z be denoted as {z1, z2, ..., z i ,...,z q}, where q is the number of gray levels in Z, and the number of pixels corresponding to each gray level in Z is denoted as {a1, a2, ..., a}. i ,…,ai,…,aq,a q},satisfy

[0055] Step 3: Incorporate the prior knowledge from Step 2 into the objective function of the FCM algorithm, and iteratively solve the objective function to obtain the membership matrix of the MCR-reconstructed UBM image; that is, introduce the local spatial gray-level statistical information of the Z-axis of the MCR-reconstructed UBM image as a prior constraint into the objective function of the FCM algorithm to improve the robustness of the FCM algorithm. The objective function of the FCM algorithm improved by the spatial gray-level statistical information of the MCR-reconstructed image is expressed by the following formula:

[0056]

[0057] Where u ij This indicates the grayscale value z of the MCR-reconstructed UBM image. i Clustering centers v of the original UBM image j The fuzzy membership degree. Using the Lagrange multiplier method, the optimization problem of objective function (4) can be transformed into an unconstrained optimization problem minimizing the following objective function:

[0058]

[0059] Where λ is the Lagrange multiplier, the membership degree u is obtained by minimizing the objective function (5). ij and class center v j Calculation formula

[0060]

[0061]

[0062] According to formula (6), the membership degree partitioning matrix U = [u ij ] q×c By iteratively calculating formulas (6) and (7), the membership partition matrix U is continuously updated until max(U) is satisfied. (r) -U (r-1) ) < e, where e is the given minimum error threshold and r is the maximum number of iterations. The grayscale value z of the UBM image reconstructed by MCR is... i For the original UBM image X, the cluster center v j Fuzzy membership degree.

[0063] Step 4: Use the membership matrix from Step 3 as prior information to fill the membership matrix of the original UBM image. Perform median filtering on the filled membership matrix to obtain a more reasonable membership matrix for the original UBM image, and use this matrix to segment the original UBM image; specifically, the grayscale value x of the original UBM image X... k For cluster center v j The fuzzy membership degree is obtained by the following formula:

[0064]

[0065] The membership partitioning matrix U′ = [u] corresponding to the original UBM image X is obtained. kj ] N×c The partition matrix U′ is corrected using a 5×5 median filter, avoiding the calculation of distances between pixels and cluster centers in local spatial neighborhoods, reducing the computational complexity of the algorithm, and enhancing its performance. The median filter used is denoted as:

[0066] U″(k,j)=med{U′(s,t)|k-2≤s≤k+2,j-2≤t≤j+2}. (9)

[0067] The original UBM image X is divided into several sub-regions using the membership matrix U″.

[0068] Step 5: Divide the original UBM image X in Step 4 into several sub-regions, use the edge detection operator to binarize the boundary curve of each region, visually display the Schlem tube in the anterior chamber angle region, and output the quantization result.

[0069] To facilitate a clear and intuitive display of the boundary of the Schlem tube in the anterior chamber angle region, this invention employs an edge detection operator to binarize the boundary curve of the Schlem tube, i.e., using... and Two weighted matrices are convolved with the UBM image to calculate estimates of the gray-level gradients in the horizontal and vertical directions, |G|. x | and | G y |. For each pixel in the UBM image, through Calculate its gradient. If the gradient G is greater than a set threshold η, then the point is considered a boundary point.

[0070] Based on the above steps, the specific computational flow of the proposed algorithm is as follows:

[0071] Step 1: Input a UBM grayscale image X containing a Schlem tube, a structuring element K, and randomly initialize the membership partition matrix U. (0) Given the minimum error threshold e, the gradient threshold η, and the loop counter r = 0;

[0072] Step 2: Based on the grayscale histogram of the UBM grayscale image X, cluster the UBM image into c classes, with the initial cluster centers denoted as V = {v1, v2, ..., v...} c}

[0073] Step 3: Calculate the MCR reconstructed image Z using formula (3), and calculate the gray-level statistical information of Z;

[0074] Step 4: Update the cluster center V according to formula (7);

[0075] Step 5: Update the partition matrix U of image Z according to formula (6). (r+1) ;

[0076] Step 6: Check if the iteration termination condition max(U) has been met. (r) -U (r-1) If r < e, stop the iteration and output the result; otherwise, let r = r + 1 and repeat steps 4 to 5.

[0077] Step 7: Calculate the partition matrix U″ of the original UBM image X according to formulas (8) and (9), and segment the original UBM image according to U″;

[0078] Step 8: Binarize the boundary curves of the segmentation results of UBM image X to visually display the Schlem tube in the anterior chamber angle region and quantify its area.

[0079] To provide a clearer explanation of the method for segmenting Schlem tubes in the above-mentioned ultrasonic biological microscope images, a specific embodiment is described below. However, it is worth noting that this embodiment is only for better illustrating the present invention and does not constitute an improper limitation of the present invention.

[0080] Taking the human eye as an example, firstly, the Suoer SW-3200L ultrasonic biological microscope was used to acquire ultrasonic biological microscopic images of the human eye from multiple angles, as shown in the reference. Figure 2-1 The image shown is an ultrasonic biological microscope image of the human eye according to a specific embodiment of the present invention.

[0081] Next, the anterior chamber angle region is extracted from the human eye ultrasound biomicroscopy image, such as... Figure 2-2 As shown;

[0082] Then, using the improved FCM algorithm, segmentation of the anterior chamber angle region in front of the human eye is achieved, such as... Figure 2-3 As shown.

[0083] Finally, the gradient edge detection operator is used to complete the edge detection of the UBM image segmentation result, and the final contour curve binarization is as follows: Figure 3-1 As shown, the quantization results are output intuitively, such as... Figure 3-2 As shown.

[0084] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for Schlem tube segmentation and quantization based on ultrasonic biological microscope images, characterized in that, The method includes: Step 1: Acquire ultrasound biomicroscopy images from multiple orientations, select appropriate ultrasound biomicroscopy images and extract the original UBM images containing Schlem tubes, perform histogram clustering on them, and initialize the membership matrix of the original UBM images according to the number of clusters. Step 2: Perform morphological closure reconstruction, i.e., MCR operation, on the original UBM image to obtain statistical information of the reconstructed UBM image as prior knowledge; Step 3: Add the prior knowledge from step (2) to the objective function of the FCM algorithm, and iteratively solve the objective function to obtain the membership partition matrix of the UBM image reconstructed by MCR; Step 4: Use the membership partition matrix from step (3) as prior information to fill the membership partition matrix of the original UBM image, obtain the filled membership partition matrix, perform median filtering on the filled membership partition matrix, obtain the filtered membership partition matrix, that is, the updated membership partition matrix of the original UBM image, and use this membership partition matrix to segment the original UBM image. Step 5: Using the gradient edge detection operator, the boundary curves of each sub-region of the original UBM image segmentation result are binarized to visually display the Schlem tube in the anterior chamber angle region and output the quantization result; In step 1, performing histogram clustering and initializing the membership matrix of the original UBM image according to the number of clusters is based on the original UBM image. The histogram information is used to count the number of peaks c in the histogram, and the gray values ​​corresponding to all peaks are recorded as follows: Then Let be the initial cluster centers, and initialize the membership matrix of the original UBM image as . N is the total number of pixels in the original UBM image.

2. The Schlemma tube segmentation and quantization method based on ultrasonic biological microscope images according to claim 1, characterized in that, In step 2, morphological closure reconstruction of the original UBM image is performed using structural elements. For UBM images The formula for MCR reconstruction is: ； in, This represents the erosion operation. This indicates the expansion operation. Indicates the use of right Perform corrosion calculations; express Compared to The morphological expansion reconstruction results; Indicates the use of right Perform an expansion operation; express Compared to The morphological corrosion reconstruction results.

3. The Schlemma tube segmentation and quantization method based on ultrasonic biological microscope images according to claim 1, characterized in that, In step 2, obtaining statistical information of the reconstructed UBM image includes grayscale levels. The number of pixels corresponding to each gray level ,satisfy q represents the number of gray levels in the reconstructed UBM image, and q is much smaller than N.

4. The Schlemma tube segmentation and quantization method based on ultrasonic biological microscope images according to claim 3, characterized in that, In step 3, the iterative solution of the objective function is expressed by the following formula: (1) Represents the grayscale values ​​of the reconstructed UBM image. Clustering centers of the original UBM image The fuzzy membership degree is obtained by iterating through the objective function (1) r times to obtain the membership degree partitioning matrix of the reconstructed UBM image. .

5. The Schlemma tube segmentation and quantization method based on ultrasonic biological microscope images according to claim 3, characterized in that, In step 4, the membership partitioning matrix includes two operations: padding and median filtering. Padding is obtained by the following formula: ，(2) Obtain the corresponding original UBM image Membership partition matrix .

6. The Schlemma tube segmentation and quantization method based on ultrasonic biological microscope images according to claim 5, characterized in that, In step 4, the median filter used is denoted as: ，(3) Based on the obtained membership degree partitioning matrix Original UBM image It is divided into several sub-regions.

7. The Schlemma tube segmentation and quantization method based on ultrasonic biological microscope images according to claim 1, characterized in that, In step 5, the boundary curves of each sub-region of the original UBM image segmentation result are binarized using the gradient edge detection operator, calculated as follows: and The edge detection operator performs a convolution operation on the segmentation results of the UBM image, through... Calculate the gradient of each pixel. If the gradient G is greater than a set threshold, the pixel is considered a boundary point.

8. The Schlemma tube segmentation and quantization method based on ultrasonic biological microscope images according to claim 1, characterized in that, In step 5, the quantization results are output. The regionprops function in Matlab is used to find the number of pixels in each region and output the area quantization results of multiple segmented regions.

Images