CN114612448B - Fundus blood vessel segmentation method based on rapid label extraction and self-adaptive topology enhancement - Google Patents
Fundus blood vessel segmentation method based on rapid label extraction and self-adaptive topology enhancement Download PDFInfo
- Publication number
- CN114612448B CN114612448B CN202210265279.0A CN202210265279A CN114612448B CN 114612448 B CN114612448 B CN 114612448B CN 202210265279 A CN202210265279 A CN 202210265279A CN 114612448 B CN114612448 B CN 114612448B
- Authority
- CN
- China
- Prior art keywords
- blood vessel
- skeleton
- vascular
- image
- label
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/136—Segmentation; Edge detection involving thresholding
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30041—Eye; Retina; Ophthalmic
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30101—Blood vessel; Artery; Vein; Vascular
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Radiology & Medical Imaging (AREA)
- Quality & Reliability (AREA)
- Eye Examination Apparatus (AREA)
Abstract
The invention discloses a fundus blood vessel segmentation method based on rapid label extraction and self-adaptive topology enhancement, which can realize automatic extraction of blood vessels in fundus images and does not depend on manual labels of experts in the model training process. This is achieved by (1) acquiring a response map on the fundus image with an optimally directional gradient flux filter. (2) And dividing the main blood vessel on the response graph by adopting an adaptive threshold method. (3) The vascular skeleton was tracked with local maxima on the response map to supplement the fine vessels. (4) Local information around the vascular skeleton is examined to eliminate false positives at the edges of optic discs and pathological tissue. (5) And taking the result of the steps as a label, and training a deep learning model by combining the self-adaptive topology enhancement loss function. According to the method, the training labels are automatically acquired, the requirement of the deep learning algorithm on the manual labels is eliminated, and the accuracy of the deep learning algorithm is increased through the self-adaptive topology enhancement loss function.
Description
Technical Field
The invention belongs to the field of medical image processing, and particularly relates to automatic extraction of fundus blood vessel labels, which is used for training a fundus blood vessel deep learning segmentation model and realizing fundus blood vessel automatic segmentation.
Background
A large number of researches show that fundus images obtained through non-invasive examination can provide important basis for diagnosis of diseases such as diabetic retinopathy, cardiovascular diseases, hypertension and the like. Segmentation of retinal blood vessels plays a critical role in clinical diagnosis for analysis of subsequent vascular properties such as tortuosity and breadth. However, manually segmenting blood vessels from fundus images is a very time consuming task, depending heavily on the experience of the physician. Therefore, a high-precision and high-speed automatic blood vessel segmentation method is highly demanded.
The fundus image vessel segmentation algorithm is always concerned by scholars at home and abroad, and a new method is always proposed. In general, existing segmentation algorithms of fundus blood vessels can be divided into three main categories, (1) segmentation methods based on conventional image processing generally perform segmentation according to morphological characteristics of blood vessels, such as a matched filtering method, a vessel tracking method, a morphological method and a multi-scale method. (2) The machine learning based segmentation method does not require rules to be predefined, but rather distinguishes between vascular and non-vascular by learning. (3) The segmentation method based on deep learning is one branch of a machine learning method, and the segmentation of the blood vessel is completed by training a specific deep network model through data.
The segmentation method is further divided into a supervised method and an unsupervised method, and almost all methods are supervised methods among existing deep learning fundus segmentation methods. These supervised methods mean that a large number of fundus pictures, manually annotated by medical professionals, are required to support training of the deep learning model. And the category labeling of the pictures pixel by pixel requires a great deal of manpower and material cost. To overcome this problem, some unsupervised deep learning methods have been proposed. However, the unsupervised method has generally less segmentation effect than the supervised learning method.
Considering that blood vessels are a linear topology, some studies have proposed a topology-enhanced deep learning approach. These topology enhancement methods have some enhancement effect on the topology consistency of the vessel segmentation results, but do not distinguish between the coarse and fine vessels.
Disclosure of Invention
In order to solve the strong dependence of the supervised deep learning segmentation method on the artificial label, the invention provides a deep learning fundus blood vessel segmentation algorithm based on rapid label extraction and self-adaptive topology enhancement, and training labels are automatically generated by utilizing the steps of optimal directional gradient flux filtering, blood vessel skeleton tracking and the like, so that the manpower and material resource consumption of manual labeling is greatly reduced. In addition, we consider the difference between the coarse blood vessel and the fine blood vessel, and put forward a self-adaptive topology-enhanced loss function to promote the blood vessel continuity and the sensitivity to the fine blood vessel of the blood vessel segmentation result of the deep learning model.
In order to achieve the above object, the method of the present invention comprises the steps of:
Step 1, carrying out optimal directional gradient flux filtering on a training set picture to obtain optimal directional gradient flux response, wherein the optimal directional gradient flux response comprises an optimal vascular metric value M of each pixel point and a corresponding vascular direction;
Step 2, dividing the image into a main vascular structure diagram by using an adaptive threshold method on vascular metric values in the optimal directional gradient flux response;
Step 3, searching blood vessels on the blood vessel metric value by using a local maximum value algorithm to obtain a blood vessel skeleton diagram s;
Step 4, counting the average gray value difference of the neighborhood at two sides of each vascular skeleton s, and deleting the vascular skeletons with large difference to eliminate false positives at the edges of the optic disc and pathological tissues;
Step 5, deleting false positives in the main vessel structure diagram by taking the vessel skeleton as constraint to obtain a crude vessel label g thick;
supplementing a tiny blood vessel part lacking in a main blood vessel diagram by using a blood vessel framework, performing AND operation with the main blood vessel diagram, performing morphological algorithm (opening and closing operation) to perfect an image and obtaining a fine binary blood vessel label g, and simultaneously obtaining a tiny blood vessel label g thin and a coarse blood vessel label g thick;
Step 7, using fundus images and corresponding fine binary blood vessel images g as training set labels to be transmitted into a deep learning model, and training a special deep learning model for fundus image blood vessel segmentation by matching with a self-adaptive topology enhancement loss function;
And 8, saving the learned model, inputting the picture into the model when a new fundus image blood vessel segmentation task needs to be processed, and outputting a segmentation result.
Further, in step 1, the optimal directional gradient flux response is a symmetric matrix Q (x, r), and by similar diagonalization, two eigenvalues λ 1(x,r),λ2 (x, r) and corresponding eigenvectors ω 1(x,r),ω2 (x, r) about the position x can be obtained, and then the optimal directional gradient flux response can be decomposed in the following manner:
Q(x,r)＝λ1(x,r)ω1(x,r)ω1 T(x,r)+λ2(x,r)ω2(x,r)ω2 T(x,r)
Let λ 1≤λ2, then for a position x of the training set picture, the feature vector ω 1 represents the vascular normal direction, ω 2 represents the vascular direction, and the vascular metric value M (x) for the position x is calculated according to the following formula:
where R scale is a series of different scales for multi-scale detection and R is the corresponding scale.
Further, in step 2, the influence of the uneven light intensity and contrast on the segmentation is avoided by an adaptive thresholding method. The adaptive thresholding computes the average value within a window of pixels around each pixel, and the appropriate threshold value for each pixel is computed for a given vascular foreground ratio. And then dividing the blood vessel measurement into binary images by a threshold method, wherein the prospect is a main blood vessel graph.
Further, in step 3, each pixel point in the map is inspected, and for one pixel point x, the direction perpendicular to the blood vessel direction ω 2 is the sameTaking two points A and B with a distance zeta from the x point:
Searching local maximum value on the line segment of the connecting line A and B, setting the local maximum value point as 1, otherwise setting as 0. Traversing each pixel in the graph to obtain a local maximum graph, removing the intersection points in the local maximum graph in an 8-neighborhood to obtain a local maximum graph without the intersection points, and finally, only retaining the skeleton with the skeleton length larger than a given threshold value 20 in the local maximum graph without the intersection points to obtain the vascular skeleton graph without the background noise.
Further, in step 4, first, neighboring areas on two sides of each skeleton are intercepted. Specifically, a single vascular skeleton is obtained by taking each skeleton pixel as a central neighborhood with the radius of l and taking the union of all pixel neighborhoods. Because the neighborhood of the skeleton can be divided into two parts by the skeleton curve and the extension lines of the two endpoints of the skeleton curve, the neighborhood of the two sides of the skeleton can be obtained. And then, calculating the average gray values of two sides of each vascular skeleton by counting the gray values of all pixels in the neighborhood and summing the gray values and dividing the gray values by the number of pixels in the neighborhood. And finally, deleting the frameworks with average gray value differences exceeding a given threshold value of 0.3, thereby completing the false positive removal of the vascular frameworks.
Further, in step 5, the skeleton is used as a reference, and the main vessel image pixels far from the skeleton are removed, which is equivalent to only preserving the vessel pixels in the vicinity of the skeleton in the main vessel image, so as to obtain a crude vessel label g thick.
Further, in step 6, the blood vessel skeleton is expanded by two pixel widths and then is subjected to AND operation with a rough blood vessel label g thick, the combined image eliminates tiny noise in the image through opening operation (firstly corroding the image and then expanding), and then voids in the image are eliminated through closing operation (firstly expanding the image and then corroding), and meanwhile, the edge of the blood vessel is smoothed, wherein the fine blood vessel label g thin＝g-gthick.
Further, in step 7, the fundus image is subjected to model prediction to obtain a prediction result y, and the difference between y and the label g is calculated through the proposed loss function L. The loss function not only considers the correctness of each pixel, but also specifically considers the difference between the coarse and fine vessels, and the weights of the coarse and fine vessels are respectively enhanced by the parameters alpha and beta:
where y is the predicted value, g is the tag value, Is the Hadamard product. L bce is the discrete cross entropy loss function:
Where N represents the total number of pixels, g i and y i represent the label and predictor, respectively, for the ith pixel.
The invention has the beneficial effects that the blood vessel in the fundus image is automatically extracted as the label by using the method based on the optimal directional gradient flux to train the deep learning model, and meanwhile, the influence of a pathological area, a low-quality imaging area and the edge of the optic disc in the fundus image on the blood vessel label is eliminated. The automatic extraction label is used for training the deep learning model, so that the dependence of the deep learning model on the manual label is greatly improved, and manpower and material resources are saved. In addition, the self-adaptive topology enhancement loss function provided by the method can effectively improve the continuity of the blood vessels and the sensitivity to the tiny blood vessels in the segmentation result.
Drawings
Fig. 1 is a flow chart of the method of the present invention.
Fig. 2a is a fundus image.
Fig. 2b is a diagram of a coarse and fine vessel label automatically generated in connection with fig. 2 a.
Fig. 3a is another fundus image
FIG. 3b is a schematic illustration of the vessel segmentation result of FIG. 3a in combination with a deep learning model.
Detailed Description
The invention is further described below with reference to the accompanying drawings.
As shown in fig. 1, the deep learning fundus blood vessel segmentation algorithm based on rapid label extraction and adaptive topology enhancement comprises the following steps:
Step 1, carrying out optimal directional gradient flux filtering on a training set picture to obtain optimal directional gradient flux response, wherein the optimal directional gradient flux response comprises an optimal vascular metric value M of each pixel point and a corresponding vascular direction;
Step 2, dividing the image into a main vascular structure diagram by using an adaptive threshold method on vascular metric values in the optimal directional gradient flux response;
Step 3, searching blood vessels on the blood vessel metric value by using a local maximum value algorithm to obtain a blood vessel skeleton diagram s;
Step 4, counting the average gray value difference of the neighborhood at two sides of each vascular skeleton s, and deleting the vascular skeletons with large difference to eliminate false positives at the edges of the optic disc and pathological tissues;
Step 5, deleting false positives in the main vessel structure diagram by taking the vessel skeleton as constraint to obtain a crude vessel label g thick;
Step 6, supplementing the tiny blood vessel part lacking in the main blood vessel diagram by using a blood vessel framework, performing AND operation with the main blood vessel diagram, performing morphological algorithm (opening and closing operation) to perfect the image, obtaining a fine binary blood vessel label g, and simultaneously obtaining a tiny blood vessel label g thin and a thick blood vessel label g thick, wherein the thicker lines represent the thick blood vessel label, and the thinner lines represent the tiny blood vessel label, as shown in the figure 2 b.
Step 7, using fundus images and corresponding fine binary blood vessel images g as training set labels to be transmitted into a deep learning model, and training a special deep learning model for fundus image blood vessel segmentation by matching with a self-adaptive topology enhancement loss function;
and 8, saving the learned model, inputting the picture into the model when a new fundus image blood vessel segmentation task needs to be processed, and outputting a segmentation result. The prediction result after the fundus image of fig. 3a is input to the model is fig. 3b.
Further, in step 1, the optimal directional gradient flux response is a symmetric matrix Q (x, r), and by similar diagonalization, two eigenvalues λ 1(x,r),λ2 (x, r) and corresponding eigenvectors ω 1(x,r),ω2 (x, r) about the position x can be obtained, and then the optimal directional gradient flux response can be decomposed in the following manner:
Q(x,r)＝λ1(x,r)ω1(x,r)ω1 T(x,r)+λ2(x,r)ω2(x,r)ω2 T(x,r)
Let λ 1≤λ2, then for a position x of the training set picture, the feature vector ω 1 represents the vascular normal direction, ω 2 represents the vascular direction, and the vascular metric value M (x) for the position x is calculated according to the following formula:
where R scale is a series of different scales for multi-scale detection and R is the corresponding scale.
Example 1:
In step 1, a range R scale = [1,8] of multi-scale detection is taken.
Further, in step 2, the influence of the uneven light intensity and contrast on the segmentation is avoided by an adaptive thresholding method. The adaptive thresholding computes the average value within a window of pixels around each pixel, and the appropriate threshold value for each pixel is computed for a given vascular foreground ratio. And then dividing the blood vessel measurement into binary images by a threshold method, wherein the prospect is a main blood vessel graph.
Further, in step 3, each pixel point in the map is inspected, and for one pixel point x, the direction perpendicular to the blood vessel direction ω 2 is the sameTaking two points A and B with a distance zeta from the x point:
Searching local maximum value on the line segment of the connecting line A and B, setting the local maximum value point as 1, otherwise setting as 0. Traversing each pixel in the graph to obtain a local maximum graph, removing the intersection points in the local maximum graph in an 8-neighborhood to obtain a local maximum graph without the intersection points, and finally, only retaining the skeleton with the skeleton length larger than a given threshold value 20 in the local maximum graph without the intersection points to obtain the vascular skeleton graph without the background noise.
Example 2:
in step 3, the search range ζ=10 is taken.
Further, in step 4, first, neighboring areas on two sides of each skeleton are intercepted. Specifically, a single vascular skeleton is obtained by taking each skeleton pixel as a central neighborhood with the radius of l and taking the union of all pixel neighborhoods. Because the neighborhood of the skeleton can be divided into two parts by the skeleton curve and the extension lines of the two endpoints of the skeleton curve, the neighborhood of the two sides of the skeleton can be obtained. And then, calculating the average gray values of two sides of each vascular skeleton by counting the gray values of all pixels in the neighborhood and summing the gray values and dividing the gray values by the number of pixels in the neighborhood. And finally, deleting the frameworks with average gray value differences exceeding a given threshold value of 0.3, thereby completing the false positive removal of the vascular frameworks.
Further, in step 5, the skeleton is used as a reference, and the main vessel image pixels far from the skeleton are removed, which is equivalent to only preserving the vessel pixels in the vicinity of the skeleton in the main vessel image, so as to obtain a crude vessel label g thick.
Further, in step 6, the blood vessel skeleton is expanded by two pixel widths and then is subjected to AND operation with a rough blood vessel label g thick, the combined image eliminates tiny noise in the image through opening operation (firstly corroding the image and then expanding), and then voids in the image are eliminated through closing operation (firstly expanding the image and then corroding), and meanwhile, the edge of the blood vessel is smoothed, wherein the fine blood vessel label g thin＝g-gthick.
Further, in step 7, the fundus image is subjected to model prediction to obtain a prediction result y, and the difference between y and the label g is calculated through the proposed loss function L. The loss function not only considers the correctness of each pixel, but also specifically considers the difference between the coarse and fine vessels, and the weights of the coarse and fine vessels are respectively enhanced by the parameters alpha and beta:
where y is the predicted value, g is the tag value, Is the Hadamard product. L bce is the discrete cross entropy loss function:
Where N represents the total number of pixels, g i and y i represent the label and predictor, respectively, for the ith pixel.
Example 4:
in step 7, the training set, the verification set and the test set are divided according to the ratio of 10:1:10.
Finally, it should be noted that the described embodiments are merely some, but not all embodiments of the present invention. Based on the embodiments of the present invention, those skilled in the art may modify the technical solutions described in the foregoing embodiments, or obtain other embodiments with equivalent substitution of some technical features, which fall within the scope of the present invention.
Claims (4)
1. The fundus blood vessel segmentation method based on the rapid label extraction and the self-adaptive topology enhancement is characterized by comprising the following steps of:
Step 1, carrying out optimal directional gradient flux filtering on a training set picture to obtain optimal directional gradient flux response, wherein the optimal directional gradient flux response comprises an optimal vascular metric value M of each pixel point and a corresponding vascular direction, and the method comprises the following steps:
The optimal directional gradient flux response is a symmetric matrix Q (x, r) that, by similarity diagonalization, yields its two eigenvalues λ 1(x,r),λ2 (x, r) and corresponding eigenvectors ω 1(x,r),ω2 (x, r) with respect to position x, then the optimal directional gradient flux response is decomposed in the following way:
Q(x,r)＝λ1(x,r)ω1(x,r)ω1 T(x,r)+λ2(x,r)ω2(x,r)ω2 T(x,r)
Let λ 1≤λ2, then for a position x of the training set picture, the feature vector ω 1 represents the vascular normal direction, ω 2 represents the vascular direction, and the vascular metric value M (x) for the position x is calculated according to the following formula:
where R scale is a series of different scales for multi-scale detection, R is the corresponding scale;
Step 2, dividing the image into a main vascular structure diagram by using an adaptive threshold method on vascular metric values in the optimal directional gradient flux response;
Step 3, searching blood vessels on the blood vessel metric value by using a local maximum value algorithm to obtain a blood vessel skeleton s;
Step 4, counting the average gray value difference of the neighborhood at two sides of each vascular skeleton s, and deleting the vascular skeletons with large difference to eliminate false positives at the edges of the optic disc and pathological tissues;
Step 5, deleting false positives in the main vessel structure diagram by taking the vessel skeleton as constraint to obtain a crude vessel label g thick;
step 6, supplementing a tiny blood vessel part lacking in a main blood vessel diagram by using a blood vessel framework, performing AND operation with the main blood vessel diagram, performing morphological algorithm to perfect an image and obtain a fine binary blood vessel label g, and simultaneously obtaining a tiny blood vessel label g thin and a coarse blood vessel label g thick, wherein the method comprises the following steps:
The method comprises the steps of performing AND operation on a blood vessel skeleton with two pixel widths and a rough blood vessel label g thick, and removing tiny noise in the image by an opening operation on the combined image, wherein the opening operation is to erode the image firstly and then expand the image, and then removing a cavity in the image by a closing operation on the image, wherein the closing operation is to dilate the image firstly and then erode the image, and simultaneously smoothing the edge of the blood vessel, wherein the fine blood vessel label g thin＝g-gthick;
And 7, using the fundus picture and the corresponding fine binary blood vessel picture g as training set labels to be transmitted into a deep learning model, and training a special deep learning model for fundus image blood vessel segmentation by matching with a self-adaptive topology enhancement loss function, wherein the method comprises the following steps:
the fundus image is subjected to model prediction to obtain a prediction result y, the difference between y and a fine binary blood vessel label g is calculated through a proposed loss function L, the loss function not only considers the correctness of each pixel, but also particularly considers the difference between thick and fine blood vessels, and the weights of the thick blood vessel and the fine blood vessel are respectively enhanced through parameters alpha and beta:
where y is the predicted value, g is the tag value, Is the hadamard product, L bce is the discrete cross entropy loss function:
where N represents the total number of pixels, g i and y i represent the label and predictor of the ith pixel, respectively;
And 8, storing the deep learning model obtained in the step 7, inputting the picture into the model when a new fundus image blood vessel segmentation task needs to be processed, and outputting a segmentation result.
2. The fundus blood vessel segmentation method according to claim 1, wherein in the step 2, an image is segmented into a main blood vessel structure map using an adaptive thresholding method on a blood vessel metric value in an optimal directional gradient flux response, comprising the steps of:
The influence of uneven light intensity and contrast on segmentation is avoided by a self-adaptive threshold method; the self-adaptive threshold method calculates the average value in the pixel window around each pixel point, calculates the proper threshold value for each pixel by the given vascular foreground proportion, and then divides the vascular metric into binary images by the threshold method, wherein the foreground is the main vascular map.
3. The fundus blood vessel segmentation method according to claim 1, wherein in the step 3, the blood vessel is searched on the blood vessel metric value by using a local maximum value algorithm to obtain a blood vessel skeleton map s, the method comprises the following steps:
Checking each pixel point in the training set picture, wherein for one pixel point x, the orthogonal direction of the blood vessel direction omega 2 is that of the pixel point x On, take two distances from the x pointPoints a, B of (a):
Searching local maximum value on the line segment of the connecting line A and B, setting the local maximum value point as 1, otherwise setting the local maximum value point as 0, traversing each pixel in the training set picture to obtain a local maximum value picture, removing the intersection point in the local maximum value picture in an 8-neighborhood to obtain a local maximum value picture without the intersection point, and finally, only reserving the skeleton with the skeleton length larger than a given threshold value 20 in the local maximum value picture without the intersection point to obtain the vascular skeleton picture without the background noise.
4. The fundus blood vessel segmentation method according to claim 1, wherein in the step 4, the average gray value difference of the neighborhood at both sides of each blood vessel skeleton s is counted, the blood vessel skeletons with large difference are deleted to eliminate false positives at the edges of optic discs and pathological tissues, and the method comprises the following steps:
Firstly, cutting off the neighborhood at two sides of each skeleton, specifically to a single vascular skeleton, namely obtaining the neighborhood of the skeleton by taking each skeleton pixel as a center and taking the union of all pixel neighborhood, wherein the neighborhood of the skeleton is divided into two parts by the skeleton curve and the extension lines of two endpoints of the skeleton, namely obtaining the neighborhood at two sides of the skeleton, then calculating the average gray value at two sides of each vascular skeleton by counting the gray values of all pixels in the neighborhood and summing up the gray values and dividing the gray values by the number of pixels in the neighborhood, and finally deleting the skeleton with the difference of the average gray values at two sides exceeding a given threshold value of 0.3, thereby completing the false positive removal of the vascular skeleton.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210265279.0A CN114612448B (en) | 2022-03-17 | 2022-03-17 | Fundus blood vessel segmentation method based on rapid label extraction and self-adaptive topology enhancement |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210265279.0A CN114612448B (en) | 2022-03-17 | 2022-03-17 | Fundus blood vessel segmentation method based on rapid label extraction and self-adaptive topology enhancement |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114612448A CN114612448A (en) | 2022-06-10 |
| CN114612448B true CN114612448B (en) | 2025-05-23 |
Family
ID=81865123
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210265279.0A Active CN114612448B (en) | 2022-03-17 | 2022-03-17 | Fundus blood vessel segmentation method based on rapid label extraction and self-adaptive topology enhancement |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114612448B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116523887A (en) * | 2023-05-08 | 2023-08-01 | 依未科技(北京)有限公司 | Image processing method and device, storage medium and electronic equipment |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106228566A (en) * | 2016-08-15 | 2016-12-14 | 哈尔滨工业大学 | Automatically detection and the recognition methods of retinal microaneurysm based on the loud analysis of gradient |
| CN116824203B (en) * | 2019-12-04 | 2025-12-23 | 深圳硅基智能科技有限公司 | A glaucoma identification device and method based on neural networks |
| CN113920078B (en) * | 2021-09-30 | 2025-08-22 | 北京鹰瞳科技发展股份有限公司 | A blood vessel segmentation method and electronic device for fundus images |
| CN114092439B (en) * | 2021-11-18 | 2025-05-09 | 深圳大学 | A multi-organ instance segmentation method and system |
-
2022
- 2022-03-17 CN CN202210265279.0A patent/CN114612448B/en active Active
Non-Patent Citations (1)
| Title |
|---|
| "基于自动标签与多任务学习的视网膜血管分割方法研究";石亦恒;《中国优秀硕士学位论文全文数据库 医药卫生科技辑》;20230115(第2023年01期);第E073-185页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114612448A (en) | 2022-06-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107644420B (en) | Vascular image segmentation method and MRI system based on centerline extraction | |
| Bülow et al. | Automatic extraction of the pulmonary artery tree from multi-slice CT data | |
| CN112716446A (en) | Method and system for measuring pathological change characteristics of hypertensive retinopathy | |
| WO2021203795A1 (en) | Pancreas ct automatic segmentation method based on saliency dense connection expansion convolutional network | |
| JP7695243B2 (en) | Method for determining severity of skin disease based on percentage of body surface area occupied by lesions | |
| SG194407A1 (en) | A method and system for determining a stage of fibrosis in a liver | |
| CN115661467A (en) | Cerebrovascular image segmentation method, device, electronic equipment and storage medium | |
| Chen et al. | A combined method for automatic identification of the breast boundary in mammograms | |
| CN113643353B (en) | Measurement method for enhancing resolution of vascular caliber of fundus image | |
| Mabrouk et al. | Multiscale graph cuts based method for coronary artery segmentation in angiograms | |
| CN113516644B (en) | Intravascular optical coherence tomography branch blood vessel identification method | |
| CN110738637A (en) | A method and system for automatic classification of breast cancer pathological sections | |
| CN116524548A (en) | Vascular structure information extraction method, device and storage medium | |
| CN118333952A (en) | Automatic segmentation method of 3D coronary arteries based on local active contour model | |
| Chen et al. | Automatic 3D coronary artery segmentation based on local region active contour model | |
| CN114612448B (en) | Fundus blood vessel segmentation method based on rapid label extraction and self-adaptive topology enhancement | |
| Mehena et al. | Brain tumor segmentation and extraction of MR images based on improved watershed transform | |
| CN115841472B (en) | Methods, devices, equipment, and storage media for identifying the high-density sign of the middle cerebral artery. | |
| Nie et al. | Improved Cascade-RCNN for automatic detection of coronary artery plaque in multi-angle fusion CPR images | |
| Su et al. | A novel method for dendritic spines detection based on directional morphological filter and shortest path | |
| Ogiela et al. | Automatic segmentation of the carotid artery bifurcation region with a region-growing approach | |
| Shang et al. | Adaptive directional region growing segmentation of the hepatic vasculature | |
| CN114764810A (en) | Medical image segmentation method | |
| Luo et al. | Extraction of brain vessels from magnetic resonance angiographic images: Concise literature review, challenges, and proposals | |
| Brindise et al. | Automated layer identification method for skin tissue histology images |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |