CN110309827B - Edema region segmentation model based on OCT image - Google Patents

Edema region segmentation model based on OCT image Download PDF

Info

Publication number
CN110309827B
CN110309827B CN201910371329.1A CN201910371329A CN110309827B CN 110309827 B CN110309827 B CN 110309827B CN 201910371329 A CN201910371329 A CN 201910371329A CN 110309827 B CN110309827 B CN 110309827B
Authority
CN
China
Prior art keywords
segmentation
oct image
edema
curve
area
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
Application number
CN201910371329.1A
Other languages
Chinese (zh)
Other versions
CN110309827A (en
Inventor
***
张文苹
陶志富
宋巍
孙亚男
颜标
李朝鹏
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shanghai Ocean University
Original Assignee
Shanghai Ocean University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Shanghai Ocean University filed Critical Shanghai Ocean University
Priority to CN201910371329.1A priority Critical patent/CN110309827B/en
Publication of CN110309827A publication Critical patent/CN110309827A/en
Application granted granted Critical
Publication of CN110309827B publication Critical patent/CN110309827B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G06F18/232—Non-hierarchical techniques
    • G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
    • G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/11—Region-based segmentation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/13—Edge detection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Data Mining & Analysis (AREA)
  • Artificial Intelligence (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Probability & Statistics with Applications (AREA)
  • Multimedia (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses an edema area segmentation algorithm based on an OCT image, which comprises the following steps: 1) Preprocessing an image, namely preprocessing an OCT image to be segmented by adopting Gaussian filtering; 2) Roughly dividing an edema area, roughly dividing an OCT image based on a K-means algorithm to obtain an interested area of the OCT image; 3) And (3) performing fine segmentation on the edema area, wherein the area of interest of the OCT image is used as an initial boundary, and the OCT image is subjected to fine segmentation based on an improved level set algorithm. The invention reduces the iteration times of boundary line detection and calculation and improves the segmentation efficiency; the time efficiency is improved by about 23%, and the iteration times are reduced by about 30%. The invention has the following advantages in terms of segmentation precision: the SPF function is used for replacing the edge stopping function to improve the existing level set model, and the obtained model can be converged to an edema area inside the ROI, so that the segmentation of the OCT image edema area is realized.

Description

Edema region segmentation model based on OCT image
Technical Field
The invention relates to the field of biomedical information and image processing, in particular to an edema area segmentation model based on an OCT image.
Background
Diabetic Macular Edema (DME) is caused by fluid leakage resulting from macular vascular damage and retinal blood barrier breakdown, and is characterized by edema and retinal thickening in the retina, which are the major causes of visual impairment in Diabetic Retinopathy (DR) patients.
Optical Coherence Tomography (OCT) is an image created by scanning based on the michelson interference principle, which can effectively display a retinal lesion site, has high resolution, can clearly identify an edema area, and accurately provide the edema area site, and thus is widely used for diagnosis of DME. Typically, medical personnel interpret the edematous region of OCT using a visual interpretation to assess the severity of DME. The disadvantages of this approach are: the edema area is visually interpreted and interpreted to have greater contingency under the influence of factors such as the experience of the medical practitioner and the OCT image quality.
In recent years, computer image aided segmentation methods are gradually applied to OCT image processing, and segmentation algorithms based on OCT images can be classified into two types: machine learning based segmentation algorithms and non-machine learning based segmentation algorithms. The segmentation algorithm based on machine learning comprises an image segmentation region iterative segmentation algorithm guided by K neighbor classification and provided by administrative and others, and a medical image with uneven segmentation strength by an adaptive fuzzy C mean method provided by Pham and the like; blanzet et al utilize a feed-forward neural network to solve the problem of segmentation of overlapping images, etc. Non-machine learning algorithm there is Vincent et al proposed waterline (watershed) algorithm for extracting image interesting regions; osher et al propose a level set method, effectively solve the problem of topological change in the curve evolution process, and are applied to image segmentation in the medical and aviation fields; the Sobel algorithm proposed by Sobel et al and the Canny algorithm proposed by Jcanny, which realize the analysis and processing of medical images by the edge detection of the images; kass et al propose that the active contour model ACM uses prior knowledge to constrain the segmentation problem, and obtain the image segmentation result with closed and smooth boundary; the segmentation method proposed by Boykov et al solves the segmentation problem in the global optimization framework.
The method initially realizes the automatic segmentation of the edema area based on the OCT image, but still has the following problems under the influence of OCT image quality, the blurring of the edema area boundary, the diversification of the edema area shape and the like:
1) The existing segmentation methods suffer from the phenomenon of over-segmentation when segmenting edema regions.
2) Due to the influence of boundary ambiguity, the existing segmentation method has the problems of high computational complexity and the like in the process of searching the boundary of the edema area.
3) Due to the influence of the shape of the edema area, the existing segmentation method has the problems of low accuracy, low calculation efficiency and the like.
Disclosure of Invention
The invention aims to overcome the defects in the prior art and provides an edema area segmentation model based on an OCT image, which is used for improving the accuracy and efficiency of the edema area segmentation based on the OCT image so as to solve the problems in the prior art.
The technical problem solved by the invention can be realized by adopting the following technical scheme:
an edema area segmentation algorithm based on an OCT image comprises the following steps:
1) Preprocessing an image, namely preprocessing an OCT image to be segmented by adopting Gaussian filtering;
2) Roughly dividing an edema area, roughly dividing the OCT image based on a K-means algorithm to obtain a region of interest of the OCT image;
3) And (4) performing fine segmentation on the edema area, wherein the area of interest of the OCT image is used as an initial boundary, and the OCT image is subjected to fine segmentation based on an improved level set algorithm.
Further, the edema area rough segmentation is based on a K-means algorithm, and the method is as follows:
x = { X for a given set of samples 1 ,x 2 ,...,x n }, randomly formulating K clustering centers C 1 ,C 2 ,...,C K Then for each sample x 1 Find the nearest cluster center C v And assign it to C v The marked class, and returning each C of the last iteration w Move to the center of its designated class and calculate the deviation D until D converges, then return to the cluster center and terminate the algorithm.
Further, the subdivision of the edema area is based on an improved level set algorithm, which is as follows:
the level set equation is:
Figure BDA0002050063340000031
wherein phi is a level set function, and t is evolution time; v (C) is a speed function of curve C evolution; using the symbol distance function SDF as the initial level set function, as shown in equation (3), the SDF function is constructed as follows:
Figure BDA0002050063340000032
wherein omega 1 Is the inner region enclosed by curve C, Ω 2 Is the area outside curve C; s is a parameter of curve C; d ((x, y), C (s, t)) represents the distance of the coordinate (x, y) to the curve C (s, t);
the derivation of the level set equation is as follows:
the evolution curve C (s, t) can be represented by a zero level set function, which has:
φ(C(s,t),t)＝0 (4)
the time t is derived based on the full differential principle, and the following results are obtained:
Figure BDA0002050063340000041
based on the curve evolution theory, the motion equation of the closed curve C (s, t) is:
Figure BDA0002050063340000042
wherein V (C) can be a constant value or a variable; the inner unit normal vector N of the closed curve C can be written as:
Figure BDA0002050063340000043
the level set equation (2) can be derived by substituting the equation (5) of the evolution curve C and the calculation equation (7) of the inner unit normal vector N into the equation (6).
Compared with the prior art, the invention has the beneficial effects that:
1) In terms of computational efficiency: firstly, segmenting an initial evolution curve, namely rough segmentation, of an edema area on the OCT image by using a k-means algorithm; and then, optimizing the boundary of the edema area on the OCT image by using the initial evolution curve as an input through an improved level set method.
Compared with other segmentation models, the edema area segmentation model provided by the invention reduces the iteration times of boundary line detection calculation and improves the segmentation efficiency; experiments show that the time efficiency is improved by about 23 percent, and the iteration times are reduced by about 30 percent.
2) In terms of segmentation accuracy: the existing level set model (SBGFRLS) is improved by replacing an edge stop function with an SPF function, and the obtained model can be converged into an edema area inside the ROI, so that the segmentation of the edema area of the OCT image is realized.
Drawings
FIG. 1a is a raw OCT image of the invention.
FIG. 1b is a denoised OCT image according to the invention.
Fig. 2 is a flowchart of an edema area segmentation model based on OCT images according to the present invention.
FIG. 3a is a diagram illustrating the segmentation result of the initial evolution curve of the edema zone according to the present invention.
FIG. 3b is a schematic diagram illustrating the optimization of the edema zone boundary according to the present invention.
FIG. 4a is a comparison graph of the segmentation result of the edema area of the OCT image according to the present invention.
FIG. 4b is a comparison graph of the segmentation result of the edema area of the OCT image according to the present invention.
Detailed Description
In order to make the technical means, the creation characteristics, the achievement purposes and the effects of the invention easy to understand, the invention is further explained by combining the specific embodiments.
Referring to fig. 2, the edema area segmentation algorithm based on OCT image according to the present invention includes the following steps:
1) Image pre-processing
The invention adopts Gaussian filtering to preprocess the OCT image to be segmented; fig. 1 (a) is an original OCT image, and fig. 1 (b) is a denoised OCT image. As can be seen from fig. 1 (a) and 1 (b), the speckle noise in the image after the gaussian filtering is effectively suppressed, the boundary information of the image is well retained, and the quality of the image is significantly improved.
2) Rough segmentation of edema area
The invention utilizes k-means algorithm to realize the segmentation of the initial evolution curve of the edema area on the OCT image;
inputting an OCT image I to be segmented, and obtaining an initial evolution curve C by utilizing a K-means algorithm, wherein C meets the requirement of
Figure BDA0002050063340000061
The initial evolution curve C divides the image into an ROI (region of interest) area and a background area, and the marked ROI boundary is used as an initial boundary for next edema area segmentation, so that the calculation iteration times are reduced, and the calculation time is shortened.
3) Subdivision and segmentation of edema area
The invention realizes the optimization of the edema zone boundary based on the improved level set method;
inputting the segmentation result based on the step 1, namely an initial evolution curve C, according to a formula
Figure BDA0002050063340000062
Calculating the average gray value c inside and outside the curve 1 ,c 2 。
According to the formula
Figure BDA0002050063340000063
The symbolic pressure function spf is solved.
And stopping iteration if the evolution curve converges to the boundary or reaches the set iteration times.
The experimental results and the precision comparison of the present invention are illustrated in the following examples of the application of the segmentation model of edema area based on OCT images.
1. OCT image edema area segmentation based on improved level set method
With OCT as original data, realizing segmentation of initial evolution curve of edema area on OCT image by using k-means algorithm, the result is shown in figure 3 (a); the edema zone boundary optimization is realized based on the improved level set method, that is, the OCT edema zone is finely segmented, and the segmentation result of the edema zone is finally obtained as shown in fig. 3 (b).
2. Comparison of segmentation results of different segmentation models on edema region boundary
The method selects 12 groups of OCT images, and respectively adopts a C-V model, a GAC model, an SBGFRLS model and the edema area segmentation model provided by the invention to compare the segmentation efficiency of the edema areas in the 12 groups of OCT images, wherein the segmentation efficiency comprises the segmentation time efficiency, the iteration times and the like.
Fig. 4 is a comparison diagram of segmentation results of edema area of 2 sets of OCT images, including the initial evolutionary curve and the boundary of edema area, wherein fig. 4 (a) is a comparison diagram of segmentation results of various models of OCT 1; fig. 4 (b) is a graph comparing the results of the model segmentation of OCT 2.
Tables 1 and 2 are the results of comparing the segmentation times and the number of iterations of the four segmentation models.
TABLE 1 comparison of calculated time for segmentation of edema zones
Figure BDA0002050063340000071
TABLE 2 comparison of iterations for segmentation of edema zones
Figure BDA0002050063340000072
Figure BDA0002050063340000081
As can be seen from tables 1 and 2,
(1) The C-V model can obtain a similar segmentation result with the improved model of the invention, but the operation time is as long as 40 minutes, and the segmentation efficiency is low;
(2) The result of the GAC model is limited by an initial curve, the GAC model cannot be converged to an ROI region and cannot mark a edema region, and the convergence speed is higher than that of the C-V model;
(3) The SBGFRLS model is highly efficient and converges to the ROI region but not to the edema region.
Compared with the first three models, the improved model can accurately segment the ROI and the edema; in terms of efficiency, compared with the SBGFRLS model with the highest segmentation efficiency in the first three models, the segmentation time of the model is reduced by 23%, and the iteration number is reduced by 30%.
Comparing the segmentation result based on the segmentation model of the invention with the expert marking result to obtain the accuracy (P) r ) Specificity (S) p ) Sensitivity (S) e ) And a similarity coefficient (R) Dice ) On the four aspects, the results of the segmentation are compared in terms of accuracy, and the specific formulas are (11), (12), (13) and (14).
Figure BDA0002050063340000082
Figure BDA0002050063340000083
Figure BDA0002050063340000091
Figure BDA0002050063340000092
Wherein, T P Number of pixels representing correctly segmented edema area, F P Indicating a wrongly splitNumber of pixels in edema region, F N Indicating the number of pixels erroneously divided into the background region, T N Indicating the number of pixels of the segmentation marker as background region, G T The number of manually marked edema region pixels is indicated, and R represents the number of edema region pixels segmented in the segmentation model of the present invention.
With manual labeling as a standard, the following was evaluated:
TABLE 3 evaluation of the accuracy of the segmentation results of the present invention
Figure BDA0002050063340000093
As can be seen from Table 3, the average accuracy, specificity, sensitivity and similarity coefficient are respectively 97.7%, 99.2%, 91.8% and 94.4%, the segmentation effect of the model is better, and qualitative and quantitative references can be provided for later-stage clinical diagnosis and treatment.
The foregoing shows and describes the general principles and features of the present invention, together with the advantages thereof. It will be understood by those skilled in the art that the present invention is not limited to the embodiments described above, which are described in the specification and illustrated only to illustrate the principle of the present invention, but that various changes and modifications may be made therein without departing from the spirit and scope of the present invention, which fall within the scope of the invention as claimed. The scope of the invention is defined by the appended claims and equivalents thereof.

Claims (1)

1. An edema area segmentation algorithm based on an OCT image is characterized by comprising the following steps:
1) Preprocessing an image, namely preprocessing an OCT image to be segmented by adopting Gaussian filtering;
2) Roughly dividing an edema area, roughly dividing an OCT image based on a K-means algorithm to obtain an interested area of the OCT image;
3) Finely dividing an edema area, namely finely dividing the OCT image based on an improved level set algorithm by taking the region of interest of the OCT image as an initial boundary;
the edema area rough segmentation is based on a K-means algorithm, and the method comprises the following steps:
x = { X) for a given set of samples 1 ,x 2 ,...,x n }, randomly formulating K clustering centers C 1 ,C 2 ,...,C K Then for each sample x 1 Find the nearest cluster center C v And assign it to C v The marked class returns each C of the last iteration w Moving to the center of the indicated class, calculating the deviation D, returning to the clustering center and terminating the algorithm until the D is converged;
the subdivision of the edema area is based on an improved level set algorithm, which is as follows:
the level set equation is:
Figure FDA0003827465530000011
wherein phi is a level set function, and t is evolution time; v (C) is a speed function of curve C evolution; using the symbol distance function SDF as the initial level set function, as shown in equation (3), the SDF function is constructed as follows:
Figure FDA0003827465530000012
wherein omega 1 Is the inner region enclosed by curve C, Ω 2 Is the area outside curve C; s is a parameter of curve C; d ((x, y), C (s, t)) represents the distance of the coordinate (x, y) to the curve C (s, t);
the derivation of the level set equation is as follows:
the evolution curve C (s, t) can be represented by a zero level set function, which has:
φ(C(s,t),t)＝0 (4)
the time t is derived based on the full differential principle, and the following results are obtained:
Figure FDA0003827465530000021
based on the curve evolution theory, the motion equation of the closed curve C (s, t) is:
Figure FDA0003827465530000022
wherein V (C) can be a constant value or a variable; the inner unit normal vector N of the closed curve C can be written as:
Figure FDA0003827465530000023
and substituting the equation (5) of the evolution curve C and the calculation equation (7) of the inner unit normal vector N into the equation (6) to obtain the level set equation (2).
CN201910371329.1A 2019-05-06 2019-05-06 Edema region segmentation model based on OCT image Active CN110309827B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910371329.1A CN110309827B (en) 2019-05-06 2019-05-06 Edema region segmentation model based on OCT image

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910371329.1A CN110309827B (en) 2019-05-06 2019-05-06 Edema region segmentation model based on OCT image

Publications (2)

Publication Number Publication Date
CN110309827A CN110309827A (en) 2019-10-08
CN110309827B true CN110309827B (en) 2022-10-14

Family

ID=68075372

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910371329.1A Active CN110309827B (en) 2019-05-06 2019-05-06 Edema region segmentation model based on OCT image

Country Status (1)

Country Link
CN (1) CN110309827B (en)

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107154047A (en) * 2017-04-24 2017-09-12 天津大学 Multi-mode brain tumor image blend dividing method and device
WO2019001208A1 (en) * 2017-06-28 2019-01-03 苏州比格威医疗科技有限公司 Segmentation algorithm for choroidal neovascularization in oct image

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107154047A (en) * 2017-04-24 2017-09-12 天津大学 Multi-mode brain tumor image blend dividing method and device
WO2019001208A1 (en) * 2017-06-28 2019-01-03 苏州比格威医疗科技有限公司 Segmentation algorithm for choroidal neovascularization in oct image

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
基于K-means聚类与改进随机游走算法的冠脉光学相干断层图像斑块分割;王光磊等;《生物医学工程学杂志》;20171225(第06期);全文 *

Also Published As

Publication number Publication date
CN110309827A (en) 2019-10-08

Similar Documents

Publication Publication Date Title
Mary et al. Retinal fundus image analysis for diagnosis of glaucoma: a comprehensive survey
Mohamed et al. An automated glaucoma screening system using cup-to-disc ratio via simple linear iterative clustering superpixel approach
Zhao et al. Saliency driven vasculature segmentation with infinite perimeter active contour model
Yin et al. Automatic segmentation and measurement of vasculature in retinal fundus images using probabilistic formulation
Villalobos-Castaldi et al. A fast, efficient and automated method to extract vessels from fundus images
CN108961261B (en) Optic disk region OCT image hierarchy segmentation method based on space continuity constraint
Panda et al. New binary Hausdorff symmetry measure based seeded region growing for retinal vessel segmentation
Xiang et al. Automatic retinal layer segmentation of OCT images with central serous retinopathy
Chen et al. Automated segmentation of the choroid in EDI-OCT images with retinal pathology using convolution neural networks
Okuwobi et al. Automated quantification of hyperreflective foci in SD-OCT with diabetic retinopathy
Ni Ni et al. Anterior Chamber Angle Shape Analysis and Classification of Glaucoma in SS‐OCT Images
Cha et al. High-accuracy retinal layer segmentation for optical coherence tomography using tracking kernels based on Gaussian mixture model
Rodrigues et al. Retinal vessel segmentation using parallel grayscale skeletonization algorithm and mathematical morphology
Vijapur et al. Sensitized glaucoma detection using a unique template based correlation filter and undecimated isotropic wavelet transform
Senapati Bright lesion detection in color fundus images based on texture features
Chen et al. Automatic segmentation of fluid-associated abnormalities and pigment epithelial detachment in retinal SD-OCT images
Akram et al. Blood vessel enhancement and segmentation using wavelet transform
Chen et al. Combination of enhanced depth imaging optical coherence tomography and fundus images for glaucoma screening
Zhou et al. Segmentation of optic disc in retinal images using an improved gradient vector flow algorithm
Zhu et al. Automated framework for intraretinal cystoid macular edema segmentation in three-dimensional optical coherence tomography images with macular hole
Krishnamoorthi et al. Hybrid feature vector based detection of Glaucoma
Shi et al. Automated choroid segmentation in three-dimensional 1-μ m wide-view OCT images with gradient and regional costs
CN110309827B (en) Edema region segmentation model based on OCT image
Ganjee et al. Automatic segmentation of abnormal capillary nonperfusion regions in optical coherence tomography angiography images using marker-controlled watershed algorithm
Wu et al. Automated segmentation of intraretinal cystoid macular edema based on Gaussian mixture model

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