CN111862121B

CN111862121B - Level set algorithm-based segmentation method and measurement method for foveal avascular region

Info

Publication number: CN111862121B
Application number: CN202010720327.1A
Authority: CN
Inventors: 陈浩宇; 林艾迪
Original assignee: Shantou University Chinese University Of Hong Kong And Shantou International Ophthalmology Center
Current assignee: Shantou University Chinese University Of Hong Kong And Shantou International Ophthalmology Center
Priority date: 2020-07-24
Filing date: 2020-07-24
Publication date: 2023-09-26
Anticipated expiration: 2040-07-24
Also published as: CN111862121A

Abstract

The invention relates to the technical fields of medical image and computer digital image processing in the medical imaging field, in particular to a method for segmenting a fovea avascular region based on a Level set algorithm.

Description

Level set algorithm-based segmentation method and measurement method for foveal avascular region

Technical Field

The invention relates to the technical fields of medical images and computer digital image processing, in particular to a method for dividing a central concave avascular zone based on a Level set algorithm and a measuring method for measuring parameters such as the area, the perimeter, the roundness and the like of the central concave avascular zone.

Background

The foveal avascular zone (FAZ, foveal avascular zone) is the area of macular fovea that is bordered by the interdigitating network of capillaries located in the inner layers of the retina. The size and the shape of the compound can objectively reflect the severity of macular ischemia and capillary blood vessel loss, and are closely related to the prognosis of vision, so that the compound provides important clinical value for the disease progress and prognosis evaluation of a plurality of retinal vascular diseases, especially diabetic retinopathy.

Optical coherence tomography vascular imaging (OCTA, optical coherence tomography angiography) is a new technology developed on the basis of optical coherence tomography for diagnosis of ocular diseases, and can provide noninvasively and rapidly three-dimensional images of retinal and choroidal blood vessels, and repeated measurement and quantitative analysis of FAZ are possible through the images of the three-dimensional images, so that the understanding of various ocular fundus diseases by ophthalmic medical researchers is deepened.

Although the traditional manual FAZ segmentation method has good reliability, when different measuring staff conduct manual FAZ segmentation, certain measurement bias possibly exists due to personal subjective judgment, and the method is particularly obvious for images with unclear FAZ boundaries.

By adopting the method of automatically dividing the FAZ by the algorithm, not only the bias caused by different measurers can be avoided, but also the continuous automatic measurement operation can be realized, the time and the manpower resources can be greatly saved, and the requirements of clinical work can be met. For machines without self-contained segmentation algorithms or unsatisfactory segmentation effects of algorithms, many researchers have designed specific automatic FAZ segmentation algorithms to fit specific OCTA machines, such as Tang et al developed MATLAB Program for DRI OCT Triton machines, ishii et al developed Kanno-Saitama Macro Program (KSM Program) for Plex ellite 9000 machines, all with more desirable FAZ segmentation effects. However, the self-made automatic segmentation FAZ algorithm is usually a corresponding algorithm aiming at a specific OCTA machine at present, and the original image is required to be processed and then segmented, so that the accuracy and efficiency of the segmentation effect are reduced. Wherein, part of the algorithm belongs to a non-open source algorithm, and an ophthalmic medical researcher is difficult to develop an FAZ automatic segmentation algorithm suitable for other OCTA machines in a mode of adjusting the algorithm or parameters;

the other part of the control system of the OCTA machine is provided with a self-contained FAZ segmentation algorithm module, and is subjected to factory debugging and correspondingly matched with the OCTA machine, such as a self-contained FAZ segmentation algorithm module of Zeiss Cirrus HD-OCT 5000, but the algorithm often leads to an error segmentation result, so that the accuracy and reliability of FAZ segmentation are greatly reduced; moreover, the original FAZ image generated by the Zeiss Cirrus HD-OCT 5000 has the characteristic that the blood vessel ring of the FAZ boundary is discontinuous, so that the pixel difference between the FAZ and the non-perfusion region of the paracentric concave is small, and if the MATLAB Program or the KSM Program is simply applied, the non-perfusion region of the paracentric concave can be misjudged as the FAZ by using a pixel two-polarization method, so that an inaccurate segmentation effect is caused.

Disclosure of Invention

The invention aims to utilize common image processing software imageJ, combine the Level set algorithm and the measurement function provided by the software, and optimize parameters of curvature parameters, convergence parameters and gray scale parameters according to different characteristics of original FAZ images generated by different OTCA machines so as to develop a segmentation method of a fovea avascular zone respectively applicable to different OTCA machines, and obtain FAZ segmentation graphs corresponding to the fovea avascular zone in an automatic and efficient manner.

In order to achieve the above purpose, the invention is implemented by adopting the following technical means:

a method for segmenting a foveal avascular zone based on a Level set algorithm is implemented by the following steps:

(1) An OCTA machine is used for obtaining an original FAZ image of a macula lutea fovea, and the original FAZ image is converted into an 8-bit gray scale image form through image processing, so that an 8-bit gray scale FAZ image is obtained;

(2) Active Contours method in a Level set algorithm of imageJ software is selected, and parameters of curvature parameters (curvatures), convergence parameters (convergence) and gray parameters (gradeales) are correspondingly set according to different characteristics of 8-bit gray scale FAZ images generated by conversion of different OCTA machines, so that a Level set optimization algorithm is formed;

(3) The 8-bit gray FAZ image is imported into imageJ software, an initial sampling ring is positioned in a central concave avascular area of the 8-bit gray FAZ image, the initial sampling ring is operated by a Level set optimization algorithm, curvature weight of contour progress is determined by the initial sampling ring according to curvature parameter set values, and a fitting sampling ring which continuously performs iterative evolution towards the periphery is formed;

(4) In the iterative evolution process of the fitting sampling ring, if a Level set algorithm function corresponding to the fitting sampling ring contour is used, when the convergence degree change value of two adjacent iteration results is smaller than the convergence degree parameter set value or the difference value between the current gray Level value and the gray Level value of the next evolution is larger than the gray Level parameter set value, the fitting sampling ring stops evolving towards the periphery, the contour of the last fitting sampling ring is taken as a FAZ sampling ring, and image segmentation is carried out along the FAZ sampling ring to obtain a FAZ segmentation map.

Further, the FAZ image can be converted into the form of an 8-bit grayscale map by ImageJ.

Further, the curvature parameter is 0.5-2.0, the convergence parameter is 0.001-0.015, and the gray-scale parameter is 10-50.

Furthermore, the OCTA machine is Zeiss Cirrus HD-OCT 5000, and the signal intensity fraction of the original FAZ image is selected to be 6-10.

Further, the curvature parameter is 1.00, the convergence parameter is 0.0100, and the gray parameter is 30.

Further, the OCTA machine is DRI OCT Triton, and the picture quality score of the original FAZ image is selected to be 60-100.

Further, the curvature parameter is 1.50, the convergence parameter is 0.0015, and the gray parameter is 30.

Further, in the step (1), the initial sampling circle is located in the geometric center or the area with high signal to noise ratio of the 8-bit gray scale FAZ image.

Furthermore, the Level set optimization algorithm is designed into a macro program. As a preferable scheme, the macro program forms a Level set optimization algorithm by starting Active Contours in the Level Sets and optimizing and setting curvature parameters, convergence parameters and gray parameters, locates an initial sampling ring at the geometric center of an 8-bit gray FAZ image and runs the Level set optimization algorithm to form a fitting sampling ring which continuously and iteratively evolves towards the periphery, takes the contour of the last fitting sampling ring as an FAZ sampling ring, and performs image segmentation along the FAZ sampling ring to obtain an FAZ segmentation map.

The FAZ segmentation map obtained by the segmentation method of the fovea avascular zone based on the Level set algorithm can more accurately measure the area of the FAZ segmentation map by using any Image area measurement technology in the computer digital Image processing field, such as measurement functions provided by software such as Image-ProPlus, MATLAB, imageJ commonly used in the medical Image and computer digital Image processing technical field; meanwhile, based on the measuring function of the imageJ, the measuring of target parameters required by the area, perimeter, roundness and the like of the FAZ segmentation map can be realized, and a basis can be provided for analyzing the state of the fovea avascular zone in an intuitive and accurate numerical mode.

In addition, the macro program designed by the invention is designed into the macro program by the Level set optimization algorithm, and after the automatic segmentation of the FAZ segmentation map is completed, the measurement function provided by the image J can be further combined, and after the FAZ segmentation map is obtained, the automatic measurement of the area and/or the perimeter and/or the roundness of the FAZ segmentation map can be realized; preferably, after obtaining the FAZ segmentation map, the macro program may position the FAZ segmentation map at the geometric center of the image by using a magic wand, then extract the FAZ contour, create a mask (create mask), set a scale according to the FAZ segmentation map, and then automatically measure the perimeter and/or roundness.

The invention has the following advantages:

1. the method can adjust parameters according to the characteristics of different images shot by different OCTA machines, thereby achieving accurate segmentation effect and having good flexibility;

2. the algorithm can achieve a good anti-noise effect without denoising treatment, is simple to operate, is an open resource, is easy to obtain and has high segmentation efficiency;

3. compared with other algorithms, the algorithm has high sensitivity, is more suitable for images with fuzzy FAZ boundaries, and can restrict the progress direction of the segmentation contour by adjusting gray scale parameters and curvature parameters in the algorithm in images with discontinuous FAZ boundaries, so that the defect of inaccurate FAZ measurement results caused by 'out-of-range leakage' of the finally obtained FAZ segmentation map can be prevented;

4. if the pixel difference between the blood vessel and the non-perfusion region is not obvious, the binarization processing of the pixels may not be suitable for the FAZ image of the Zeiss Cirrus HD-OCT 5000 OCTA machine, and a Level set algorithm is adopted to achieve a more accurate and reliable segmentation result;

5. the macro program of the ImageJ combines the Level set algorithm with the measurement function, so that the effects of fully automatically obtaining the FAZ segmentation map and measuring the corresponding parameters can be achieved, and the operability and the automation degree of the method are further improved.

Drawings

FIG. 1 shows the FAZ segmentation effect of 8-bit gray-scale FAZ images obtained by different combinations of curvature parameters and convergence parameters in LSM Program on a Zeiss Cirrus HD-OCT 5000 machine, wherein the gray-scale variation inside and outside the FAZ boundary is large, and no region with higher local signal-to-noise ratio exists in the FAZ;

FIG. 2 shows the FAZ segmentation effect of an 8-bit gray-scale FAZ image obtained by setting different gray-scale parameters in an LSM Program on a Zeiss Cirrus HD-OCT 5000 machine, wherein the gray-scale change inside and outside the FAZ boundary is large and no region with high local signal-to-noise ratio exists in the FAZ;

FIG. 3 shows the FAZ segmentation effect of an 8-bit gray-scale FAZ image obtained by setting different gray-scale parameters in an LSM Program on a Zeiss Cirrus HD-OCT 5000 machine, wherein the difference of gray-scale values inside and outside the FAZ boundary is small, and no region with higher local signal-to-noise ratio exists in the FAZ;

FIG. 4 illustrates a segmentation process of an LSM Program for an 8-bit gray scale FAZ image obtained by a Zeiss Cirrus HD-OCT 5000 machine;

FIG. 5 shows a comparison of the effects of 8-bit gray scale FAZ images obtained by a Zeiss Cirrus HD-OCT 5000 machine for manual segmentation by two measurement personnel and FAZ segmentation by four automatic segmentation algorithms;

FIG. 6 shows a box plot of FAZ segmentation map area (area), perimeter (perimeter), roundness (circularity) measurements measured by different measurement methods in example 1;

FIG. 7 shows a linear regression plot of area measurements of FAZ segmentation plots from different measurement methods in example 1;

FIG. 8 shows Bland-Altman plots of area measurements of FAZ segmentation plots measured by different methods of measurement in example 1;

FIG. 9 shows a linear regression plot of perimeter measurements of FAZ segmentation plots measured by different methods of measurement in example 1;

FIG. 10 shows Bland-Altman plots of perimeter measurements of FAZ segmentation plots measured by different methods of measurement in example 1;

FIG. 11 is a linear regression diagram showing the roundness measurements of FAZ segmentation maps obtained by the different measurement methods of example 1;

FIG. 12 is a Bland-Altman plot showing the roundness measurements of FAZ segmentation maps obtained by the different methods of measurement in example 1;

FIG. 13 illustrates the FAZ segmentation effect of 8-bit gray scale FAZ images obtained for DRI OCT Triton machines from different combinations of curvature parameters and convergence parameters in LSM Program, with large difference in gray scale values inside and outside the FAZ boundary and no local high signal-to-noise ratio region in the FAZ;

FIG. 14 shows the FAZ segmentation effect of an 8-bit gray FAZ image obtained by setting different gray scale parameters in an LSM Program for a DRI OCT Triton machine, wherein the gray scale value difference between the inside and outside of the FAZ boundary is large and no local high signal-to-noise ratio region exists in the FAZ;

FIG. 15 shows the FAZ segmentation effect of an 8-bit gray FAZ image obtained by setting different gray scale parameters in an LSM Program on a DRI OCT Triton machine, wherein the gray scale value difference inside and outside the FAZ boundary is large and a local high signal to noise ratio region exists in the FAZ;

FIG. 16 illustrates a segmentation process of the DRI OCT Triton machine by LSM Program to obtain 8-bit gray scale FAZ images;

FIG. 17 shows a comparison of the effect of a DRI OCT Triton machine to obtain an 8-bit gray scale FAZ image for manual segmentation by two measurement personnel and for segmentation of the FAZ by two automatic segmentation algorithms;

FIG. 18 is a box plot showing the area (area), perimeter (circumference), and roundness (circularity) measurements of FAZ segmentation plots measured by different measurement methods in example 2;

FIG. 19 is a linear regression graph showing the area measurements of the FAZ segmentation map obtained by the different measurement methods of example 2;

FIG. 20 is a Bland-Altman plot showing area measurements of FAZ segmentation plots from different measurements in example 2;

FIG. 21 is a linear regression graph showing perimeter measurements of FAZ segmentation maps obtained by the different measurement methods of example 2;

FIG. 22 shows Bland-Altman plots of perimeter measurements of FAZ segmentation plots measured by different methods of measurement in example 2;

FIG. 23 is a linear regression graph showing the roundness measurements of FAZ cut maps obtained by the different measurement methods of example 2;

FIG. 24 is a Bland-Altman plot showing the roundness measurements of FAZ segmentation maps obtained by the different methods of measurement in example 2;

in the above figures:

M ₁ representing the measurement result of the tester 1;

M ₂ representing the measurement result of the tester 2;

m represents M ₁ 、M ₂ An average value of the manual measurement results;

l represents an automatic measurement result by the LSM Program;

z represents the automatic measurement by the FAZ segmentation algorithm (Cirrus inbuilt algorithm) carried by the Zeiss Cirrus HD-OCT 5000 machine;

k represents the automatic measurement result by KSM Program;

t represents the results of the automatic measurement by MATLAB Program.

Detailed Description

For the purpose of making the objects, technical solutions and advantages of the present invention more apparent, the following description of embodiments and reliability evaluation of the present invention will be given with reference to the accompanying drawings:

example 1

(1) Establishment and measurement of LSM Program for Zeiss Cirrus HD-OCT 5000 machine

(1) An en-face image of a central concave shallow capillary vessel layer is acquired by using a Zeiss Cirrus HD-OCT 5000 machine, and a scanning mode of macula lutea 3mm multiplied by 3mm is selected to acquire a test area; based on a machine scoring mode of a Zeiss Cirrus HD-OCT 5000 machine, an original FAZ image with a signal intensity score range of 6-10 can be selected to be used as a picture to be measured for segmenting and measuring FAZ;

(2) Active Contours method in a Level set algorithm of imageJ software is selected, curvature parameters are 0.5, convergence parameters are 0.0050 and gray scale parameters are 30 to serve as parameter adjustment examples, and the Level set algorithm is adjusted to obtain a corresponding Level set test algorithm;

(3) During testing, an original FAZ image is imported into imageJ software in the form of an 8-bit gray Level image, an initial sampling ring is positioned in the center of a central concave avascular zone of the 8-bit gray Level FAZ image, the Level set test algorithm is operated, the initial sampling ring determines curvature weight of contour progress according to curvature parameter set values, and iterative evolution is continuously carried out to the periphery to form a fitting sampling ring; in the iterative evolution process of the fitting sampling ring, if a Level set algorithm function corresponding to the profile of the fitting sampling ring is used, when the convergence degree change value of two adjacent iteration results is smaller than the convergence degree parameter set value or the difference value between the current gray Level value and the gray Level value of the next evolution is larger than the gray Level parameter set value, the fitting sampling ring stops evolving towards the periphery, the profile of the last fitting sampling ring is taken as the FAZ sampling ring, image segmentation is carried out along the FAZ sampling ring, and the FAZ segmentation test chart is obtained. If the gray scale contrast inside and outside the FAZ boundary has obvious difference and no local area with high signal to noise ratio exists in the FAZ, different gray scale values are selected to have small influence on the segmentation effect, otherwise, if the gray scale difference inside and outside the FAZ boundary is not obvious or the local area with high signal to noise ratio exists in the FAZ, the accuracy of the segmentation of the FAZ is improved by optimizing the gray scale parameter values;

(4) For the convenience of observation, the central concave avascular area of the 8-bit gray-scale FAZ image is magnified. Referring to fig. 1 in detail, in this embodiment, an image of a region with a large gray scale change inside and outside the FAZ boundary and no local signal-to-noise ratio in the FAZ is selected as the target test chart. Selecting moderate gray scale parameters such as 30 as constant value, adjusting different curvature parameters and convergence parameters as shown in fig. 1, and screening different FAZ segmentation test patterns formed by the same 8-bit gray scale FAZ image to determine the optimal combination of curvature and convergence parameters;

(5) And (3) under the condition that the curvature parameter is 1.00 and the convergence parameter is 0.0100, carrying out gray parameter optimization on the target side view. Selecting a Level set test algorithm with gray parameters of 10, 30 and 50 correspondingly, and importing the Level set test algorithm with gray parameters modified again, wherein the measured result is as shown in fig. 2, so that the segmentation effect of different gray parameters is accurate;

(6) For an original FAZ image obtained by a Zeiss Cirrus HD-OCT 5000 machine, after the original FAZ image with partial signal intensity fraction in the range of 6-8 is converted into an 8-bit gray FAZ image, the gray value difference inside and outside the FAZ boundary can possibly be smaller, so that the optimized gray value parameter mainly aims at an image with insignificant gray value difference inside and outside the FAZ. Referring to fig. 3 in detail, in this embodiment, an image with a smaller difference between gray values inside and outside the FAZ boundary is selected as a target test image, and the gray parameters are optimized under the condition that the curvature parameter is 1.00 and the convergence parameter is 0.0100. Selecting a Level set test algorithm with gray parameters of 10, 30 and 50 correspondingly, importing the Level set test algorithm with gray parameters modified again, measuring the result as shown in figure 3, and determining the optimal gray parameter as 30;

(7) Through evaluation, a Level set optimization algorithm is established by selecting a curvature parameter of 1.00, a convergence parameter of 0.0100 and a gray scale parameter of 30 as optimized parameter values, and a macro Program (LSM Program) is established based on the Level set optimization algorithm, and is specifically as follows:

run("Level Sets"， "method=[Active Contours] use_level_sets grey_value_threshold=50 distance_threshold=0.50 advection=2.20 propagation=1 curvature=1.00 grayscale=30 convergence=0.0100 region=outside")；

//setTool("wand")；

doWand(360， 360)；

run("Create Mask")；

run("Set Scale..."， "distance=719 known=3 unit=mm")；

run("Set Measurements..."， "area perimeter shape redirect=None decimal=4")；

run("Analyze Particles..."， "display")；

close()；

run("Open Next")；

in the macro program, besides adjusting the curvature parameter, the convergence parameter and the gray scale parameter, the geometric center of the image is determined according to the actual pixel sizes of the original FAZ images obtained by different OCTA machines, so that the segmented FAZ is extracted, and further measurement is facilitated. As in this example 1, the original FAZ pixel size of the Zeiss Cirrus HD-OCT 5000 machine is 719 x 719 pixels with image geometry centered coordinates (360 ), and dowands inserted coordinates (360 ). Since the original FAZ image size of the Zeiss Cirrus HD-OCT 5000 machine is 719X 719 pixels, representing a 3X 3mm region of the fovea, the following is set in the macro procedure: "Set scale.," distance=719 known=3 unit=mm ", so that the measurement result is measured in" mm ". The above parameter settings are reasonably adjusted according to the original FAZ image size, which is a technical knowledge well known to those skilled in the art of computer digital image processing, and will be better understood by brief description herein.

By this macro procedure it will be possible to constantly "constrain" the boundaries of the fitted sampling circle according to the set parameters, showing an evolving fitted sampling circle profile in the window "Segemntation progress". When it detects the boundary of the fovea avascular zone, the procedure will stop, and fig. 4 shows the FAZ sampling circle profile (FAZ contour) and the FAZ segmentation map (FAZ segmentation) which are finally formed after the iterative evolution of the initial sampling circle profile, respectively.

After the FAZ segmentation is finished, the LSM Program can automatically output the measurement results of the parameters such as the area, the perimeter, the roundness and the like of the FAZ segmentation map based on the measurement function provided by the imageJ, and automatically open the next FAZ image to be segmented for the next automatic segmentation and measurement; of course, after obtaining the FAZ segmentation map as shown in fig. 4c each time, the FAZ segmentation map may be imported into a medical Image having an Image area measurement function such as Image-ProPlus, MATLAB, or a computer digital Image processing software to perform the area measurement of the FAZ, and the area of the FAZ segmentation map may be accurately measured.

After the FAZ segmentation is finished, the LSM Program automatically outputs the measurement results of the area, perimeter, roundness and other parameters of the FAZ segmentation map based on the measurement function provided by the imageJ, and automatically opens the next FAZ image to be segmented for the next automatic segmentation and test.

(2) Assessment of FAZ segmentation effect based on different measurement methods

An original FAZ image is obtained through a Zeiss Cirrus HD-OCT 5000 machine, an 8-bit gray-scale FAZ image is formed through imageJ processing, the FAZ is manually segmented by a tester 1 and a tester 2, and the FAZ segmentation is correspondingly carried out on the same original FAZ image through an image processing method corresponding to an LSM Program, a KSM Program and a MATLAB Program provided by the Zeiss Cirrus HD-OCT 5000 machine.

As shown in fig. 5, the segmentation result of the LSM Program is very similar to the artificial segmentation result, while Cirrus inbuilt algorithm is wrong in the segmentation process, and the KSM Program and the MATLAB Program have a "void" phenomenon or a "boundary crossing" phenomenon in the segmentation process to different degrees, resulting in inaccurate segmentation results. As shown in fig. 5, the segmentation result of the LSM Program is less prone to "cavitation";

since it is difficult for the manual segmentation to achieve sufficient capture of each point on the FAZ boundary, the outer contour of the FAZ segmentation map obtained by the manual segmentation tends more to be "smooth", the outer contour shape tends more to be "round", numerically appears to be more prone to 1 in measured roundness, and shows a shortage of being small in circumference relative to the true fovea avascular zone;

compared with a manual segmentation method, the LSM Program can capture each point on the FAZ boundary more sensitively and form a zigzag FAZ boundary more easily, so that marking and sampling can be carried out along the gaps among blood vessels at the edge of the central concave avascular zone, and the measured perimeter is more accurate; in addition, due to reasonable setting of curvature parameters, the outer contour in the iterative evolution process of the FAZ sampling ring is restrained, so that the FAZ sampling ring can be more accurately segmented along the FAZ boundary without generating the phenomenon of 'out-of-boundary leakage';

(3) repeatability assessment of measurement results based on different measurement methods

Selecting 37 healthy subjects, selecting one eye after sufficient mydriasis, repeatedly shooting four times by using a Zeiss Cirrus HD-OCT 5000 machine, and obtaining original FAZ images, wherein 2 eyes cannot detect the FAZ boundary due to an automatic FAZ segmentation algorithm provided by the Zeiss Cirrus HD-OCT 5000 machine, and 1 eye is excluded due to poor image quality (the signal intensity fraction of the image is less than 6), so that the original FAZ images of 34 eyes of 34 healthy subjects are taken as test targets.

After the original FAZ images are disordered, the 8-bit gray-scale FAZ images are processed through image processing software, the FAZ images are manually segmented by a tester 1 and a tester 2, two measured values of each image are averaged, and as a result of manual measurement, the image processing methods corresponding to an automatic FAZ segmentation algorithm (Cirrus inbuilt algorithm), a KSM Program and a MATLAB Program provided by an LSM Program and a Zeiss Cirrus HD-OCT 5000 machine are correspondingly tested, and specific test results are shown in table 1:

TABLE 1 repeatability assessment of image measurements taken by various FAZ segmentation methods on Zeiss Cirrus HD-OCT 5000

Wherein, area: an area; perimeter: perimeter; circling quality: roundness; m is M ₁ : measurement results of test person 1, M ₂ : the test person 2 measures the result; m: m is M ₁ 、M ₂ Average value of measurement results; l: LSM Program; z: cirrus inbuilt algorithm; k: KSM Program; t: MATLAB Program; mean: an average value; SD: standard deviation; coV: coefficient of variation; ICC: intra-group correlation coefficients; 95% CI:95% confidence interval.

As is clear from the data in table 1, the LSM Program has the best repeatability among all the automatic segmentation algorithms, and the distribution of the LSM Program measurement results is closest to the distribution of the manual measurement results among the three measurement parameters of the area, perimeter, and roundness of the FAZ segmentation map as shown in fig. 6.

(4) Consistency analysis of measurement results based on different measurement methods:

TABLE 2 evaluation of the consistency of the image measurements taken by the various FAZ segmentation methods on Zeiss Cirrus HD-OCT 5000

Wherein, area: an area; perimeter: perimeter; circling quality: roundness; m is M ₁ : measurement results of test person 1, M ₂ : the test person 2 measures the result; m: m is M ₁ 、M ₂ Average value of measurement results; l: LSM Program; z: cirrus inbuilt algorithm; k: KSM Program; t: MATLAB Program; paired t-test: pairing t-test; ICC: intra-group correlation coefficients; 95% CI:95% confidence interval; 95% limits of agreement:95% consistency limit; lower bound: a lower limit; upper bound: an upper limit; bias: bias.

By all automatic segmentation algorithms for Zeiss Cirrus HD-OCT 5000, as shown in table 2 and fig. 7-12, the LSM Program is best consistent with the manual segmentation results.

Example 2

(1) Establishment and measurement of DRI OCT Triton machine LSM Program

(1) Selecting a DRI OCT Triton machine to obtain an en-face image of a central concave shallow capillary vessel layer, and selecting a scanning mode of macula lutea 3mm multiplied by 3mm to obtain a test area; based on the machine scoring mode of the DRI OCT Triton machine, the original FAZ image with the image quality score range of 60-100 can be selected as the target test chart of the embodiment;

(2) Active Contours method in a Level set algorithm of imageJ software is selected, a curvature parameter is 1.0, a convergence parameter is 0.0010, and a gray scale parameter 30 is taken as a parameter adjustment example, and the Level set algorithm is adjusted to obtain a corresponding Level set test algorithm;

(3) During testing, converting an original FAZ image into an 8-bit gray Level image form through image processing software to obtain an 8-bit gray Level FAZ image, importing the 8-bit gray Level FAZ image into imageJ software, positioning an initial sampling ring in the center of a central concave avascular zone of the 8-bit gray Level FAZ image, running a Level set test algorithm for modifying parameters, determining curvature weight of contour progress by the initial sampling ring according to curvature parameter set values, and continuously performing iterative evolution to the periphery to form a fitting sampling ring; in the iterative evolution process of the fitting sampling ring, if a Level set algorithm function corresponding to the profile of the fitting sampling ring is used, when the convergence degree change value of two adjacent iteration results is smaller than a convergence degree parameter set value or the difference value between the current gray Level value and the gray Level value of the next evolution is larger than the gray Level parameter set value, stopping the fitting sampling ring from evolving towards the periphery, taking the profile of the last fitting sampling ring as an FAZ sampling ring, and carrying out image segmentation along the FAZ sampling ring to obtain an FAZ segmentation test chart;

(4) Referring to fig. 13 in detail, in this embodiment, an image of a region with larger gray level variation inside and outside the FAZ boundary and no local signal-to-noise ratio in the FAZ is selected as a target test chart, a proper fixed gray level parameter such as 30 is selected as a fixed value, different curvature parameters and convergence parameters are adjusted as shown in fig. 13, different FAZ segmentation test charts are formed on the same 8-bit gray level FAZ image, and an optimal combination of curvature and convergence parameters is determined; under the condition that the curvature parameter is 1.5 and the convergence parameter is 0.0015, the gray scale parameters are optimized, in the embodiment, the gray scale parameters are 10, 30 and 50, a Level set test algorithm after the gray scale parameters are modified is imported again, and the measured result is as shown in fig. 14, so that the segmentation effect of different gray scale parameters is accurate;

(5) For an original FAZ image obtained by a DRI OCT Triton machine, the obtained original FAZ image with partial image quality score ranging from 60 to 80 has the characteristic that a local high signal to noise ratio area is easy to generate in the FAZ, and the obtained original FAZ image still exists in the local high signal to noise ratio area in the FAZ after being converted into an 8-bit gray FAZ image, so that the optimization of gray value parameters mainly aims at the problem of high signal to noise ratio. Referring to fig. 15 in detail, in this embodiment, an image with a local high signal-to-noise ratio region in the FAZ is selected as a target test image;

(6) Under the condition that the curvature parameter is 1.5 and the convergence parameter is 0.0015, the gray scale parameter is optimized, in the embodiment, the gray scale parameters are 10, 30 and 50, a Level set test algorithm after the gray scale parameter modification is imported again, the result is shown in fig. 15, and the optimal gray scale parameter is determined to be 30;

(7) Through evaluation, a Level set optimization algorithm is established by selecting parameter values with curvature parameter of 1.5, convergence parameter of 0.0015 and gray scale parameter of 30 as optimized parameter values, and a macro Program (LSM Program) suitable for the FAZ image of the OCTA machine is established based on the Level set optimization algorithm, and is specifically as follows:

run("Level Sets"， "method=[Active Contours] use_level_sets grey_value_threshold=50 distance_threshold=0.50 advection=2.20 propagation=1 curvature=1.50 grayscale=30 convergence=0.0015 region=outside")；

//setTool("wand")；

doWand(160， 160)；

run("Create Mask")；

run("Set Scale..."， "distance=320 known=3 unit=mm")；

run("Set Measurements..."， "area perimeter shape redirect=None decimal=4")；

run("Analyze Particles..."， "display")；

close()；

run("Open Next")；

in this example 2, the DRI OCT Triton machine has an original FAZ pixel size of 320×320 pixels, its image geometry center coordinates (160 ), and doWand inserted coordinates (160 ). Since the DRI OCT Triton machine original FAZ image size is 320×320 pixels, representing a 3×3mm region of the fovea, the macro procedure is set as follows: "Set scale.," distance=320 known=3 unit=mm ", and the measurement result is measured in" mm ".

By this macro procedure it will be possible to constantly "constrain" the boundaries of the fitted sampling circle according to the set parameters, showing an evolving fitted sampling circle profile in the window "Segemntation progress". When the central fovea avascular zone boundary is detected, the program stops running, and an FAZ sampling circle contour (FAZ contour) and an FAZ segmentation map (FAZ segmentation) which are finally formed after the iterative evolution of the initial sampling circle contour are respectively shown in FIG. 16; for the image obtained by DRI OCT Triton, as shown in the central concave avascular region of the 8-bit gray scale FAZ image corresponding to FIG. 15, a region with higher signal to noise ratio exists, and for other algorithms such as KSM Program, MATLAB Program and the like, the region needs to be subjected to denoising treatment so as to achieve a better segmentation effect, but the LSM Program algorithm does not need to perform the operation, if the original FAZ image exists, the segmentation effect is influenced by the region with particularly high local signal to noise ratio, and the region can be included when the initial sampling ring is positioned, namely the equivalent denoising effect can be achieved without denoising treatment.

Similarly, after obtaining the FAZ segmentation map as shown in fig. 16c, the FAZ segmentation map may be imported into a medical Image having an Image area measurement function such as Image-ProPlus, MATLAB, or a computer digital Image processing software to perform the area measurement of the FAZ, so that the area of the FAZ segmentation map may be accurately measured.

An original FAZ image is obtained through a DRI OCT Triton machine, an 8-bit gray-scale FAZ image is formed through image processing software, the FAZ is manually segmented by a tester 1 and a tester 2, and the FAZ segmentation is correspondingly carried out on the same original FAZ image through an image processing method corresponding to an LSM Program and an MATLAB Program, wherein the specific effect is as shown in figure 17.

As shown in fig. 17, the FAZ boundary of the Triton DRI OCTA image is more distinct and continuous, and the FAZ profile of the LSM Program segmentation is smoother and very similar to the artificial segmentation result, compared to the FAZ segmentation map obtained by the LSM Program segmentation of example 1.

Taking 37 eyes of 37 healthy subjects as test targets, taking four times of continuous shooting by using a DRI OCT Triton machine by each subject, obtaining an original FAZ image with the picture quality fraction of 60-100, correspondingly importing image J, then converting the image into an 8-bit gray-scale FAZ image, manually dividing the FAZ by using a tester 1 and a tester 2, correspondingly testing by using an image processing method corresponding to an LSM Program and a MATLAB Program, and specifically testing the FAZ image according to the specific test results shown in Table 3:

TABLE 3 repeatability assessment of DRI OCT Triton acquired image measurements by various FAZ segmentation methods

Wherein, area: an area; perimeter: perimeter; circling quality: roundness; m is M ₁ : measurement results of test person 1, M ₂ : the test person 2 measures the result; m: m is M ₁ 、M ₂ Average value of measurement results; l: LSM Program; t: MATLAB Program; mean: an average value; SD: standard deviation; coV: coefficient of variation; ICC: intra-group correlation coefficients; 95% CI:95% confidence interval.

As is clear from the data in table 3, the LSM Program has the best repeatability among all the automatic segmentation algorithms, and the distribution of the LSM Program measurement results is closest to the distribution of the manual measurement results among three measurement parameters of the area, perimeter, and roundness of the FAZ segmentation map as shown in fig. 18.

TABLE 4 consistency assessment of DRI OCT Triton acquired image measurements by various FAZ segmentation methods

area: an area; perimeter: perimeter; circling quality: roundness; m is M ₁ : measurement results of test person 1, M ₂ : the test person 2 measures the result; m: m is M ₁ 、M ₂ Average value of measurement results; l: LSM Program; t: MATLAB Program; paired t-test: pairing t-test; ICC: group ofAn inner correlation coefficient; 95% CI:95% confidence interval; 95% limits of agreement:95% consistency limit; lower bound: a lower limit; upper bound: an upper limit; bias: bias.

By all automatic segmentation algorithms for DRI OCT Triton, the segmentation results of LSM Program and MATLAB Program are very close to the manual segmentation results, as shown in table 4 and fig. 19-24.

The basic principle, main characteristics and advantages of the invention are described above, and the reliability of the invention is proved by comparison with manual measurement and other automatic algorithm measurement. The above description describes the application of the method to both Zeiss Cirrus HD-OCT 5000 and DRI OCT Triton OCTA machines, whereby the principles and applications of the present invention are described, and the present invention may also be adapted to corresponding parameter adjustments and technical adjustments according to the different features of the images taken by other OCTA machines without departing from the spirit and scope of the present invention, which shall be considered to fall within the scope of the invention as claimed and by the appended claims and their equivalents.

Claims

1. A method for segmenting a foveal avascular zone based on a Level set algorithm is characterized by comprising the following steps:

an OCTA machine is used for obtaining an original FAZ image of a macula lutea fovea, and the original FAZ image is converted into an 8-bit gray scale image form through image processing, so that an 8-bit gray scale FAZ image is obtained;

active Contours method in a Level set algorithm of imageJ software is selected, and parameters of curvature parameters, convergence parameters and gray parameters are correspondingly set according to different characteristics of 8-bit gray FAZ images generated by different OCTA machines, so that a Level set optimization algorithm is formed;

importing the 8-bit gray FAZ image into imageJ software, positioning an initial sampling ring in a central concave avascular area of the 8-bit gray FAZ image, operating by a Level set optimization algorithm, and adjusting curvature weights of contour iterative evolution by the initial sampling ring according to curvature parameter set values to form a fitting sampling ring which continuously performs iterative evolution towards the periphery;

in the iterative evolution process of the fitting sampling ring, if a Level set algorithm function corresponding to the fitting sampling ring contour is used, when the convergence degree change value of two adjacent iteration results is smaller than the convergence degree parameter set value or the difference value between the current gray Level value and the gray Level value of the next evolution is larger than the gray Level parameter set value, the fitting sampling ring stops evolving towards the periphery, the contour of the last fitting sampling ring is taken as a FAZ sampling ring, and image segmentation is carried out along the FAZ sampling ring to obtain a FAZ segmentation map.

2. The method for segmenting the foveal avascular zone based on the Level set algorithm as claimed in claim 1, wherein the method comprises the following steps: the FAZ image was converted into the form of an 8-bit gray scale map by ImageJ.

3. The method for segmenting the foveal avascular zone based on the Level set algorithm as claimed in claim 1, wherein the method comprises the following steps: the curvature parameter is 0.5-2.0, the convergence degree parameter is 0.001-0.015, and the gray scale parameter is 10-50.

4. The method for segmenting the foveal avascular zone based on the Level set algorithm as claimed in claim 3, wherein the method comprises the following steps of: the OCTA machine is Zeiss Cirrus HD-OCT 5000, and the signal intensity fraction of the original FAZ image is 6-10.

5. The method for segmenting the foveal avascular zone based on the Level set algorithm as claimed in claim 4, wherein the method comprises the following steps: the curvature parameter is 1.00, the convergence parameter is 0.0100, and the gray parameter is 30.

6. The method for segmenting the foveal avascular zone based on the Level set algorithm as claimed in claim 3, wherein the method comprises the following steps of: the OCTA machine is DRI OCT Triton, and the picture quality score of the original FAZ image is 60-100.

7. The method for segmenting the foveal avascular zone based on the Level set algorithm as claimed in claim 6, wherein the method comprises the following steps: the curvature parameter is 1.50, the convergence parameter is 0.0015, and the gray scale parameter is 30.

8. The method for segmenting the foveal avascular zone based on the Level set algorithm as set forth in any one of claims 1 to 7, wherein: the initial sampling ring is positioned in the geometric center of the 8-bit gray FAZ image or in the area with high signal to noise ratio.

9. The method for segmenting the foveal avascular zone based on the Level set algorithm as set forth in any one of claims 1 to 8, wherein: and designing the Level set optimization algorithm into a macro program.

10. A measuring method of a fovea avascular zone based on a Level set algorithm is characterized by comprising the following steps of: the area of the FAZ segmentation map obtained by the segmentation method as set forth in any one of claims 1 to 9 is measured by any image area measurement technique in the field of computer digital image processing.

11. The Level set algorithm-based foveal avascular zone measurement method of claim 10, wherein: by utilizing the measurement function provided by ImageJ, a macro program is designed to automatically measure the area and/or circumference and/or roundness of the FAZ segmentation map.