Triton光学相干断层扫描血管成像图像中正常人的中心凹无血管区的自动测量方法

• 作 者: 林艾迪 方丹齐 吴苇杭 容毅标 陈浩宇
• 机 构: 汕头大学·香港中文大学联合汕头国际眼科中心眼外伤科,广东 汕头 515041;汕头大学电子信息工程系,广东 汕头 515063
• 出 处: 眼科学报
• 发表日期: 2022年1期
• 关键词: foveal avascular zone optical coherence tomography angiography automated measure

背景:目前已有研究报道了一种MATLAB的定制算法,用于Triton光学相干断层扫描血管成像(optical coherence tomography angiography,OCTA)图像的中心凹无血管区(fovea avascular zone,FAZ)的自动测量.由于这种算法非开源,且难以获取,因而大大限制了其在临床实践和科学研究中的应用.本研究提出一种用于Triton OCTA图像的FAZ自动分割的开源算法,即Smooth Level Sets macro(SLSM)算法,并将其测量结果与MATLAB和人工方法相比较,评估该算法分割的准确性和可靠性.方法:纳入35位健康受试者的35只健眼,选用Triton OCTA机器中的3 mm×3 mm扫描模式,对其黄斑区进行连续4次扫描.分别用人工和自动方法(包括MATLAB和SLSM),测量浅层毛细血管图像中FA Z的面积、周长和圆度.分析各种自动算法的准确性、重复性,以及与人工方法结果的一致性.结果:SLSM算法的准确性仅低于人工方法,而高于MATLAB算法(Dice系数:人工方法,0.9568;SLSM,0.9506;MATLAB,0.9483).SLSM和MATLAB测量FAZ面积的重复性均很高[组内相关系数(intraclass correlation coefficient,ICC):SLSM,0.987;MATLAB,0.983].SLSM、MATLAB测量FAZ面积的结果均与人工方法呈很高的一致性(ICC:SLSM,0.973;MATLAB,0.968).结论:SLSM在Triton OCTA图像的FAZ自动分割中的准确性高于MATLAB,其测量结果与人工测量结果很相近.作为免费和开源的资源,SLSM有望成为Triton OCTA图像中有效可靠的FA Z自动分割和测量方法.“,”Background: Previous studies have proposed an automated customized program named MATLAB used in the foveal avascular zone (FAZ) measurements in Triton optical coherence tomography angiography (OCTA) images. But it is not open-source and not easy to obtain, which will largely restrict its application in clinical practice and medical research. In this study, we aimed to investigate the feasibility of the Smooth Level Sets macro (SLSM), a free and open-source program, and compared with the manual measurements and MATLAB in the FAZ quantification in Triton OCTA. Methods: Thirty-five eyes of 35 healthy subjects were scanned four times continuously using Triton OCTA. Manual and automated methods including the SLSM and MATLAB were used in the FAZ metrics (area, perimeter, and circularity) of the superficial capillary plexus. The accuracy, repeatability of all methods, and agreement between automated and manual methods were analyzed. Results: The SLSM presented higher accuracy with a higher average Dice coefficient (0.9506) than MATLAB (0.9483), which was just second to the manual method (0.9568). Both the SLSM [intraclass correlation coefficient (ICC) =0.987; coefficient of variation (CoV) =3.935％] and MATLAB (ICC =0.983; CoV =4.165％) showed excellent repeatability for the FAZ area. They also had excellent agreement with manual measurement (SLSM, ICC =0.973; MATLAB, ICC =0.968). Conclusion: The SLSM exhibits better accuracy than MATLAB in the automated FAZ measurement in Triton OCTA, the results of which were comparable to those obtained by manual measurement. This free and open-source program may be an accessible and feasible option for automated FAZ segmentation on Triton OCTA images.