Department of Eye and Vision Science, Institute of Ageing and Chronic Disease, University of Liverpool, Liverpool, United Kingdom.
Am J Ophthalmol. 2013 Feb;155(2):277-286.e1. doi: 10.1016/j.ajo.2012.07.030. Epub 2012 Oct 27.
To evaluate a new computerized segmentation technique for the quantification of intraretinal and subretinal fluid in spectral-domain optical coherence tomography (SD OCT) images of the retina.
Prospective, cross-sectional study.
Thirty-seven B-scan images of 37 patients with exudative age-related macular degeneration were chosen randomly from SD OCT volume scans (1 per volume scan). All hyporeflective areas in the image first were segmented automatically as candidate regions by the program. Researchers who were masked to the candidate region information selected each fluid region from the original image using a single mouse click. The program then delineated the boundary of each region selected and calculated quantitative parameters, including total area of fluid regions if multiple regions were selected. The performance of our technique was validated by comparing the results with the measurements obtained from boundaries manually delineated by 2 masked observers. Time efficiency, agreement with manual delineation, and intraobserver and interobserver agreement of using the program were evaluated.
The proposed technique reduced the average processing time per image approximately 6-fold (15 seconds for computerized segmentation vs 90 seconds for manual delineation). There was good agreement between computerized segmentation and manual delineation measured by intraclass correlation coefficient (range, 0.897 to 0.979) and the Dice coefficient (range, 0.721 to 0.785). The proposed technique has excellent intraobserver and interobserver agreement (intraclass correlation coefficient range, 0.998 to 0.999; Dice coefficient range. 0.959 to 0.981).
This computerized segmentation method allows for accurate and fast quantification of fluid in retinal SD OCT images and could assist in monitoring disease progression and evaluating therapeutic intervention.
评估一种新的计算机化分割技术,用于对视网膜的光谱域光学相干断层扫描(SD OCT)图像中的视网膜内和视网膜下液进行定量分析。
前瞻性、横断面研究。
从 SD OCT 容积扫描中随机选择 37 例渗出性年龄相关性黄斑变性患者的 37 个 B 扫描图像(每个容积扫描 1 个)。首先,程序自动将图像中的所有低反射区域分割为候选区域。研究人员对候选区域信息进行屏蔽,然后使用单个鼠标点击从原始图像中选择每个液区。程序随后描绘每个所选区域的边界,并计算定量参数,如果选择了多个区域,则计算总液区面积。通过将结果与 2 名屏蔽观察者手动描绘的边界进行比较,验证了我们技术的性能。评估了该技术的时间效率、与手动描绘的一致性以及使用程序的观察者内和观察者间一致性。
与手动描绘相比,该技术将每个图像的平均处理时间缩短了约 6 倍(计算机化分割 15 秒,手动描绘 90 秒)。通过组内相关系数(范围为 0.897 至 0.979)和 Dice 系数(范围为 0.721 至 0.785)测量,计算机化分割与手动描绘具有良好的一致性。该技术具有极好的观察者内和观察者间一致性(组内相关系数范围为 0.998 至 0.999;Dice 系数范围为 0.959 至 0.981)。
这种计算机化分割方法可准确、快速地对视网膜 SD OCT 图像中的液区进行定量分析,有助于监测疾病进展和评估治疗干预效果。