Abstract
Purpose :
Metrics quantifying the Foveal Avascular Zone (FAZ) in Optical Coherence Tomography Angiography (OCTA) images may be useful biomarkers for retinal vascular diseases. However, robust estimation of FAZ metrics is challenged by image noise and blur. This study investigated an automated frame averaging and segmentation technique for FAZ metric estimation.
Methods :
Ten OCTA image frames were acquired for each of 19 subjects using the Optovue’s AngioVue (Fremont, CA). For each subject, the five frames with highest image quality, as quantified by mean gradient magnitude, were sorted by image quality and spatially registered (StackReg: Rigid Body, ImageJ). Average images were calculated as the mean of one to five registered frames. The averaged images were input to an automated FAZ segmentation algorithm based on vessel edge detection. For each subject, three masked, expert readers manually segmented the FAZ on the averaged image of five frames. The expert segmentations were statistically combined to form a ground truth FAZ region. The agreement of the algorithm FAZ segmentation to the ground truth was quantified by Dice coefficient. Metrics of FAZ area, perimeter, and centroid were calculated for algorithm FAZ segmentations and compared to ground truth values.
Results :
The automated algorithm identified the correct FAZ region in 18 of 19 subjects. One subject presented an ambiguous FAZ with small avascular regions in the central fovea. The mean Dice coefficients of agreement between the algorithm and expert segmentations were 0.94±0.04 when using the best frame and 0.96±0.04 when averaging three or more frames. Averaging three or more frames reduced the error in the FAZ metrics (see Table). For example, the error in the estimated FAZ area was 6.2% when using one frame compared to 3.2% for three frames and 2.6% error when averaging five frames.
Conclusions :
Averaging three to five OCTA frames improved the accuracy of FAZ metrics estimated by the automated FAZ segmentation algorithm. The proposed technique may facilitate robust estimation of FAZ biomarkers for evaluating retinal vascular diseases and monitoring therapeutic interventions.
This is an abstract that was submitted for the 2017 ARVO Annual Meeting, held in Baltimore, MD, May 7-11, 2017.