Abstract
Purpose :
Quantitative analysis of choriocapillaris in optical coherence tomography angiography (OCTA) is challenging and has been mostly based on binary analysis of lighter versus darker pixels. Computerized texture analysis has been used in other medical imaging areas and allows discrimination of higher-order patterns in image data, including inter-pixel correlation, skew, and coarseness. We investigated the feasibility of applying this technique to OCTA images of choriocapillaris in normal subjects to extract quantitative metrics representing vascular patterns and perfusion.
Methods :
Observational cross-sectional study involving young healthy subjects with refractive error range from -1D to +1D. RTVue-XR Avanti was used to obtain en-face 3x3mm OCTA images of superficial choriocapillaris using automated segmentation. Images were analyzed using computerized texture-analysis software to extract texture features in five categories based on: 1) absolute value of gray levels (average gray level, gray level yielding 70% threshold or 30% threshold of the area under the histogram; 2) gray-level histogram analysis (balance and skewness); 3) spatial relationship among gray levels (coarseness, contrast, energy, entropy); 4) edge frequency (mean gradient as a function of distance between pixels); 5) fourier transform analysis (root-mean-square variation [RMS], first moment of power spectrum [FMP]).
Results :
Nineteen eyes of 10 young, healthy white subjects (3 men and 7 women) with age of 24.7± 2.5 yrs were included. The mean and standard deviation from mean were measured to assess the distribution and dispersion of data points in each metric among eyes as follows: Average gray level: 106.80 ± 4.8, gray level yielding 70% threshold: 78.3 ±4.2, 30% threshold: 130.89 ±6.3, balance: -0.96 0±0.005, skewness: 0.30± 0.09, energy: 0.01± 0.0005, entropy: 5.28± 0.055, contrast: 248156.4 ±23327.2, edge mean gradient: 0.03±0.035, RMS: 16086715.5± 447937, FMP: 0.04 ±0.001.
Conclusions :
To our knowledge, this is the first report using computerized texture analysis for en-face OCTA choriocapillaris images, showing it is feasible providing values that are coherent and tightly distributed around the mean (except for mean gradient) in a homogeneous healthy group. The next step is characterizing how well such analysis allows for discrimination of normal and abnormal vascular patterns.
This is an abstract that was submitted for the 2018 ARVO Annual Meeting, held in Honolulu, Hawaii, April 29 - May 3, 2018.