Assessment of potential vessel segmentation pitfalls in the analysis of blood flow velocity using the Retinal Function Imager

作者：Somfai Gabor Mark; Tian Jing; Debuc Delia Cabrera^*

来源：Graefes Archive for Clinical and Experimental Ophthalmology, 2016, 254(6): 1075-1081.

DOI：10.1007/s00417-015-3166-0

• Retinal blood flow
• Retinal Function Imager
• Reproducibility
• Image analysis

The purpose of our study was to investigate the potential pitfalls associated with different vessel segmentation methods using the built-in software of the Retinal Function Imager (RFI) for the analysis of retinal blood flow velocity (BFV).
Ten eyes of nine healthy subjects were enrolled in the study. Retinal blood flow measurements were obtained with the RFI device with a 20A degrees field of view imaging. The same grader segmented the retinal vasculature using the RFI software in both sessions, with segments ranging in length from 50 to 100 pixels ("short segments") and 100-200 pixels ("long segments"). The blood flow velocities for the arteriolar and venular system were calculated, and the percentage of excluded vessel segments with high coefficients of variation (> 45 %) was recorded and compared by paired t test. Spearman's correlation was used to analyze the relationship between measurements by the two vessel segmentation methods.
The number of analyzed vessel segments did not differ significantly between the two groups (28.6 +/- 2.6 short and 26.7 +/- 4.6 long segments), while the percentage of acceptable segments was significantly higher in the long segment group (65.2 +/- 11.4 % vs 85.2 +/- 5.87 %, p = 0.001). In the short segment group, more than 15 % of vessel segments were rejected in all subjects, while in the long segment group only three subjects had a rejection rate of greater than 15 % (16.7 %, 18.7 % and 28 %). Both arteriolar and venular velocities were lower in the short segment group, although it reached significance only for arteriolar velocities (3.93 +/- 0.55 vs. 4.45 +/- 0.76 mm/s, p = 0.036 and 2.95 +/- 0.56 vs. 3.17 +/- 0.84 mm/s, p = 0.201 for arterioles and venules, respectively). Only venular velocities showed significant correlation (p = 0.003, R (2) = 0.67) between the two groups.
Our results suggest that BFV measurements by the RFI may be affected by segment length, and care should therefore be taken in choosing vessel segment lengths used during the analysis of RFI data. Long segments of 100-200 pixels (400-800 mu m) seem to provide more robust measurements, which can be explained by the analysis methodology of the RFI device.

• 出版日期2016-6