Àá½Ã¸¸ ±â´Ù·Á ÁÖ¼¼¿ä. ·ÎµùÁßÀÔ´Ï´Ù.
KMID : 1191420120110010091
Korean Journal of Optometry and Vision Science
2012 Volume.11 No. 1 p.91 ~ p.95
A Statistical Approach on Fast Retinal Vessel Detection
Nam Hyung-Sung

Kang Uk
Seo Jong-Mo
Lee Hae-Young
Kei Shinoda
Khwarg Sang-In
Abstract
Purpose: In fundus photographs, pixel data were statistically analyzed in order to detect retinal vessel rapidly.

Methods: Red channel data was used to detect the optic nerve head (ONH) and make the region of interest (ROI). ROI was established between the circle of radius 2RONH and the circle of radius 3RONH when ONH has radius RONH. After that, intensity profiles along each concentric circle can be obtained in the ROI by using green channel data. Each concentric circle has radius from 2RONH to 3RONH at 5 pixel intervals. Local maximum data were derived from each intensity profile, and data smoothing and cubic spline interpolation were utilized to acquire trend lines. Using these intensity profiles and trend lines, the pixel distribution information was derived by subtracting each intensity profile from the corresponding trend line. Mean and standard deviation (SD) values were calculated for each new intensity profile, and three threshold values (1.96, 1.65 and 1.00 SD from the mean) were used. Segmentation results were analyzed by two retina specialists.

Results: Several venules were detected at the threshold 97.5%, venules and clear arterioles were detected at the the threshold 95%, and most venules & arterioles were detected at the threshold 84%. The diameters of the retinal vessels were corresponded with physician¡¯s interpretation for venules at the threshold 97.5%, and for arterioles at the threshold 84%.

Conclusions: Meaningful result was derived with the fast and simple statistical method which is different from existing complicated vessel tracking algorithms. It would be possible to measure the blood vessel diameter more accurately with the arteriole/venule classification and the compensation of vascular direction.
KEYWORD
Computer-assisted image interpretation, Retinal vessel
FullTexts / Linksout information
Listed journal information