Limits...
Analysis of Fundus Fluorescein Angiogram Based on the Hessian Matrix of Directional Curvelet Sub-bands and Distance Regularized Level Set Evolution.

Soltanipour A, Sadri S, Rabbani H, Akhlaghi MR - J Med Signals Sens (2015 Jul-Sep)

Bottom Line: In order to extract blood vessel centerlines, the algorithm of vessel extraction starts with the analysis of directional images resulting from sub-bands of fast discrete curvelet transform (FDCT) in the similar directions and different scales.The final vessel segmentation is obtained using a simple region growing algorithm iteratively, which merges centerline images with the contents of images resulting from modified top-hat transform followed by bit plane slicing.The experimental results show the accuracy more than 93% for vessel segmentation and more than 87% for OD boundary extraction.

View Article: PubMed Central - PubMed

Affiliation: Department of Electrical and Computer Engineering, Isfahan University of Technology, Isfahan, Iran.

ABSTRACT
This paper presents a new procedure for automatic extraction of the blood vessels and optic disk (OD) in fundus fluorescein angiogram (FFA). In order to extract blood vessel centerlines, the algorithm of vessel extraction starts with the analysis of directional images resulting from sub-bands of fast discrete curvelet transform (FDCT) in the similar directions and different scales. For this purpose, each directional image is processed by using information of the first order derivative and eigenvalues obtained from the Hessian matrix. The final vessel segmentation is obtained using a simple region growing algorithm iteratively, which merges centerline images with the contents of images resulting from modified top-hat transform followed by bit plane slicing. After extracting blood vessels from FFA image, candidates regions for OD are enhanced by removing blood vessels from the FFA image, using multi-structure elements morphology, and modification of FDCT coefficients. Then, canny edge detector and Hough transform are applied to the reconstructed image to extract the boundary of candidate regions. At the next step, the information of the main arc of the retinal vessels surrounding the OD region is used to extract the actual location of the OD. Finally, the OD boundary is detected by applying distance regularized level set evolution. The proposed method was tested on the FFA images from angiography unit of Isfahan Feiz Hospital, containing 70 FFA images from different diabetic retinopathy stages. The experimental results show the accuracy more than 93% for vessel segmentation and more than 87% for OD boundary extraction.

No MeSH data available.


Related in: MedlinePlus

(a) Detected vessel centerlines. (b) Final vessel segmentation
© Copyright Policy - open-access
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4528352&req=5

Figure 12: (a) Detected vessel centerlines. (b) Final vessel segmentation

Mentions: According to Figure 11, in the first iteration, the vessel centerlines image is used as seed points and the reconstructed image from top-hat transform with the smallest structuring element for opening operator is used as an aggregated image for a simple region growing algorithm. In each of the subsequence three iterations, according to Figure 11, the reconstructed images from top-hat transform with increasing size of structuring element are used as seed points and outputs of the previous region growing steps are used as aggregated image. The final vascular tree segmentation is obtained after implementation of the region growing algorithm in the fourth iteration. Figure 12 shows vessel centerlines and final vessel segmentation using the proposed method.


Analysis of Fundus Fluorescein Angiogram Based on the Hessian Matrix of Directional Curvelet Sub-bands and Distance Regularized Level Set Evolution.

Soltanipour A, Sadri S, Rabbani H, Akhlaghi MR - J Med Signals Sens (2015 Jul-Sep)

(a) Detected vessel centerlines. (b) Final vessel segmentation
© Copyright Policy - open-access
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4528352&req=5

Figure 12: (a) Detected vessel centerlines. (b) Final vessel segmentation
Mentions: According to Figure 11, in the first iteration, the vessel centerlines image is used as seed points and the reconstructed image from top-hat transform with the smallest structuring element for opening operator is used as an aggregated image for a simple region growing algorithm. In each of the subsequence three iterations, according to Figure 11, the reconstructed images from top-hat transform with increasing size of structuring element are used as seed points and outputs of the previous region growing steps are used as aggregated image. The final vascular tree segmentation is obtained after implementation of the region growing algorithm in the fourth iteration. Figure 12 shows vessel centerlines and final vessel segmentation using the proposed method.

Bottom Line: In order to extract blood vessel centerlines, the algorithm of vessel extraction starts with the analysis of directional images resulting from sub-bands of fast discrete curvelet transform (FDCT) in the similar directions and different scales.The final vessel segmentation is obtained using a simple region growing algorithm iteratively, which merges centerline images with the contents of images resulting from modified top-hat transform followed by bit plane slicing.The experimental results show the accuracy more than 93% for vessel segmentation and more than 87% for OD boundary extraction.

View Article: PubMed Central - PubMed

Affiliation: Department of Electrical and Computer Engineering, Isfahan University of Technology, Isfahan, Iran.

ABSTRACT
This paper presents a new procedure for automatic extraction of the blood vessels and optic disk (OD) in fundus fluorescein angiogram (FFA). In order to extract blood vessel centerlines, the algorithm of vessel extraction starts with the analysis of directional images resulting from sub-bands of fast discrete curvelet transform (FDCT) in the similar directions and different scales. For this purpose, each directional image is processed by using information of the first order derivative and eigenvalues obtained from the Hessian matrix. The final vessel segmentation is obtained using a simple region growing algorithm iteratively, which merges centerline images with the contents of images resulting from modified top-hat transform followed by bit plane slicing. After extracting blood vessels from FFA image, candidates regions for OD are enhanced by removing blood vessels from the FFA image, using multi-structure elements morphology, and modification of FDCT coefficients. Then, canny edge detector and Hough transform are applied to the reconstructed image to extract the boundary of candidate regions. At the next step, the information of the main arc of the retinal vessels surrounding the OD region is used to extract the actual location of the OD. Finally, the OD boundary is detected by applying distance regularized level set evolution. The proposed method was tested on the FFA images from angiography unit of Isfahan Feiz Hospital, containing 70 FFA images from different diabetic retinopathy stages. The experimental results show the accuracy more than 93% for vessel segmentation and more than 87% for OD boundary extraction.

No MeSH data available.


Related in: MedlinePlus