A multi-scale tensor voting approach for small retinal vessel segmentation in high resolution fundus images

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Background:

• Accurate segmentation of retinal vessels is vital for computer-aided detection (CAD) of diabetic retinopathy.
• Current methods often struggle to segment the smallest retinal vessels, potentially leading to false positives in lesion detection.
• Detecting small vessels is essential for comprehensive analysis and early diagnosis.

Purpose of the Study:

• To propose a new hybrid method for the accurate segmentation of the smallest retinal vessels.
• To improve the sensitivity and accuracy of retinal vessel segmentation, particularly for fine vasculature.
• To enhance the performance of CAD systems for diabetic retinopathy by addressing limitations in small vessel detection.

Main Methods:

• A hybrid approach combining multi-scale line detection and perceptual organization techniques.
• Reconstruction of small vessels using a perceptual-based approach with tracking and pixel painting.
• Validation on a high-resolution fundus image database of healthy and diabetic subjects.

Main Results:

• The proposed hybrid method achieved a sensitivity rate of 85.06%, outperforming the original multi-scale line detection method (81.06%).
• A significant improvement of 6.47% in sensitivity was observed for the smallest vessels.
• The method demonstrated a 7.8% improvement in vasculature detection based on perceptual measures.

Conclusions:

• The novel hybrid method effectively segments small retinal vessels, enhancing diagnostic accuracy for diabetic retinopathy.
• This approach offers a significant improvement over existing methods, particularly in detecting fine vasculature.
• Accurate segmentation of small vessels is critical for reliable CAD systems in ophthalmology.