A Benchmark Framework for Multiregion Analysis of Vesselness Filters

Area of Science:

• Medical image analysis
• Computer vision
• Vascular imaging

Background:

• Vesselness filters enhance tubular structures in angiographic images for segmentation.
• Selecting appropriate filters and parameters is complex and requires expertise.
• Existing benchmarks for comparing vesselness filters are limited.

Purpose of the Study:

• To present a generic framework for comparing vesselness filters.
• To benchmark seven gold-standard vesselness filters using the proposed framework.
• To provide an open-source tool for reproducible research and community use.

Main Methods:

• Developed a generic framework for vesselness filter comparison.
• Experimented on three public datasets: hepatic (CT), brain (MRA), and synthetic.
• Quantitatively and qualitatively analyzed seven filters, focusing on performance across different vessel sizes and bifurcations.

Main Results:

• Compared seven vesselness filters on diverse datasets.
• Assessed filter performance on organ, vascular network, and vessel-specific neighborhoods.
• Evaluated filter efficacy on vessel bifurcations, often missed in segmentation.

Conclusions:

• The benchmark framework facilitates objective comparison of vesselness filters.
• Results provide insights into filter performance for specific vascular structures and datasets.
• The provided code and demonstrator support the selection and application of vesselness filters.