Dual-pass feature extraction on human vessel images

Area of Science:

• Medical Imaging
• Computational Anatomy
• Image Processing

Background:

• Vessel imaging analysis is crucial for diagnosing cardiovascular conditions.
• Artifacts from fat deposits and non-elliptical vessel shapes complicate automated feature extraction.
• Accurate segmentation of the vessel lumen is essential for quantitative analysis.

Purpose of the Study:

• To develop and validate a novel, automated algorithm for extracting human vessel cavity features.
• To address challenges posed by image artifacts and irregular vessel geometries.
• To provide an accurate and efficient tool for vessel lumen segmentation.

Main Methods:

• A two-stage approach: 1) Bounding segmentation mask using circular region filling and Principal Component Analysis (PCA) with auto-correction.
• 2) Precise cavity enclosure via micro-dilation filter and edge-walking scheme.
• Algorithm validated on 30 computed tomography angiography (CTA) scans of lower body vessels with varying wall distortions.

Main Results:

• High accuracy achieved: 98% sensitivity, 8% false positive rate, 93% positive predictive value compared to specialist annotations.
• Exceptional efficiency: Average execution time of 18-24 ms per 15 cm section (approx. 1 ms per contour).
• Excellent reproducibility demonstrated on synthetic images with no variation in computed diameters.

Conclusions:

• The novel algorithm reliably extracts human vessel cavity features, even with challenging artifacts and shapes.
• Its high accuracy and efficiency make it suitable for interactive medical imaging software.
• The method provides a robust solution for quantitative vessel lumen analysis in clinical settings.