A comparison between two robust techniques for segmentation of blood vessels

Area of Science:

• Medical image analysis
• Computer vision
• Biomedical imaging

Background:

• Image segmentation is crucial for image analysis, but no single method suits all applications.
• The goal of segmentation is observer-dependent, necessitating method comparison.
• Accurate blood vessel (BV) quantification is vital in medical imaging.

Purpose of the Study:

• To compare the efficacy of the mean shift segmentation method with a novel algorithm against the spectral method for blood vessel detection.
• To evaluate segmentation performance against manual segmentation benchmarks.
• To identify the most accurate and efficient image segmentation technique for BV analysis.

Main Methods:

• Implementation of a novel algorithm for the mean shift segmentation method.
• Application of the spectral segmentation method.
• Comparative analysis of both methods against manual segmentation for blood vessel counting.
• Quantification of segmentation accuracy using false positive (FP) and false negative (FN) rates.

Main Results:

• Mean shift segmentation achieved an error rate of less than 20% for false positives (FP) and 0% for false negatives (FN).
• The spectral method exhibited a higher error rate, with over 45% for FP and 0% for FN.
• Manual segmentation served as the ground truth for performance evaluation.

Conclusions:

• The proposed mean shift segmentation method demonstrates superior performance and accuracy for blood vessel detection compared to the spectral method.
• Mean shift segmentation offers a more reliable approach for quantitative BV analysis in medical images.
• The study highlights the importance of method selection in achieving accurate image segmentation for specific analytical goals.