Interactive live-wire boundary extraction

Area of Science:

• Medical image analysis
• Computational imaging
• Image segmentation

Background:

• Accurate and reproducible boundary extraction is crucial for quantitative image analysis.
• Manual tracing is time-consuming, subjective, and prone to inter-observer variability.
• Existing automated segmentation methods often require extensive parameter tuning or user intervention.

Purpose of the Study:

• To introduce and evaluate the live-wire segmentation technique for interactive boundary extraction.
• To assess the efficiency, accuracy, and reproducibility of live-wire compared to manual tracing.
• To present novel enhancements: boundary cooling and on-the-fly training.

Main Methods:

• Live-wire segmentation utilizes an interactive approach where optimal boundaries are computed dynamically as the user moves the mouse from a seed point.
• The 'live-wire' boundary snaps to object edges, requiring minimal user guidance.
• Boundary cooling automates seed point generation, and on-the-fly training adapts the algorithm to specific image features.

Main Results:

• Live-wire segmentation achieved boundary extraction in one-fifth the time of manual tracing.
• Accuracy was 4.4 times greater, and reproducibility was 4.8 times greater than manual methods.
• Interobserver reproducibility with live-wire was 3.8 times higher than intraobserver reproducibility with manual tracing.

Conclusions:

• The live-wire segmentation tool offers a highly efficient, accurate, and reproducible method for boundary extraction.
• It significantly reduces user input and variability in image segmentation tasks.
• The enhancements of boundary cooling and on-the-fly training further optimize its performance and applicability.