Filtering Spatial Point Patterns Using Kernel Densities.

Area of Science:

• Spatial statistics
• Biomedical image analysis
• Computational statistics

Background:

• Spatial inhomogeneity and clustering are crucial in disease monitoring and medical image analysis.
• Existing methods struggle with variations in cluster size and density.
• Classifying points as features or background noise is a common challenge.

Purpose of the Study:

• To develop a novel method for separating feature points from noise in spatial point patterns.
• To address limitations of existing methods when dealing with inhomogeneous cluster characteristics.
• To apply the method to analyze spatial distribution of emphysema in lung CT scans.

Main Methods:

• Employing kernel density estimates of point process intensity.
• Utilizing a data-driven approach for bandwidth selection.
• Constructing a null distribution via asymptotic properties or Monte Carlo simulation for comparison.

Main Results:

• The proposed Kernel Density and Simulation based Filtering (KDS-Filt) method demonstrated superior performance.
• KDS-Filt effectively handles variations in cluster size and density.
• The method proved useful for identifying clinically relevant spatial information in emphysema CT scans.

Conclusions:

• KDS-Filt offers an improved approach for analyzing spatial point patterns, particularly with inhomogeneous clusters.
• The methodology provides a robust tool for biomedical image analysis, exemplified by emphysema detection.
• The KDS-Filt method is accessible through the sncp R package.