High-Throughput Segmentation of Tiled Biological Structures using Random-Walk Distance Transforms.

Area of Science:

• Biomedical imaging
• Computational biology
• Materials science

Background:

• 3D imaging generates grayscale image stacks requiring segmentation for quantitative analysis.
• Euclidean distance transform (EDT) is common for segmenting touching objects but is noise-sensitive and limited to roundish shapes.
• Biological data often contains noise, hindering accurate segmentation with traditional EDT methods.

Purpose of the Study:

• To introduce and validate the random-walk distance transform (RWT) as an alternative to EDT for segmenting biological structures.
• To demonstrate RWT's effectiveness in high-throughput segmentation of microCT datasets of biological tilings.
• To showcase RWT's applicability to complex, anisotropic structures often found in biological materials.

Main Methods:

• Development and application of the random-walk distance transform algorithm.
• Segmentation of three microCT datasets: stingray cartilage, starfish endoskeleton, and bivalve mollusc shell.
• Utilizing core image processing concepts for segmentation, visualization, and analysis.

Main Results:

• The random-walk transform proved less susceptible to noise compared to EDT, leading to more accurate segmentations.
• RWT effectively segmented biological tilings with anisotropic shapes, overcoming EDT's limitations.
• High-throughput analysis enabled rapid segmentation, visualization, and data extraction from complex biological structures.

Conclusions:

• The random-walk distance transform is a robust and versatile tool for segmenting challenging biological structures.
• This method offers significant advantages over traditional Euclidean distance transform for noisy and complex datasets.
• The study highlights the potential for rapid, large-scale quantitative analysis of biological microstructures.