A pulse coupled neural network segmentation algorithm for reflectance confocal images of epithelial tissue

Area of Science:

• Biomedical Imaging
• Computational Pathology
• Cancer Diagnostics

Background:

• Reflectance confocal microscopy offers in vivo, sub-cellular resolution 3D imaging of epithelial tissue.
• Nuclear-to-cytoplasmic ratio is a critical biomarker for epithelial precancer detection and staging.
• Challenges include low contrast, reduced resolution at depth, and signal variations, hindering accurate nuclear segmentation.

Purpose of the Study:

• To develop an automated method for segmenting nuclei in reflectance confocal microscopy images.
• To improve the quantification of nuclear-to-cytoplasmic ratio for epithelial precancer assessment.

Main Methods:

• Utilized a pulse coupled neural network (spiking cortical model) and an artificial neural network classifier for automated segmentation.
• Applied the algorithm to simulated nuclei with varying contrast levels.
• Validated the method on in vivo confocal images of porcine and human oral mucosa.

Main Results:

• The automated segmentation method achieved high accuracy in simulated data.
• Over 90% of simulated nuclei were detected with a contrast ratio of 2.0 or greater.
• Segmentation accuracy was validated against manual segmentation on real tissue samples.

Conclusions:

• The developed automated segmentation method is effective for analyzing reflectance confocal microscopy images.
• This technique facilitates accurate nuclear quantification, crucial for early epithelial cancer detection.
• The method shows promise for clinical applications in diagnosing and staging epithelial cancers.