Refining competition in the self-organising tree map for unsupervised biofilm image segmentation

Area of Science:

• Microbiology
• Computer Science
• Image Analysis

Background:

• Accurate segmentation of micro-organisms from microscopy images is crucial for biological research.
• Existing methods may struggle with complex cellular structures and overlapping objects.
• The Self Organising Tree Map (SOTM) offers a novel approach to unsupervised feature learning and data clustering.

Purpose of the Study:

• To investigate the efficacy of the Self Organising Tree Map (SOTM) neural network for segmenting micro-organisms in confocal microscopy data.
• To explore the impact of various image features (intensity, phase congruency, spatial proximity) on segmentation performance.
• To propose refinements to the SOTM for improved microbial image segmentation.

Main Methods:

• Utilized the Self Organising Tree Map (SOTM) neural network for image segmentation.
• Analyzed pixel and regional intensity, phase congruency, and spatial proximity features.
• Developed a refined competitive search strategy and a data-dependent stop criterion for the SOTM.
• Conducted preliminary experiments to evaluate segmentation improvements.

Main Results:

• Demonstrated that the SOTM can effectively segment micro-organisms from confocal microscopy images.
• Identified that prioritizing intensity features in early learning stages, followed by relaxing proximity constraints, enhances segmentation.
• The proposed refinements to the SOTM improved sensitivity to regional associations of microbial matter.

Conclusions:

• The SOTM is a flexible and efficient tool for micro-organism segmentation, preserving data topology.
• Feature weighting during the SOTM learning process significantly impacts segmentation quality.
• The proposed SOTM refinements offer a generalizable mechanism for improved microbial image analysis.