Spherical harmonics texture extraction for versatile analysis of biological objects

Area of Science:

• Bioimage analysis
• Computational biology
• Microscopy data analysis

Background:

• Phenotype characterization from microscopy relies on image intensity distribution.
• Existing texture extraction methods often fail to adapt to 3D microscopy data.
• Novel approaches are needed for quantitative analysis of 3D image texture.

Purpose of the Study:

• To introduce Spherical Texture extraction for analyzing 3D microscopy data.
• To provide a quantitative method for texture analysis adaptable to various biological systems.
• To offer a user-friendly implementation for broad accessibility.

Main Methods:

• Spherical Texture extraction measures intensity variance per angular wavelength using Spherical Harmonics or Fourier power spectrum.
• Calculates a 20-value feature set characterizing intensity distribution scale.
• Applies the method to 2D and 3D microscopy datasets, including a plugin for ilastik and a Python package.

Main Results:

• Successfully characterized gene expression patterns in Drosophila melanogaster embryos.
• Quantified morphological differences in Caenorhabditis elegans germline nuclei.
• Demonstrated superior classification performance over convolutional neural networks with limited training data for nuclei classification.
• Extracted polarization direction and marker alignment in 2D cell migration data.

Conclusions:

• Spherical Texture extraction is a versatile and effective method for quantitative feature extraction from microscopy data.
• The method provides robust characterization of biological patterns and morphology.
• Its performance, especially with limited data, highlights its utility in various bioimage analysis applications.