Nuclear pattern recognition by two-parameter texture analysis

Abstract

The present paper describes a simple procedure for the analysis of chromatin texture. High-resolution digitized images of nuclei are first standardized to render gray values invariant to staining and illumination conditions. Subsequently, the nucleus is subdivided by a square grid into 0.4 × 0.4 μm 2 quadrats and standard deviations of gray values within each quadrat are estimated. Finally, the overal mean and standard deviation of quadrat standard deviations are calculated. These values may be considered as pure descriptors of the nuclear texture, as they represent the distribution of chromatin changes, disregarding any absolute densitometric and morphometric feature. Using the above descriptors it is possible to recognize at least seven chromatin patterns in a mixed population of developing and degenerating neurons. Results are visually verified by mapping the original pictures at the corresponding bivariate plot points. Comparison with the Markovian texture analysis is discussed.