Classifying structural alterations of the cytoskeleton by spectrum enhancement and descriptor fusion

A classifier capable of ranking structural alterations of the cytoskeleton is developed. Images of cytoskeletal microtubules obtained from the epifluorescence microscopy of primary culture rat hepatocytes are analyzed. Morphological descriptors are extracted by contour and mass fractal analysis, direct methods, and spectrum enhancement. All methods are designed and tuned to make the extracted morphological descriptors insensitive to absolute fluorescence intensities. Spectrum enhancement is a nonlinear filter that involves spatial differentiation of the gray-scale image followed by conversion of power spectral density to the logarithmic scale and averaging over arcs in the reciprocal domain. Enhanced spectra exhibit local maxima that correspond to the structured microtubule bundles of a normal cytoskeleton. Descriptor fusion for classification is achieved by means of multivariate analysis. The classifier is trained by image sets representing normal ("negative control") microtubules and those altered by exposure to a fungicide at the highest dose of the experiment design. Some sensitivity and validation tests, including discriminant functions analysis, are applied to the classifier. The latter is applied to recognize images of microtubules not used in the training stage and comes from treatments at lower concentrations and shorter times. As a result, structural alterations are ranked and structural recovery after treatment is quantified. The method has potential use in quantitative, morphology-based tests on the cytoskeleton treated either by anticancer drugs or by cytotoxic agents.