Fuzzy information fusion of classification models for high-throughput image screening of cancer cells in time-lapse microscopy

Abstract

Bioimaging at molecular and cellular levels requires specific image analysis methods to help life scientists develop methodologies and hypotheses in biology and biomedicine. In particular, this is true when dealing with microscopic images of cells and vessels. To facilitate the automation of cell screening, we have developed methods based on vector quantization and Markov model for classification of cellular phases using time-lapse fluorescence microscopic image sequences. Because of ambiguity inherently existing in the labeling of cell-phase feature vectors, we proposed to use relaxation labeling technique to reduce uncertainty among cell-phase models having overlapping properties. To further improve the classification rate we applied a fuzzy fusion strategy for combining individual results obtained from multiple classifiers. Our proposed image-classification methods can be useful for the task of high-content cell-cycle screening which is essential for biomedical research in the study of structures and functions of cells and molecules.