Abstract
Cellular imaging is an exciting area of research in computational life sciences, which provides an essential tool for the study of diseases at the cellular level. In particular, to faciliate the usefulness of cellular imaging for high content screening, image analysis and classification need to be automated. In fact the task of image classification is an important component for any computerized imaging system which aims to automate the screening of high-content, high-throughput fluorescent images of mitotic cells. It can help biomedical and biological researchers to speed up the analysis of mitotic data at dynamic ranges for various applications including the study of the complexity of cell processes, and the screening of novel anti-mitotic drugs as potential cancer therapeutic agents. We propose in this paper a novel image feature based on a spatial linear predictive model. This type of image feature can be effectively used for vector-quantization based classification of nuclear phases. We used a dataset of HeLa cells line to evaluate and compare the proposed method on the classification of nuclear phases. Experimental results obtained from the new feature are found to be superior to some recently published results using the same dataset.