Classify cellular phenotype in high-throughput fluorescence microcopy images for RNAi genome-wide screening

As we know, the genes could cause the cell phenotypes to change dramatically. Currently, biologists attempt to perform the genome-wide RNAi screening to identify various image phenotypes. It is a challenging task to recognize the phenotypes automatically because of the noisy background and low contrast of fluorescence images. In this work, we applied two cellular segmentation techniques, deformable model and Cellprofiler software, for the preprocess of cellular segmentation. Then five kinds of features including wavelet feature, moments feature, haralick co-occurrence feature, region property feature, and problem-specific shape descriptor are extracted from the cellular patches. The genetic algorithm (GA) is applied to select a subset of the most discriminate features to remove the irrelevance and redundancy. We use linear discriminant analysis (LDA) as the tool for training the statistical classification model. Experimental results show the proposed approach works well in RNAi screening