Automatic annotation of drosophila developmental stages using association classification and information integration

In current developmental research, one of the challenging tasks is to understand the spatio-temporal gene expression patterns and the relationships among different genes. In situ hybridization (ISH) assay which shows mRNA spatio-temporal expression patterns in cells and tissues directly is currently widely utilized in the bench work. With the increasing of available ISH images, automatic annotation systems are highly demanded. In this paper, an automatic classification system is proposed for annotating the in situ hybridization images with respect to the developmental stages. The embryo is first segmented from the original image, registered and normalized. The segmented embryo image is then divided into 100 blocks from which the pixel intensity and texture features are extracted and discretized. The multiple correspondence analysis (MCA) based association classification approach is proposed to generate classification rules for different stages based on the training data set. The testing instance is classified by applying the rules generated in the training process and a classification coordination module is incorporated to resolve the conflicts utilizing the weights derived from angle values in the MCA procedure. Experimental results show that our proposed method achieves promising results and outperforms other state-of-the-art algorithms.