Biological Image Temporal Stage Classification via Multi-layer Model Collaboration

In current biological image analysis, the temporal stage information, such as the developmental stage in the Drosophila development in situ hybridization images, is important for biological knowledge discovery. Such information is usually gained through visual inspection by experts. However, as the high-throughput imaging technology becomes increasingly popular, the demand for labor effort on annotating, labeling, and organizing the images for efficient image retrieval has increased tremendously, making manual data processing infeasible. In this paper, a novel multi-layer classification framework is proposed to discover the temporal information of the biological images automatically. Rather than solving the problem directly, the proposed framework uses the idea of ``divide and conquer'' to create some middle level classes, which are relatively easy to annotate, and to train the proposed subspace-based classifiers on the subsets of data belonging to these categories. Next, the results from these classifiers are integrated to improve the final classification performance. In order to appropriately integrate the outputs from different classifiers, a multi-class based closed form quadratic cost function is defined as the optimization target and the parameters are estimated using the gradient descent algorithm. Our proposed framework is tested on three biological image data sets and compared with other state-of-the-art algorithms. The experimental results demonstrate that the proposed middle-level classes and the proper integration of the results from the corresponding classifiers are promising for mining the temporal stage information of the biological images.