Bioinformatics
Performance Model Selection for Learning-based Biological Image Analysis on a Cluster
Proceedings of the International Conference on Bioinformatics, Computational Biology and Biomedical Informatics
| Hi-index | 0.00 |
Cell counting is an important problem in many microscopy related applications of cellular functional analysis or disease diagnosis. While automatic 2D image processing and pattern recognition approaches for cell counting have been studied, their 3D counterparts are less visited, especially when the cells are crowded. Here we propose a discriminative model to count cells automatically and apply it to for 3D confocal images of the Drosophila melanogaster (fruit fly)'s larval nervous system. We design and implement the process of cell counting based on the detection of nuclear centers; we formulate a binary classification task that incorporates both 3D morphological and texture features. Compared with unsupervised cell counting based on template matching, our model produces more reliable results. This supervised model also requires a smaller number of parameters than the unsupervised approach, and at the mean time improves the robustness. Our cell counting method is also computationally efficient, which is important for high-throughput quantification of 3D cellular images.