Convolutional Neural Network-Based Automatic Classification for Algal Morphogenesis

抄録

Convolutional neural networks (CNNs) are used for various data analyses and resemble the human brain cognition system. CNN algorithms are composed of multiple layers of convolution and pooling layers. Recent studies have shown that applying CNN algorithms to classify biological images are feasible. The alga, which has various morphological features, is the one of the interesting targets to classify using CNN. As an example, we targeted unicellular Cyanidioschyzon merolae. C. merolae is a primitive red alga that is not only used for studying organelles but also to proceed the study of the production of lipids. Measuring the division rate requires the classification of the interphase and mitotic phase of C. merolae. In this study, we constructed an automatic classifier for interphase and mitotic C. merolae. By tuning the hyper-parameter, number of layers, and number of insertions of dropout functions, we classified the interphase and mitotic C. merolae with the high accuracy of 92%.