A convolutional neural network model for semantic segmentation of mitotic events in microscopy images

Abstract

Mitosis, which has important effects such as healing and growing for human body, has attracted considerable attention in recent years. Especially, cell division characteristics contain useful information for regenerative medicine. However, the analysis of this complex structure is very challenging process for experts, because many cells are scattered at random times and at different speeds. Therefore, we propose an automatic mitosis event detection method using convolutional neural network (CNN). In the proposed method, semantic segmentation has been applied with the help of CNN in order to make the complex mitosis images more easily understandable. The CNN structure consists of four convolution layers, four pooling layers, one rectified linear unit layer and softmax layer. Generally, the aim of CNN structure is to reduce the image size, but in this study, the image size is preserved for the semantic segmentation which provides high-level information. For this, the size of the images at each layer output is calculated and updated with the appropriate padding parameters. Thus, real-size images presented at the network output can be easily understood. BAEC and C2C12 phase-contrast microscopy image sequences are used for experiments. The precision, recall and F-score parameters are used for evaluating the success of the proposed method and compared with the other methods using the same datasets.