Cell Image Segmentation by Integrating Multiple CNNs

Yuki Hiramatsu, Kazuhiro Hotta, Ayako Imanishi, Michiyuki Matsuda, Kenta Terai; Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2018, pp. 2205-2211

Convolutional Neural Network is valid for segmentation of objects in an image. In recent years, it is beginning to be applied to the field of medicine and cell biology. In semantic segmentation, the accuracy has been improved by using deeper single deep neural network. However, the accuracy is saturated for difficult segmentation tasks. In this paper, we propose a semantic segmentation method by integrating multiple CNNs adaptively. This method consists of a gating network and multiple expert networks. Expert network outputs a segmentation result for an input image. Gating network automatically divides the input image into several sub-problems and assigns them to expert networks. Thus, each expert network solves only the specific problem, and our proposed method is possible to learn more efficiently than single deep neural network. We evaluate the proposed method on segmentation problem of cell membrane and nucleus. The proposed method improved the segmentation accuracy in comparison with single deep neural network.