In this study, we will use autoencoders for the development of a cell-type-independent, highly versatile and accurate cell segmentation algorithm, which has not yet been realized with existing imaging analysis/machine learning algorithms. Whereas deep learning can automatically detect features of images, which is an attractive advantage, it requires the preparation of voluminous data—on the order of hundreds or thousands—to reach the correct answer. Thus, in terms of segmentation, a tremendous amount of time is required for building data. This is why highly accurate segmentation algorithms have not been developed despite their being the base technology in cell image analysis. In this research, we will focus on solving the above-mentioned problems and develop a highly accurate segmentation algorithm for various cell types from ‘correct answer’ data obtained from one cell type.