Breakthroughs in bioimaging technology have paved the way for the acquisition of large amounts of information from biological images. In addition to 3D spatial information, biological images contain temporal and spectral information along multiple axes. Identifying ‘core’ information as well as the relationships between various axes by representing such an enormous quantity of multi-dimensional data in a unified manner is a principal challenge in the area of biological image information processing. In this study, we will develop a new method of multilinear sparse and low-rank tensor decomposition and apply it to the phenotype analysis of 4D biological data. By decomposing biological dynamic images (4th-order tensor) into a sparse tensor and a low-rank tensor, we aim to (a) identify and detect the place and time of a specific change (e.g., cell division), and (b) enable robust statistical analysis and machine learning even if there is noise in the data or when pieces of data are missing.