A novel approach to clustering time series data in large spatial arrays.

Many biological and medical imaging modalities produce large arrays of spatially-correlated time series. For example, these include EEG, fMRI, and calcium imaging. A common problem in the analysis of such problems is to organise the image nodes/pixels/voxels into groups that share similarities and that are different from other groups. In machine learning and statistics, this problem is known as clustering. A common approach to clustering is via the use of mixture models, which produce model-based clusterings of the data. Model-based clustering techniques can be slow due to memory-intensive and computationally-expensive algorithms. We propose an approach based on trimmed algorithms, which can be implemented with less memory and computational strain. An example of a zebrafish brain calcium image is presented to demonstrate the approach.

Dr Hien D. Nguyen is a Postdoctoral Research Fellow at the Centre for Advanced Imaging and School of Mathematics and Physics, UQ.