Application of time series techniques to data mining and analysis of spatial patterns in 3D images

Abstract

Analysis of spatial patterns in images can provide valuable information in many application domains, such as in geography, meteorology and medicine. We propose to apply techniques from the time series domain to analyze the spatial patterns extracted from 3D images. After traversing an image using a space-filling curve, we discover discriminative patterns by analyzing the spatial sequence in the transformed domain. Because of the similarity of the sequences with time series, we propose the use of existing time series similarity analysis techniques, including Euclidean distance, and dimensionality reduction techniques, such as singular value decomposition and piecewise aggregate approximation, for further analysis of the spatial patterns. As a case study, we analyze an fMRI dataset. Experimental results verify that the discovered spatial patterns have strong discriminative power among different classes and the overall accuracy for clustering and similarity retrieval is above 90% and as high as 100% for certain experimental settings.