A MULTIDIMENSIONAL DATA VISUALIZATION AND CLUSTERING METHOD: CONSENSUS SIMILARITY MAPPING

摘要

Multiparametric Magnetic Resonance Imaging (MRI) produces large amounts of high dimensional data for radiologists to read. Currently, radiologists integrate multiparametric MRI data visually to identify meaningful structures within the data. In this paper, we present a novel visualization and clustering technique called "Consensus Similarity Mapping (CSM)" for integration of multidimensional radiological data. The CSM algorithm computes an ensemble of stable clustering results obtained from multiple runs of the k-means algorithm. The CSM algorithm uses a unique method called cluster stability index (CSI) to identify the stable clustering configurations required to create the k-means ensemble. The CSM algorithm transforms the stable clustering ensemble into a matrix of pairwise similarities, which, uncovers the intrinsic classes within the high dimensional input data. We demonstrate the performance of CSM on well-known synthetic datasets as well as multiparametric magnetic resonance imaging (MRI) data.