Clustering consistency in neuroimaging data analysis

Clustering techniques have been applied to neuroscience data analysis for decades. New algorithms keep being developed and applied to address different problems. However, when it comes to the applications of clustering, it is often hard to select the appropriate algorithm and evaluate the quality of clustering results due to the unknown ground truth. It is also the case that conclusions might be biased based on only one specific algorithm because each algorithm has its own assumption of the structure of the data, which might not be the same as the real data. In this paper, we explore the benefits of integrating the clustering results from multiple clustering algorithms by a tunable consensus clustering strategy and demonstrate the importance and necessity of consistency in neuroimaging data analysis.