An automatic three-way clustering method based on shadowed sets and its application to FMT

Abstract

Fluorescence molecular tomography (FMT) is a promising medical imaging technique that enables non-invasive, specific, and sensitive detection of the fluorophore. However, multiple-target resolving remains a challenge in FMT. Its key issue is how to determine the number of targets, which is difficult to obtain in practical applications. To address this issue, a three-way clustering algorithm based on shadowed sets is proposed in this work. It not only provides the appropriate number of clusters automatically but also determines the partition thresholds. First, we have defined a shadowed set similarity to quantify the similarity between fuzzy information samples. Second, based on the concept of shadowed set similarity, the intra-class similarity and inter-class similarity are introduced to quantify the change of the relationship between the sample and the clusters. Furthermore, an improved clustering validity index (CVI) is created to measure the clustering performance. At the same time, the optimal cluster number is determined according to the CVI. Finally, the proposed method has been applied to FMT for adaptively determining the number of fluorescent targets. The experimental data of FMT indicate that the proposed algorithm can accurately determine the number of fluorescent targets from the effective reconstruction results. University of California Irvine (UCI) datasets further demonstrated the potential of the method.