Fuzzy clustering in biochemical analysis of cancer cells

Currently there is considerable effort investigating whether infrared spectroscopy can be used as a diagnostic probe to identify early stages of cancer since these techniques are sensitive to biological changes within cells. Cluster analysis is often used to try and unravel complex vibrational images. In this paper, a Fuzzy C-Means (FCM) based model selection algorithm was used to automatically cluster sets of infrared spectral data taken from lymph node tissue sections. Initial results were often prone to the creation of excessive clusters in comparison with clinical diagnosis which is partly due to the complexities of biological tissue. A new method to merge clusters was developed with the ability to successfully find and combine the two most similar clusters in order to address this problem. This new method was applied to four sets of infrared spectral data, from two types of human cancers (lymph node and oral). The experimental results clearly show that this new method can successfully combine the most similar clusters together and potentially improve the accuracy of diagnosis.