K-means and Hierarchical Cluster Analysis as segmentation algorithms of FTIR hyperspectral images collected from cutaneous tissue

Fourier Transform Infrared (FTIR) spectroscopy is a rapid and label-free analytical technique whose potential as a diagnostic tool has been well demonstrated. The combination of spectroscopy and microscopy technologies enable wide-field scanning of a sample, providing a hyperspectral image with tens of thousands of spectra in a few minutes. In order to increase the information content of FTIR images, different clustering algorithms have been proposed as segmentation methods. However, systematic comparative tests of these techniques are still missing. Thus, the present paper aims to compare the ability of K-means Cluster Analysis (KMCA) and Hierarchical Cluster Analysis (HCA) as clustering algorithms to reconstruct FTIR hyperspectral images. Spectra for cluster analysis were acquired from healthy cutaneous tissue and the pseudo-color reconstructed images were compared to standard histopathology in order to assess the number of clusters required by both methods to correctly identify the morphological skin components (stratum corneum, epithelium, dermis and hypodermis).