In Vivo Tumor Spatial Classification using PCA and K-Means with NIR-Hyperspectral Data

Abstract

This paper presents new method of spatial classification and wavelength bands reduction of near-inferred (NIR) hyperspectral imaging data for medical application. Hyperspectral imaging data have more than several hundred wavelength bands. However hyperspectral data have sometimes redundant information to detect region of interest. The aim of this research is to archive became possible that a region of interest is distinguished by observing the particular wavelength bands without any markers in order to develop a special application such as a surgery supporting system. NIR light with wavelengths of 800-2000 nm, called as the ‘biological window,’ has received particular attention given that water and biological tissues have minimal optical loss caused by scattering and absorption at these wavelengths. NIR light can penetrate/see through deep tissues. NIR endoscope have a great potential as the surgery supporting system, however wavelength bands needs to reduce according to the limitation of NIR endoscope hardware performance. To consider only several wavelength bands are sometimes much effective case than to consider all wavelength bands. In this paper, we proposed the method of spatial classification and reduction a number of wavelength bands simultaneously by combined PCA and k-means, and assessed the cancer-caring nude mouse. The experimental results demonstrate that the proposed method can select valuable wavelength bands to distinguish the region of interest with comparable accuracy of the conventional method.