Automatic Live and Dead Cell Classification via Hyperspectral Imaging

Abstract

Classification of live and dead human ovarian cancer cells (SKOV3) is an important task in biomedicine research. Fluorescence is the most common technique to distinguish live and dead cells. However, it requires physical contact to the cells, which affects the appearance of the cells and their features to a certain degree. There is an urgent need to develop a novel non-invasive technology for this application. This paper proposes the first hyperspectral image (HSI) based system to address this issue. A microscopic hyperspectral imaging system is built to capture cell images. Then morphological methods are employed to extract spectral-spatial features for an SVM classifier that is capable of automatically distinguishing live and dead SKOV3 cells. We show that tensor morphology profile (TMP) leads to the best discriminative capability when compared to other features. At the same time, our work also shows that HSI works better than RGB images for this novel application thanks to its fine spectral information.