Challenges in Computational Histopathology: The Feasibility of FTIR Spectroscopy in Clustering

The aim of this project is to find suitable, reliable and fast methods for computational cancer histopathology. Clustering of FTIR spectra has been done before, but this paper uses features instead of whole spectra or its second derivative. The resolution of FTIR spectroscopic image is not good enough, because pathologists need better resolution and different magnifications for current histopathologic recognition. When there is no way to compare the result of FTIR and light microscopic image at high resolution, it is also impossible to compare the results of clustering using FTIR spectroscopy. A computational histopathological method using both FTIR spectroscopy and light microscopic images is proposed in this study. Only important parts of tissue samples are measured using FTIR spectroscopy, which conserves time. Then the measured spectra is compared to preconceived spectral database. Spectral database is formed by measuring perfectly cultivated samples of tissue types using FTIR spectroscopy.