Pinky A. BAUTISTA, Tokiya ABE, Masahiro YAMAGUCHI, Yukako YAGI¹ and Nagaaki OHYAMA²

Tokyo Institute of Technology, Imaging Science & Engineering Laboratory, 4259 Nagatsuta-cho, Midori-ku, Yokohama 226-8503, Japan
¹University of Pittsburgh Medical Center, Health System, Department of Pathology, Division of Pathology Informatics, Pittsburgh Pennsylvania, USA
²Tokyo Institute of Technology, Frontier Collaborative Research Center, 4259 Nagatsuta-cho, Midori-ku, Yokohama 226-8503, Japan

(Received June 1, 2004; Accepted November 22, 2004; Revised October 15, 2004)

Histological structures of a pathological tissue sample convey information relevant to the diagnosis of the disease that might have afflicted the person. To reveal the morphology of these structures clearly, pathological tissues are stained. In this paper, a digital staining methodology for pathological tissue samples is introduced. Digital staining implies the application of digital processing techniques to transform the image of an unstained sample to its stained image counterpart. In the method, the transmittance spectra of the unstained and Hematoxylin and Eosin (H&E) stained multispectral images (16 bands) of specific tissue components are utilized. Two experiments were conducted to probe the possibility of the digital staining framework: the linear mapping of spectral transmittances, and the classification of spectral transmittances in conjunction with the linear mapping of specific transmittance data sets. The method classified the four tissue components, e.g. nucleus, cytoplasm, red blood cells, and the white region (region devoid of tissue structures), while the misclassifications between components with spectral transmittances that are closely similar were not completely rectified.

Key words: Multispectral, digital staining, spectral transmittance, linear mapping, classification