Color-Based Tumor Tissue Segmentation for the Automated Estimation of Oral Cancer Parameters YUNG-NIEN SUN, 1 * YI-YING WANG, 1 SHAO-CHIEN CHANG, 1 LI-WHA WU, 2,3 AND SEN-TIEN TSAI 4 1 Department of Computer Science and Information Engineering, National Cheng Kung University, Tainan, Taiwan Province, Republic of China 2 Cardiovascular Research Center, College of Medicine, National Cheng Kung University, Tainan, Taiwan Province, Republic of China 3 Institute of Molecular Medicine, College of Medicine, National Cheng Kung University, Tainan, Taiwan Province, Republic of China 4 Department of Otolaryngology, National Cheng Kung University, Tainan, Taiwan Province, Republic of China KEY WORDS feature extraction; principal component analysis; tumor tissue classification ABSTRACT This article presents an automatic color-based feature extraction system for parameter estimation of oral cancer from optical microscopic images. The system first reduces image-to-image variations by means of color normalization. We then construct a database which consists of typical cancer images. The color parameters extracted from this database are then used in automated online sampling from oral cancer images. Principal component analysis is subse- quently used to divide the color features into four tissue types. Each pixel in the cancer image is then classified into the corresponding tissue types based on the Mahalanobis distance. The afore- mentioned procedures are all fully automated; in particular, the automated sampling step greatly reduces the need for intensive labor in manual sampling and training. Experiments reveal high lev- els of consistency among the results achieved using the manual, semiautomatic, and fully auto- matic methods. Parameter comparisons between the four cancer stages are conducted, and only the mean parameters between early and late cancer stages are statistically different. In summary, the proposed system provides a useful and convenient tool for automatic segmentation and evaluation for stained biopsy samples of oral cancer. This tool can also be modified and applied to other tissue images with similar staining conditions. Microsc. Res. Tech. 73:5–13, 2010. V V C 2009 Wiley-Liss, Inc. INTRODUCTION Oral cancer is one leading cause of cancer death in the world. Betel quid chewing, alcohol drinking, and smoking are three well-recognized risk factors for oral cancer by World Health Organization. Because of the prevalent use of betel quid among young males, the incidence rate of this cancer is increasing at an alarm- ing rate in Taiwan (Tsai et al., 2004). Moreover, the 5- year survival rate for oral cancer still lags behind most of other cancer types despite the therapeutic regimen advancement. Therefore, early detection of oral cancer holds the key to find against this disease. To have a bet- ter management for such cancer patients, we have a considerable interest in developing a better staging system for oral cancer. The stage of a cancer is a clinical descriptor, typically using the numbers I–IV, and taking into account such factors as tumor size, how deeply the tumor has pene- trated, the invasion of adjacent organs, and the state of metastasis. Cancer staging is important, because the stage at diagnosis is the most powerful predictor of sur- vival, and treatments often are based on stage. Tissue evaluation based on microscopic images is an impor- tant part of the diagnosis, staging, and prognostic tracking of cancer. One aspect of optical microscopic cancer diagnosis/staging is related to the presence and extent of a special local vascular network induced by tumors to supply oxygen and nutrients for tumor growth (Axelsson et al., 1995; Folkman, 1990; Gasin- ska et al., 2002; Ikeda et al., 1999; Khan et al., 2002; Liotta et al., 1974; Arora et al., 2002). Recent studies have attempted to improve and digi- tally automate the image-based diagnosis of cancer. Carlson et al. (2007) developed a near-real-time confo- cal microscope which combines reflectance and fluores- cence imaging technology. Such dual-mode imaging provides in vivo information regarding morphological and molecular expression during cancer progression. Assis et al. (2005) studied the expression of bcl-2 in rat tongue mucosa exposed to cigarette smoke. Jung et al. (2005) studied the early diagnosis of oral cancer based on 2D and 3D OCT (optical coherence tomography) images. To identify areas of oral cancer, Jiang et al. (2004) developed a fluorescent image system combined with a color image fusion algorithm. Rodrı ´guez et al. *Correspondence to: Yung-Nien Sun, Department of Computer Science and Information Engineering, National Cheng Kung University, University Road No. 1, Tainan, Taiwan Province, Republic of China. E-mail: ynsun@mail.ncku. edu.tw Received 9 April 2008; accepted in revised form 5 May 2009 Contract grant sponsor: National Science Council, Taiwan; Contract grant number: NSC 94-2614-E-006-075; Contract grant sponsor: Program of Top 100 Universities Advancement, Ministry of Education, Taiwan. DOI 10.1002/jemt.20746 Published online 12 June 2009 in Wiley InterScience (www.interscience.wiley.com). V V C 2009 WILEY-LISS, INC. MICROSCOPY RESEARCH AND TECHNIQUE 73:5–13 (2010)