IEEE TRANSACTIONS ON INFORMATION TECHNOLOGY INBIOMEDICINE, VOL. 16, NO. 6, NOVEMBER 2012 1127 Multiparametric Decision Support System for the Prediction of Oral Cancer Reoccurrence Konstantinos P. Exarchos, Yorgos Goletsis, Member, IEEE, and Dimitrios I. Fotiadis, Senior Member, IEEE Abstract—Oral squamous cell carcinoma (OSCC) constitutes the predominant neoplasm of the head and neck region, featuring par- ticularly aggressive nature, associated with quite unfavorable prog- nosis. In this paper, we formulate a decision support system that integrates a multitude of heterogeneous data (clinical, imaging, and genomic), thus, framing all manifestations of the disease. Our pri- mary aim is to identify the factors that dictate OSCC progression and subsequently predict potential relapses (local or metastatic) of the disease. The discrimination potential of each source of data is initially explored separately, and afterward the individual predic- tions are combined to yield a consensus decision achieving complete discrimination between patients with and without a disease relapse. Index Terms—Classification, decision support system (DSS), gene expression, oral cancer, reoccurrence prediction. I. INTRODUCTION M ALIGNANT neoplasms constitute the second cause of death in the western world; oral cancer, in particular, is the 8th most common neoplasm according to the worldwide cancer incidence ranking [1]. In the literature, several risk fac- tors have been associated with the induction and progression of oral cancer; excessive smoking and especially when coupled with alcohol consumption constitute predominant factors of oral squamous cell carcinoma (OSCC) development. A significant correlation has also been identified between oral cancer and the sex of the patients, with men facing twice the risk of being diagnosed with OSCC [1]. Moreover, extensive sun exposure has been proven a proliferating factor of OSCC development, particularly for neoplasms of the lip [1]. Besides the low quality of everyday life that is inherent to OSCC patients, a major issue is the high occurrence rate of lo- coregional relapses, which often leads to considerable inability and even death. Specifically, after successful treatment of the disease, there is a state called remission where the patient is considered cancer free, and particles of the primary tumor mass are no longer detectable. Patients in remission have a proba- bility of 25–48% to develop a disease relapse [2] owed to the Manuscript received April 15, 2011; revised July 9, 2011; accepted August 6, 2011. Date of publication August 18, 2011; date of current version November 16, 2012. This work was supported by the European Commission NeoMark Project (FP7-ICT-2007-224483—ICT enabled prediction of cancer reoccurrence). K. P. Exarchos and D. I. Fotiadis are with the Department of Materials Science and Engineering, Unit of Medical Technology and Intelligent Infor- mation Systems, University of Ioannina, Ioannina, GR 45110, Greece (e-mail: kexarcho@cc.uoi.gr; fotiadis@cs.uoi.gr). Y. Goletsis is with the Department of Economics, University of Ioannina, Ioannina, GR 45110, Greece (e-mail: goletsis@cc.uoi.gr). Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/TITB.2011.2165076 deeply infiltrative nature of the disease. Due to the aggressive nature of OSCC and high rates of locoregional relapses, early identification of a potential disease reoccurrence can prove very beneficial for the prognosis of the patient [3] and the subsequent fine tuning of the followup treatment. To this end, several studies have been conducted in the liter- ature aiming to predict a relapse after the disease has reached remission; yet, the overall results remain unsatisfactory. Specif- ically, Roepman et al. [4], [5] have extracted a subset of genes able to identify lymph node metastasis from a primary head and neck squamous cell carcinoma. In a similar manner, Rickman et al. [6] exploit gene expression information toward the predic- tion of future metastasis of a neoplasm originating in the region of the head and neck. In the same context, Zhou et al., Watanabe et al., and Nagata et al. [7]–[9] have focused on carcinomas of the tongue, yielding a subset of genes able to predict nodal metastasis. In [10], an artificial neural network (ANN) is em- ployed in order to predict the metastasis of primary esophageal cancer in the adjacent lymph nodes. In [11], a gene subset is ex- tracted aiming to predict lymph node metastasis from an initial oral squamous cell carcinoma. In this paper, we propose an orchestrated approach in order to systematically study and analyze the multifactorial basis of oral cancer evolvement. Three sources of data are employed, namely clinical, imaging, and genomic in pursue of the most prominent OSCC proliferating factors. This multifaceted approach and the complementary analysis of the data are prone to capture all pos- sible manifestations of the disease. Subsequently, the identified factors are utilized in a classification scheme in order to predict a potential disease reoccurrence and discriminate the patients in terms of relapse probability. The timely and accurate identifica- tion of a disease relapse, can substantially determine the most proper treatment, based on each patient’s specific prognosis. II. MATERIALS AND METHODS The flowchart of the proposed methodology is shown in Fig. 1. Initially, all sources of data (clinical, imaging, and ge- nomic) are subject to certain basic preprocessing steps in order to enhance the quality of the input data that are further fed to the next steps. Next, we either feed the input data directly for classification or we employ a feature selection algorithm to omit potentially redundant features. Besides the individ- ual clinical/imaging/genomic-based classifiers, we implement a consensus classifier that combines the aforementioned predic- tions in a complementary manner in order to procure a more accurate outcome. 1089-7771/$31.00 © 2012 IEEE