Optimizing a Classification System for HEp-2 Cells by Evolutionary Computation

1st!International*Contest*on!! HEp!2"Cells"Classification" hosted'by'the'! 21th%International%Conference%on%Pattern%Recognition!(ICPR&2012)! ! ! ! ! ! Organizers* Gennaro*Percannella!"!([email protected],/University/of/Salerno,/Italy)! Pasquale*Foggia!"!([email protected],/University/of/Salerno,/Italy)! Paolo*Soda*5!([email protected],/University/"Campus/Bio?Medico"/of/Roma,/Italy)/ ! http://mivia.unisa.it/hep2contest! ! ! Table&of&content& ! ! Assigned team name Members and affiliation Page CHEPLYGINA V. Cheplygina, A. Ibba, D.M.J. Tax, Pattern Recognition Laboratory, Delft University of Technology, The Netherlands 8 DI CATALDO S. Di Cataldo, A. Bottino, E. Ficarra, E. Macii, Department of Control and Computer Engineering, Politecnico di Torino, Italy 9 ERSOY I. Ersoy1, F. Bunyak1, J. Peng2, K. Palaniappan1, 1Department of Computer Science, University of MissouriColumbia, Columbia, MO 65211 USA, 2Department of Computer Science, Montclair State University, Montclair, NJ 07043 US. 10 FIASCHI L. Fiaschi, F.A. Hamprecht, IWR/HCI Heidelberg. 11 GHOSH S. Ghosh, Department of Computer Science and Engineering, State University of New York (SUNY) at Buffalo, NY 14260 12 GILBERT Lim Yong San, Gilbert, National University of Singapore. 13 HASSAINE A. Hassaine, Computer Science and Engineering Department College of Engineering, Qatar University Doha, Qatar. 14 KASTANIOTIS I. Theodorakopoulos, D. Kastaniotis, Electronics Laboratory, Physics Department, University of Patras, Greece. 15 KAZANOV G.V. Ponomarev, V.L. Arlazarov, M.D. Kazanov, Research and Training Center on Bioinformatics, Institute for Information Transmission Problems, RAS, Bolshoy Karetny per. 19, Moscow, 127994, Russia. 16 KOVACS G. Kovacs, University of Debrecen Hungary. 17 KUAN 1Kuan Li, 1Jianping Yin, 2Zhi Lu, 2Xiangfei Kong, 2Rui Zhang, 2Wenyin Liu, 1National University of Defense Technology, Changsha, China, 2City University of Hong Kong. 18 MALON P.F. Laquerre, C. Malon, Department of Machine Learning, NEC Laboratories America. 19 MAREE R. Mare, GIGA Bioinformatics Core Facility and GIGA-R Bioinformatics and Modeling, Montefiore Institute, University of Lige, Belgium 20 MATEOSGARCIA D. Mateos-Garcia, J. Garcia-Gutierrez, J.C. Riquelme-Santos, Department of Computer Science, University of Seville. Avda. Reina Mercedes S/N, 41012 Seville (Spain). 21 NANNI M. Paci1, L. Nanni2, S. Severi1, 1DEIS, University of Bologna, 2DEI, University of Padua. 23 NOSAKA R. Nosaka, K. Fukui, University of Tsukuba, Japan. 24 PERSSON B. Persson, FOSS Analytical AB. 25 REZVANI M. Naghshnejad, Z. Rezvani, Shahid Beheshti University. 26 1st International Contest on HEp-2 Cells Classification 2 RUUSUVUORI T. Manninen, H. Huttunen, P. Ruusuvuori, Department of Signal Processing, Tampere University of Technology. 27 SHEN L. Shen, J. Lin, School of Computer Science and Software Engineering, Shenzhen University, Shenzhen 518060, China. 28 SNELL V. Snell, W. Christmas, J. Kittler, CVSSP, University of Surrey, Guildford, UK. 29 STOKLASA R. Stoklasa1, T. Majtner1, David Svoboda1, Michal Batko2, 1Center for Biomedical Image Processing, Masaryk University, Brno, Czech Republic, 2Laboratory of Searching and Dialogue, Masaryk University, Brno, Czech Republic. 30 STRANDMARK P. Strandmark, J. Ulen, F. Kahl, Centre for Mathematical Sciences, Lund University, Sweden. 31 THIBAULT G. Thibault, J. Angulo, CMM, MINES-ParisTech. 32 WAFA Wafa Bel haj ali, M. Barlaud, U. Nice - Sophia Antipolis / CNRS, France. 33 WANG L. Liu1, L. Wang2, 1CECS, Australian National University ACT 0200, Canberra, Australia, 2School of Computer Science and Software Engineering University of Wollongong, NSW 2522, Australia. 34 WILIEM A. Wiliem1, Y. Wong1,3, C. S1,3, P. Hobson2, S. Chen1,3, B.C. Lovell1,3, 1NICTA, 2Sullivan Nicolaides Pathology, Australia, 3School of ITEE, University of Queensland, Australia. 36 XIANGFEI Xiangfei Kong, Kuan Li, Zhi Lu, Liu Wenyin, City University of Hong Kong 37 1st International Contest on HEp-2 Cells Classification 3 Optimizing a Classification System for HEp-2 Cells by Evolutionary Computation Daniel Mateos-García, Jorge García-Gutiérrez, J. Cristóbal Riquelme-Santos Department of Computer Science, University of Seville. Avda. Reina Mercedes S/N, 41012 Seville (Spain) [email protected]s (Daniel Mateos-García), jorgarc[email protected] (Jorge García Gutiérrez), [email protected] (J. Cristóbal Riquelme-Santos) Abstract In this work, we describe a classification system to automatically recognize the pattern of HEp-2 cells within IIF images. For this purpose we have carried out several steps to preprocess the data, select a proper predictor and generate a model. The use of Evolutionary Computation in the feature selection step and the optimization of the classification system is the main contribution of this work. 1. Method The proposed pattern recognition system has been carried out following the next steps (see Figure 1): 1. We have chosen a standard number of pixels for all the data. In our case the considered resolution for the training and testing steps has been of 100x100. The system resizes the image data to the mentioned resolution. 2. With respect to the training data, every image was transformed to one instance and included into a unique training file with ARFF format [4]. 3. The number of features (pixels) of the training file was reduced with the CFS evaluator and considering the Best First algorithm as searching method [1]. In a second phase, a Genetic Algorithm (GA) improved the quality of the final set of features. The fitness of the GA is the accuracy of the classification subsystem that will be described in the next step. To obtain the average of the accuracy, the training file was randomly divided into two folds and then they were used as training and testing data twice (2 x holdout 85%-15%). 4. The classification subsystem used for label predictions is J48 (C4.5) [2]. This predictor has been boosted by the AdaBoost M1 method [3]. This decision has been made by discarding the rest of the classifiers that integrate the framework used (Weka [4]) and taking into account the observed accuracy of each one. Furthermore, a GA optimized the AdaBoost parameters. Regarding the GA design, each individual represents a combination of values of these parameters, and the fitness is the accuracy of the classification subsystem using holdout (2 x holdout 85%-15%) with the pre-processed training data (features selected). 5. The final classifier system was trained with the full pre-processed training data and the optimized AdaBoost + C4.5 stack. In the testing step, each testing image will be resized, transformed in an ARFF file with only one instance and unknown label, and filtered with the same feature set calculated in the training step. References [1] M. A. Hall (1998). Correlation-based Feature Subset Selection for Machine Learning. Hamilton, New Zealand. 1st International Contest on HEp-2 Cells Classification 21 [2] Ross Quinlan (1993). C4.5: Programs for Machine Learning. Morgan Kaufmann Publishers, San Mateo, CA. [3] Yoav Freund, Robert E. Schapire: Experiments with a new boosting algorithm. In: Thirteenth International Conference on Machine Learning, San Francisco, 148156, 1996. [4] Weka Machine Learning Project. Weka. URL http://www.cs.waikato.ac.nz/ml/weka/ Figure 1. Training and testing processes 1st International Contest on HEp-2 Cells Classification 22