Feature selection for computer-aided angle closure glaucoma mechanism detection

Abstract

Selection of relevant features is of fundamental importance in building robust classifiers for computer-aided detection (CAD) of angle closure glaucoma mechanism. Typically one is interested in determining which, of a large number of potentially redundant or noisy features, are most discriminative for classification. The objective of the paper is to exploit machine learning algorithms for automated classification of different angle closure mechanisms based on the quantitative assessment of Anterior Segment Optical Coherence Tomography (AS-OCT). In this paper, we propose an effective combination of Minimum Redundancy Maximum Relevance (MRMR), a mutual information feature selection, with AdaBoost, an adaptive boosting to detect angle closure glaucoma. The proposed method effectively combines the best feature selection and classifier to the problem. A sequential forward search was conducted to determine the optimal feature subset by the proposed criteria. The optimal selected feature subset, using only 11.90% of the entire available feature, for our angle closure glaucoma dataset outperforms the result using all 84 features. It also achieves better prediction compared to 4 other methods, i.e., Classification Tree, Support Vector Machine (SVM), Random Forest and Naïve Bayes. The reduced set of features avoids computation of unnecessary features and thus improves the efficiency. Furthermore, 9 out of the 10 selected features have been clinically proven to be important in determining the type of angle closure glaucoma. The algorithm show promising and encouraging results to detect and determine the type of angle closure glaucoma which may help early recognition and treatment of the disease. The reduced complexity of the generated models achieves better generalization and improves the efficiency. Copyright © 2012 American Scientific Publishers. All rights reserved.