Automated Decision Tree Classification of Keratoconus From Videokeratography

Classification and prediction are common tasks in the biomedical sciences. Several machine learning classification methods exist including Bayesian classifiers, decision trees, neural networks, statistical regression, and others. Each approach has unique strengths that make it more or less appropriate for a particular classification problem. Decision trees have emerged as one of the most versatile and robust classification methods, and have been widely applied to medical diagnosis problems. We have developed a quantitative method of corneal shape classification to discriminate keratocons from normal videokeratograhy. This decision tree induction method can be applied to the raw data output from any videokeratography instrument platform, and could be used to analyze both group and longitudinal data. In this study we describe an alternative method of corneal shape classification based upon Zernike polynomials and compare performance of this classification method with existing classifiers for a sample of normal eyes and eyes diagnosed with keratoconus.