Comparative Study of Feature Selection for White Blood Cell Differential Counts in Low Resolution Images

Features that are widely used in digital image analysis and pattern recognition tasks are from three main categories: shape, intensity, and texture invariant features. For computer-aided diagnosis in medical imaging for many specific types of medical problem, the most effective choice of a subset of these features through feature selection is still an open problem. In this work, we consider the problem of white blood cell (leukocyte) recognition into their five primary types: Neutrophils, Lymphocytes, Eosinophils, Monocytes and Basophils using a Support Vector Machine classifier. For features, we use four main intensity histogram calculations, set of 11 invariant moments, the relative area, co-occurrence and run-length matrices, dual tree complex wavelet transform, Haralick and Tamura features. Global sensitivity analysis using Sobol's RS-HDMR which can deal with independent and dependent input variables is used to assess dominate discriminatory power and the reliability of feature models in presence of high dimensional input feature data to build an efficient feature selection. Both the numerical and empirical results of experiments are compared with forward sequential feature selection. Finally, the results obtained from the preliminary analysis of white blood cell classification are presented in confusion matrices and interpreted using Cohen's kappa (i¾?) with the classification framework being validated with experiments conducted on poor quality white blood cell images.