Feature subset selection using ICA for classifying emphysema in HRCT images

Feature subset selection, applied as a pre-processing step to machine learning, is valuable in dimensionality reduction, eliminating irrelevant data and improving classifier performance. In recent years, data in some applications has increased in both the number of instances and features. It is in this context that we introduce a novel approach to reduce both instance and feature space through independent component analysis (ICA) for the classification of emphysema in high resolution computer tomography (HRCT) images. The technique was tested successfully on 60 HRCT scans having emphysema using three different classifiers (Naive Bayes, C4.5 and Seeded K Means). The results were also compared against "density mask", a standard approach used for emphysema detection in medical image analysis. In addition, the results were visually validated by radiologists.