Classification of prostatic tissues using feature selection methods

Abstract

This paper proposes the use of sequential feature selection for classification of prostatic tissues. The technique aims to classify microscopic samples taken by needle biopsy for the purpose of prostate Cancer diagnosis. Four major classes (representing different grades of abnormality from normal to cancer respectively: Stroma, BPH, PIN, PCa) have to be discriminated. To achieve that, the same feature vector, based on texture measurements, was derived for each class. Haralick features have been used to describe textures. Sequential forward selection (SFS) and sequential backward selection (SBS) has been used to reduce the dimensionality of the generated feature vector into a manageable size. Tests have been carried out using k nearest neighbor (kNN) method and have shown that the use of feature selection algorithms SFS and SBS can significantly improve the classification performance.