Evolution oriented semi-supervised approach for segmentation of medical images

Abstract

Segmentation of medical images plays a central role in intelligent image analysis and understanding. This paper presents a novel evolution oriented semi-supervised (EOS) approach for segmentation and labelling of medical images. The segmentation method is based on a semi supervised classifier. The classifier, which can evolve with the introduction of new classes and can accommodate corrections made by human experts in the existing class, is developed using adaptive K-means clustering and ripple down rule (RDR) concepts. For classifying pixels of the image to obtain homogeneous segments of a specific class we use feature vectors derived from DCT coefficients. We tested the method on high resolution computed tomography (HRCT) lung images which contain patterns of emphysema and ground glass opacities.