Image registration for varicose ulcer classification using KNN classifier

Abstract

This paper proposes a clinically supportive tissue classification scheme for varicose ulcer classification using medical image processing and knowledge engineering techniques. Among various leg ulcers varicose ulcer accounts for over 90% of all cases. In this paper, we have focused on a complex image registration issue that occurs while the dependencies between intensities of images to be registered are not spatially homogeneous. In this paper, the classification scheme is divided into four distinct parts: pre-processing that includes image registration and difference image calculation, segmentation, feature extraction, KNN classification. In feature extraction step, color features like color correlogram, color moments; texture features like homogeneity, contrast, energy, correlation; shape features like solidity, eccentricity, major axis and minor axis are extracted. These extracted features are classified using K-Nearest Neighbor classifier to differentiate the different stages of wound. Our proposed approach gives efficient performance rates of average sensitivity (95.29%), specificity (94.4%), and accuracy (94.85%). Experiments conducted on varicose ulcer wound images show that the classifying registration improves both the registration and the detection, especially when the deformations are small. In future the same classification can be done using microscopic images (varicose ulcer tissue images).