Retrieval in Dermatology using Intelligent Techniques

Abstract

Identification of skin disease has become a challenging task with the origination of various skin diseases. The architecture used in this work is content-based image retrieval to facilitate medical diagnosis. In this work, analysis of feature vectors based on fusion of colour and texture and shape features are carried out. The feature vector data are fed into an optimization framework. The results proved that using precision recall curve the shape feature vectors and Mahalanobis distance measure have high contribution to computer-aided diagnosis of skin lesions. Experiments on a set of 2420 images yielded a specificity of 97.04% and a sensitivity of 80.81%. Our empirical evaluation has a superior retrieval and diagnosis performance when compared to the performance of other works. The work is tested with 40 skin diseases.