A Probabilistic SVM Approach to Annotation of Calcification Mammograms

Abstract

Due to the increasing use of digital medical images, a need exists to develop an approach for automatic image annotation, which provides textual labels for images. Thus added labels can be used to access images using textual queries. Automatic image annotation can be separated into two individual tasks: feature extraction and image classification. In this paper, the authors present feature extraction methods for calcification mammograms. The resultant features, based on BI-RADS standards, make annotated image contents represent the correct medical meaning and tag correspondent terms. Furthermore, this paper also proposes a probabilistic SVM approach to image classification. Finally, the experimental results indicate that the probabilistic SVM approach to image annotation can achieve 79.5% in the average accuracy rate.