SIFT with associative classifier for mammogram classification

In this proposed work, a Scale-Invariant Feature Transform (SIFT) with improved associative classifier is used for effective classification of mammograms. SIFT is used to extract distinctive invariant features, from region of interest (ROI) of mammograms. However, SIFT features are of very high dimension, and large number of features are generated for a mammogram, resulting in an increase in feature space and search space for matching. Authors proposed PreARM algorithm, to optimize the number of SIFT features. Transaction database consists of optimized feature vector and class of training images, and is given as input to association rule mining. Multi-fitness function Genetic algorithm is used to optimize association rules generated using Apriori algorithm. An experimental result shows that PreARM algorithm achieves 91% reduction in features and Genetic algorithm achieves 90% reduction in association rules. Standard DDSM medical image dataset is used to validate the proposed method. Optimized rules are used for classification of mammograms. Proposed SIFT based associative classifier gives classification accuracy as 93.75% and the area under receiver operating characteristic (ROC) curve value as 0.932.