Texton analysis for mass classification in mammograms

• A mass classification method is proposed without segmenting mass contours.
• Texton is applied to classify mass in mammograms.
• Subsamples with unequal intervals capture different discriminating structures.
• Combined with different subsamples, texton becomes less scale dependent.

In this paper, a novel method is proposed to classify masses as benign or malignant in mammograms without segmenting the contour of each mass. This method combines the scheme of texton analysis with multiple subsampling strategies. Before performing texton-based classification, intensity and rotation normalization are applied. Then subsampling strategies with either uniform or non-uniform intervals are generated and a k-nearest-neighbor (KNN) classifier is trained for each subsampling strategy. Every subsampling strategy captures one discriminating structure. To take advantage of all possible discriminating structures, the final classification result is obtained by performing majority voting over all subsampling strategies. The proposed method is tested on 114 mass regions from Digital Database for Screening Mammography (DDSM) database. The classification accuracy rate reaches 85.96%, which is higher than some existing texture-based methods.