Mammogram classification using dynamic time warping

This paper presents a new approach for breast cancer classification using time series analysis. In particular, the region of interest (ROI) in mammogram images is classified as normal or abnormal using dynamic time warping (DTW) as a similarity measure. According to the analogous case in time series analysis, the DTW subsumes Euclidean distance (ED) as a specific case with increased robustness due to DTW flexibility to address local horizontal/vertical deformations. This method is especially attractive for biomedical image analysis and is applied to mammogram classification for the first time in this paper. The current study concludes that varying the size of the ROI images and the restriction on the search criteria for the warping path do not affect the performance because the method produces good classification results with reduced computational complexity. The method is tested on the IRMA and MIAS dataset using the k-nearest neighbour classifier for different k values, which produces an area under curve (AUC) value of 0.9713 for one of the best scenarios.