Mammogram retrieval based on heterogeneous attributes using LeGall 5/3 wavelet and CART algorithm

Mammogram retrieval systems and techniques have a key role in supporting the radiologists and physicians in their decision making. This paper presents an approach based on the biorthogonal wavelet and decision tree methods for the retrieval of mammograms from digital database for screening mammogram (DDSM) database. Classification and regression tree (CART) algorithm and LeGall 5/3 wavelet are considered for decision tree construction and image decomposition, respectively. For each mammogram the contextual attributes derived from the patient files of the DDSM database and statistical modeling of wavelet coefficient distributions forms the heterogeneous feature set. The mean precision rate attained in this work is 84%. The inclusion of decision tree has reduced the similarity measurement computation and speeds up the retrieval process. The results show that the performance of LeGall 5/3 wavelet towards mammogram retrieval is comparatively much better than using Daubechies wavelet.