Research on Translation-Invariant Wavelet Transform for Classification in Mammograms

Classification of benign and mat microcalcifications in mammograms through computer-aided diagnosis (CADx) is vital for the early diagnosis of the breast cancer. To this end, wavelet-based textural feature has been proved to be an effective feature extraction method. However, a majority of these methods is restricted to decimated wavelet transform, which lacks the property of translation invariance that is useful in signal processing. In this paper, we apply the translation-invariant (TI) wavelet transform to microcalcifications classification. A set of features, combining the TI wavelet based features and co-occurrence features, is employed to get better classification results than the conventional methods. The area under ROC curve ranged from 0.87 to 0.91 when using the proposed method. Experimental results show that the TI-wavelet method outperforms the one based on multiwavelet, which achieved the best results in 2004 on the same database as ours.