Wavelet transform for directional feature extraction in medical imaging

Abstract

Directional features are extremely important in diagnostic imaging. They can be used either in the searching for important indications of abnormality or for the reference in image alignment. The wavelet transform (WT) as an efficient method for multiresolution representation has been used extensively in medical imaging. However, from the view point of directional selectivity it is not so efficient in feature extraction. In this paper, an analysis and comparison of directional selectivity of wavelet transforms with three different frequency decompositions were undertaken. WTs with radial-angular decomposition were used for two medical imaging problems: mass detection in digital mammography and lung nodule detection in chest radiography. The results demonstrated its efficiency.