Mammographic Feature Characterization Using Two-Dimensional Empirical Mode Decomposition

Abstract

Empirical mode decomposition (EMD) is a self adaptive method that decomposes non-linear non-stationary signals into different frequency components Bi-dimensional EMD (BEMD) is introduced to deal with the image data in a two-dimensional space However with BEMD the counting of pixel relationship is constrained within two (x and y) or limited directions during the interpolation of upper and lower envelopes Hence we propose a new two-dimensional method to solve the situation The new 2DEMD algorithm has been successfully developed and applied in digital mammography to analyze image features and for lesion enhancement The new 2DEMD we proposed in this paper combines the techniques of image projection / back-projection and conventional 1DEMD Radon Transform is used to project the two-dimensional image into one-dimensional space in every angle Subsequently we apply the conventional EMD in one-dimensional space to get the intrinsic mode functions (IMFs) Then inverse Radon Transform is used to rebuild the two-dimensional IMFs to accomplish the 2DEMD calculations In order to validate the utility of this method normal mammography images with different densities and abnormal mammography with a variety of lesions were collected These images were then interpreted and analyzed in terms of their IMF distribution Normal fibroglandular tissue is mainly seen in IMF4 to IMF6 the middle frequency components The thin lines of spiculated margin and architectural distortion are mainly visible in high frequency components namely IMF2 and IMF3 Thick lines are visible in subsequent IMF4 Microcalcifications and margins of coarse calcifications are mainly visible in IMF1 and IMF2 followed by small coarse calcifications in IMF3 and larger coarse calcifications in IMF4 Masses of either benign or malignant nature are mainly visible in IMF5 and IMF6 Using a combination of Radon transform and empirical mode decomposition we successfully establish a new 2DEMD technique to tackle the problematic pixel relationship in different angles This novel 2DEMD technique allows us to characterize the IMF features of normal structures and lesions Furthermore these features can be used to enhance the mammographic lesions for better visual perception This achievement will contribute to mammographic imaging diagnosis and also provide the fundamental EMD concepts for future research in mammography

Date of Award	2014 Sept 2
Original language	English
Supervisor	Kuo-Sheng Cheng (Supervisor)