Multiresolution/multiorientation based nonlinear filters for image enhancement and detection in digital mammography

Issue title: Special Issue on Clinical Applications of Modern Imaging

Article type: Research Article

Authors: Qian, Wei | Sun, Xuejun | Clark, Robert

Affiliations: Department of Interdisciplinary Oncology and Radiology, College of Medicine and the H. Lee Moffitt Cancer and Research Institute, University of South Florida, Tampa, FL 33612, USA

Note: [] Corresponding author: Wei Qian, PhD, Department of Interdisciplinary Oncology, H. Lee Moffitt Cancer Center & Research Institute, University of South Florida, 12902 Magnolia Drive, Tampa, FL 33612-9497, USA. Tel.: +1 813 975 7874; E-mail: qianw@moffitt.usf.edu

Abstract: In this paper a novel hybrid filter architecture is presented that incorporates an adaptive multistage nonlinear filter (AMNF) and Multiresolution/Multiorientation Wavelet Transform (MMWT) specifically designed for image enhancement in medical imaging. The MMWT is used for multiresolution and/or multiorientation decomposition and reconstruction of the mammograms where selective reconstruction of subimages is used to perform further enhancement of the MCCs and masses. The specific clinical application is enhancement of microcalcification clusters (MCCs) and masses in digitized mammograms and the AMNF is used for noise suppression and image enhancement. This novel filter was evaluated using diagnostic images from two databases: (a) and (b). Database (a) has 100 single view mammograms at a resolution of 105 micron and 12 bits gray level. It contains 50 cases of normal mammograms and 50 cases of biopsy proven malignant MCCs. Database (b) has 100 single view mammograms at a resolution of 180 micron and 12 bits gray value. It contains 50 normal and 50 biopsy proven abnormal masses. Visual inspection of the enhanced images demonstrated improved visualization of MCCs/masses and the morphology of the individual subtle microcalcifications and masses compared with unprocessed images. The performance of the enhancement method was quantitatively evaluated using a Kalman Filtering Neural Network for MCCs and mass detection and applied to the digital mammograms of the database (a) and (b). The sensitivities (True Positive (TP) detection rate) of MCC cluster detection were 93% images respectively, and the average false positive (FP) detection rate was 1.35 FP and 1.47 MCCs/image respectively. The sensitivities of mass detection were 85% unprocessed images respectively and the average FP rates were 1.81 and 2.75 masses/image respectively. The Research results in this article demonstrate the importance of the hybrid filter architecture for enhancement in the computerized detection of MCCs and mass.

Journal: Journal of X-Ray Science and Technology, vol. 10, no. 1-2, pp. 1-15, 2002