There is a growing interest in using multiresolution noise filters in a variety of medical imaging applications. We review recent wavelet denoising techniques for medical ultrasound and for magnetic resonance images and discuss some of their potential applications in the clinical investigations of the brain. Our goal is to present and evaluate noise suppression methods based on both image processing and clinical expertise. We analyze two types of filters for magnetic resonance images (MRI): noise suppression in magnitude MRI images and denoising blood oxygen level-dependent (BOLD) response in functional MRI images (fMRI). The noise distribution in magnitude MRI images is Rician, while the noise distribution in BOLD images has been recently shown to follow a Gaussian model well. We evaluate different methods based on signal to noise ratio improvement and based on the preservation of the shape of the activated regions in fMRI. A critical view on the problem of speckle filtering in ultrasound images is given where we discuss some of the issues that are overlooked in many speckle filters like the relevance of the "speckled texture", expert-defined features of interest and the reliability of the common speckle models. We analyze the use of multiresolution speckle filters to improve the automatic processing steps in the clinical research of non-cystic periventricular leukomalacia. In particular we apply speckle filters to ultrasound neonatal brain images and we evaluate the influence of the filtering on the effectiveness of the subsequent classification and segmentation of flares of affected tissue in comparison with the manual delineation of clinicians.
Keywords: Image denoising, wavelets, magnetic resonance imaging, ultrasound, statistical parametric mapping, false discovery rate control