Comparison of preprocessing methods for diffusion tensor estimation in brain imaging

Abstract

The Magnetic Diffusion Images or Diffusion Weighted Images (DWIs) are based on Magnetic Resonance (MR) techniques to study water particles' diffusion in human brains. These images are used for determining which neuron pathways were used for the communication among the principal regions of the brain by estimating the diffusion tensors (DTs). DTs contain all the information of water diffusion for each individual voxel of the image. The filtering of these images is relevant to remove the level of noise of each image and improve the DTs estimation. Moreover, the smoothing methods may be used to reduce noise in medical images. However, certain smoothing filters may blur important features such as edges and also affect structures, so it is essential to preserve the fine features using anisotropic diffusion filtering. Therefore, we need to preprocess this type of brain images by removing noise, smoothing surfaces and enhancing edges, are necessary to improve the results of estimating the DTs. This paper formalizes and compares the advantages and disadvantages obtained by applying different kinds of preprocessing techniques for removing noise, smoothing surfaces and enhancing edges techniques include Median Filter (MF), Perona-Malik algorithm and Gaussian filter (GF). Then, in order to determine the potential benefits of the mentioned pre-processing, the DTs are estimated with and without using the filter stage. In addition, several metrics are used for the evaluation and comparison of the DWI preprocessing methods. Finally, we discuss the quality of these methods and we also define what are the appropriate conditions for each preprocessing method.