Noise removal in magnetic resonance diffusion tensor imaging.

Although promising for visualizing the structure of ordered tissues, MR diffusion tensor imaging (DTI) has been hampered by long acquisition time and low spatial resolution associated with its inherently low signal-to-noise ratio (SNR). Moreover, the uncertainty in the DTI measurements has a direct impact on the accuracy of structural renderings such as fiber streamline tracking. Noise removal techniques can be used to improve the SNR of DTI without requiring additional acquisitions, albeit most low-pass filtering methods are accompanied by undesirable image blurring. In the present study, a modified vector-based partial-differential-equation (PDE) filtering formalism was implemented for smoothing DTI vector fields. Using an image residual-energy criterion to equate the degree of smoothing and error metrics empirically derived from DTI data to quantify the relative performances, the effectiveness in denoising DTI data is compared among image-based and vector-based PDE and fixed and adaptive low-pass k-space filtering. The results demonstrate that the edge-preservation feature of the PDE approach can be highly advantageous in enhancing DTI measurements, particularly for vector-based PDE filtering in applications relying on DTI directional information. These findings suggest a potential role for the postprocessing enhancement technique to improve the practical utility of DTI.