DWI filtering using joint information for DTI and HARDI.

Area of Science:

• Medical Imaging
• Neuroimaging
• Diffusion MRI

Background:

• Diffusion Weighted Images (DWI) filtering is crucial before diffusion tensor or Orientation Distribution Function (ODF) estimation.
• Ignoring filtering or filtering gradients separately introduces unrecoverable errors in tensor field estimation.

Purpose of the Study:

• To develop a novel methodology for filtering DWI data by leveraging joint information across all gradient images.
• To adapt and evaluate this joint filtering approach using Linear Minimum Mean Squared Error (LMMSE) and Unbiased Non-Local Means (UNLM) filters.

Main Methods:

• A new methodology that filters all DWI gradient images simultaneously, exploiting shared first and second-order information.
• Adaptation of LMMSE and UNLM filters to process DWI data jointly.
• Testing the proposed filters on both synthetic and real diffusion MRI datasets.

Main Results:

• The joint filtering approach significantly improves the accuracy of diffusion tensor and ODF estimation.
• Joint LMMSE demonstrates accuracy comparable or superior to UNLM, particularly for High Angular Resolution Diffusion Imaging (HARDI).
• The joint LMMSE method offers a substantial reduction in computational load compared to UNLM.

Conclusions:

• Jointly filtering DWI volumes is superior to individual gradient filtering for accurate diffusion modeling.
• The proposed joint LMMSE filter is an efficient and accurate method for DWI data processing, especially beneficial for HARDI.
• This approach enhances the reliability and reduces the computational burden of diffusion MRI analysis.