Filtering Techniques for Diffusion Tensor Data

  1. What are we doing ?

    We are trying to build robust filters for removing noise from Diffusion Tensor Data.

  2. Why is filtering DT data important ?

    • With the advent of DT-MRI imaging as a non invasive technique for studying tissue structure, Diffusion Tensor data is becoming more and more common.
    • The nature of the Magnetic Resonance DT-imaging process results in inherently low SNR (Signal to Noise Ratio) in the estimated Difusion Tensors.
    • There is a need for robust and effective filters for DT-Data for DT based application to be useful.

  3. What characteristics should our filter possess ?

    • We want to ensure that we preserve feature discontinuities in tensor data.
    • We want the filter to be computationally efficient.
    • We want the filtered tensors to be symmetric and positive definite.

  4. What Filtering Techniques have we tried out? How are the results?

    We have experimented with 3 different filtering ideas till now on a synthetically generated data set.
    These are some latest results on real data. The corpus collosum of the brain is visible.
    1. Data Set -(HUVA00102397) Results
    2. Data Set -(HUVA98189432) Results
    3. Data Set -(UNC -Repeated Scans) Results