Randomized iterative spherical-deconvolution informed tractogram filtering.

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Diffusion MRI

Background:

• Brain connectivity studies rely on tractography for nerve fiber reconstruction.
• Current tractography methods produce many anatomically implausible streamlines, impacting reliability.
• Existing tractogram filtering methods aim to remove these faulty connections post-processing.

Purpose of the Study:

• To address the limitations of Spherical-deconvolution Informed Filtering of Tractograms (SIFT) in assessing individual streamline compliance.
• To develop a method for identifying anatomically plausible streamlines with high confidence.
• To create a classifier trained on reliable streamline data for improved tractogram quality.

Main Methods:

• Applying SIFT to randomly selected tractogram subsets to obtain multiple compliance assessments per streamline.
• Utilizing consistently filtered streamlines from subset analyses as pseudo-ground truths.
• Training a classifier to distinguish between compliant and non-compliant streamlines based on acquired diffusion MRI data.

Main Results:

• The proposed method generates multiple SIFT-based assessments for each streamline.
• Consistently filtered streamlines were effectively used as training data.
• The trained classifier achieved an accuracy exceeding 80% in differentiating plausible from implausible streamlines.

Conclusions:

• The novel approach enhances the reliability of tractogram filtering by enabling individual streamline assessment.
• This method significantly improves the ability to identify anatomically plausible nerve fiber reconstructions.
• The findings contribute to more accurate and dependable brain connectivity analyses using diffusion MRI.