Related Experiment Video
Updated: Jul 27, 2026
17:06
Co-analysis of Brain Structure and Function using fMRI and Diffusion-weighted Imaging
Published on: November 8, 2012
26.9K
Manual vs. AI-based tractography: Assessing fractional anisotropy consistency, applicability and methodological
Adrian Korbecki1, Maja Gewald2, Krzysztof Winiarczyk1
1Department of General and Interventional Radiology and Neuroradiology, Medical University Hospital, Wroclaw, Poland; Hetalox sp. z o.o., Wroclaw, Poland.
Computers in Biology and Medicine
|October 14, 2025
Summary
Comparing manual diffusion tensor imaging (DTI) tractography and AI-based TractSeg reveals significant differences in fractional anisotropy (FA) values. These FA metrics are not interchangeable, highlighting the need for standardization in white matter pathway analysis.
Area of Science:
- Neuroimaging
- Computational Neuroscience
- Medical Image Analysis
Background:
- Diffusion Tensor Imaging (DTI) and Constrained Spherical Deconvolution (CSD) are crucial for mapping white matter tracts.
- Discrepancies in Fractional Anisotropy (FA) values arise from differing tractography methodologies.
- Understanding these differences is vital for accurate interpretation of brain connectivity studies.
Purpose of the Study:
- To compare Fractional Anisotropy (FA) measurements between manual DTI deterministic tractography and an automated AI-based approach (TractSeg).
- To assess the interchangeability of FA values derived from these distinct tractography methods.
- To identify potential biases and discrepancies in white matter pathway reconstruction.
Main Methods:
- Brain MRI scans from 30 healthy adults were analyzed.
- Nine major white matter tracts were reconstructed using both DTI-based vendor software and CSD with TractSeg.
- FA values were compared using inter-rater reliability metrics, including Intraclass Correlation Coefficients (ICCs), and normalized values.
Main Results:
- Substantial differences in FA values were observed between manual DTI and AI-based TractSeg methods, with poor to moderate ICCs for most tracts.
- Manual DTI methods generally yielded higher FA values, particularly in the Corpus Callosum (CC) and inferior fronto-occipital fasciculus.
- AI-based TractSeg showed higher FA for the uncinate fasciculus and consistently produced larger tract volumes, though volume did not correlate with FA reliability.
Conclusions:
- FA values obtained from manual DTI and AI-based TractSeg are not interchangeable due to significant inter-method discrepancies.
- The findings underscore the need for methodological standardization to ensure reliable and comparable results across neuroimaging studies.
- AI-based methods may offer advantages for smaller, complex fibers, but require further validation for consistent FA quantification.
