Schilling, Kurt G.; Cieslak, Matthew; Descoteaux, Maxime; Landman, Bennett A.; Pestilli, Franco; Rokem, Ariel; Sotiropoulos, Stamatios N.; Tournier, Jacques-Donald; Veraart, Jelle (2026).听.听Brain Structure and Function, 231(3), 48.听
Diffusion MRI tractography, a method used to map the brain鈥檚 wiring by tracking water movement along nerve fibers, is highly sensitive to noise and imaging errors. These issues can distort the data and lead to inaccurate results if not properly addressed. Evidence shows that applying modern聽preprocessing聽techniques鈥攎ethods used to clean and correct raw imaging data鈥攕ignificantly improves both the accuracy of the resulting brain maps and their consistency when scans are repeated.
Key preprocessing steps include聽denoising聽(removing random noise), correcting for motion and scanner-related distortions such as聽eddy currents听补苍诲听EPI distortions聽(which can warp images), and removing artifacts like聽Gibbs ringing聽(false edge patterns). Newer tools now combine these steps into standardized, easy-to-use workflows. Overall, careful data preparation, along with good scanning and data-handling practices, is essential for producing reliable and meaningful maps of brain connectivity.
Fig 1: We show the impact of individual preprocessing steps (denoising, Gibbs ringing correction, susceptibility-induced geometric or 鈥淓PI鈥� distortion correction, and eddy-current (EC) distortion correction on fiber orientation distribution functions (fODFs). For each preprocessing step (rows), an example voxel location is indicated on an anatomical image (cross-hairs), and the corresponding fODF is shown before (left) and after (right) applying the indicated correction. Denoising has been shown to reduce the number of spurious fibers (Veraart et al.听,听), Gibbs ringing correction results in sharper signal kernels and fODFs (Kellner et al.听), susceptibility-induced geometric distortion correction improves alignment of diffusion and structural MRI data (Andersson et al.听), lowering the risk of streamline traversing the ventricles for example, and eddy current distortions sharpen the fODFs by reducing signal scattering across dMRI data