gms | German Medical Science

Objective: Up to now, manual tractography by choosing and outlining suitable inclusion and exclusion regions of interest is still the standard approach in neurosurgical planning strongly depending on the user’s skills and experience. Therefore, automatic tractography with less or even without manual interaction should be taken more into account as some supervised and unsupervised approaches already exist. In this study, an automated tractography approach is therefore compared to traditional fiber tractography manually defining proper regions of interest in case of the corticospinal tract (CST).

Methods: 15 healthy volunteers were included in the study. For each volunteer five repetitive diffusion weighted imaging data sets were acquired at a 3T MRI. After preprocessing applying motion and eddy current correction, manual fiber tractography was performed in iPlan (Brainlab, Feldkirchen, Germany). Automated fiber tractography was performed in AFQ (automatic fiber quantification, VistaLab, Stanford University, USA), both utilizing a deterministic tractography approach. To compare both methods, tract volumes and fractional anisotropy (FA) measures were compared across groups. Additionally, the intraclass correlation coefficient (ICC) was applied across groups for tract volume and FA as measure of reliability.

Results: In case of the left CST, mean tract volume was 13.37±2.14 ml using the traditional approach in contrast to 11.75±2.75 ml using the automated approach (p<0.001). For the right CST, the user-based approach led to a mean tract volume of 14.04±2.38 ml in contrast to 11.86±3.13 ml for the automated approach (p<0.001). Analyzing FA within the reconstructions of the CST, mean FA values were significantly larger in automated reconstructions for both hemispheres (p<0.001). Applying the ICC as measure of reliability across repeated measures, the ICC was 0.742 (manual) and 0.899 (automated) for the left CST and 0.829 (manual) and 0.940 (automated) for the right CST. ICCs regarding FA were 0.873 and 0.834 for the left and right CST applying the manual approach, whereas corresponding ICCS were 0.806 and 0.741 applying the automated approach.

Conclusion: In automated tractography the tract volume was significantly decreased with increased FA values within the reconstructed tract and showed more reproducible results. Even the volume of major matter tracts might be underestimated, more automatic tractography approaches should be developed to overcome the limitation of user bias.