M INISTERIO DE D EFENSA N ACIONAL
RESOLUCIÓN NÚMERO 0173 DE 2013 (febrero 22)
This section has presented a tractography algorithm which combines structural infor-mation drawn from voxels in the immediate neighbourhood of the tracking location with information from a local model capturing sub-voxel fibre orientation dispersion. By cap-turing fibre dispersion in the model underpinning the tractography algorithm, we can ad-dress the underestimation of connectivity caused by underestimation of the true spread of directions available in a voxel exhibiting orientation dispersion in the underlying fi-bre architecture. Furthermore, the results show that by forming a joint distribution in a neighbourhood exploration scheme which propagates trajectories from the local model into the immediate voxel neighbourhood and examines their coherence with forthcoming structure, we can directly address one of the key ambiguities of antipodally symmetric fODFs: fanning polarity.
Experiments on simulated data show that by addressing the ambiguities of utilis-ing fODFs capturutilis-ing dispersion, this enables us to exploit the full range of trajectories
Figure 5.11: ROIs used for the in vivo experiments. ROIs are manually outlined to cover the whole of the corticospinal tract and the pons in the region of the cerebral peduncle, and a location in the white matter in the superior portion of the pre-central gyrus.
From cerebral peduncle From superior precentral gyrus
NODDI-NIT
ND-track
DT-PICo
MRtrix
Figure 5.12: Tractography from a seed region in the cerebral peduncle and a seed re-gion in the superior portion of the pre-central gyrus with four different algorithms. The red arrows show the anterior connections which are recovered by NODDI-NIT, but are missed or under-represented by alternative algorithms. The yellow arrows point to the region where false positives occur if neighbourhood exploration is not used.
From cerebral peduncle From superior precentral gyrus
Left hemisphere
Right hemisphere
Figure 5.13: Tracking from a seed region in the cerebral peduncle and a seed region in the superior portion of the pre-central gyrus for each hemisphere of one subject.
ambiguities in the fODF. Due to the NODDI model using a single Bingham distribu-tion to model fibre architecture, this presents problems for the NODDI-NIT algorithm in crossing fibre regions, as demonstrated by the results on the crossing phantom shown in Figure 5.9. NODDI-NIT produces false positives in this synthetic dataset, as the neigh-bourhood exploration framework follows the Bingham mean directions, which can align with the 2nd fibre population which crosses the one the algorithm is initially following.
This represents a failure case for the current formulation of the algorithm. Future work to explicitly model multiple fibre populations in the NODDI model would fix this prob-lem. The neighbourhood exploration framework could be extended to take into account multiple Bingham distributions per voxel.
Experiments on in vivo data examined the behaviour of the algorithm in real data when passing though an area known to exhibit significant dispersion with a clear and easily identifiable polarity: the internal capsule and corona radiata, focusing on the con-nections between the cerebral peduncle and the superior, anterior and posterior regions
NODDI-NIT MRtrix
Subject 1
Subject 2
Subject 3
Subject 4
Figure 5.14: Tractography from a seed region in the cerebral peduncle in four subjects, using NODDI-NIT and MRtrix.
NODDI-NIT MRtrix
Subject 1
Subject 2
Subject 3
Subject 4
Figure 5.15: Tractography from a seed region in the superior portion of the pre-central gyrus in four subjects, using NODDI-NIT and MRtrix.
Cingulum ILF IFOF
NODDI-NIT
MRtrix
Subject 1
NODDI-NIT
MRtrix
Subject 2
NODDI-NIT
MRtrix
Subject 3
NODDI-NIT
MRtrix
Subject 4
Figure 5.16: Tractography in 3 major WM fascicles for validation: the cingulum, the inferior longitudinal fasciculus (ILF) and the inferior fronto-occipital fasciculus (IFOF).
of the cortex. In this region, there is clear and significant spreading of fibres as they travel superiorly, while in the opposing direction there is clear convergence. This struc-tural arrangement is obvious in both histological specimens and the gross structure clear in visualisations of directional information derived from DW-MRI, exemplified in Figure 5.10.
The results presented in Figure 5.12 demonstrate that by utilising models dispersion in tractography, we recover the full range of connectivity throughout the corona radiata, both with and without neighbourhood exploration, while tracking from a seed region located in the cerebral peduncle. A significantly greater degree of streamlines project towards the anterior portions of the cortex, these connections are far less strongly rep-resented in the MRtrix result and not captured at all by DT-PICo. We hypothesize that these may be fronto-pontine fibres. Figures 5.13, 5.14 and 5.15 further demonstrate that this is a repeatable result across multiple subjects.
The advantage of utilising neighbourhood exploration becomes clear when tracking in the direction opposing dispersion, from a superior region of the cortex. Utilising dis-persion models without neighbourhood exploration produces false positive results, with spreading both posterior and anterior which is in conflict with known anatomy. By utilising the neighbourhood exploration scheme, these false positives are eliminated, and the tracks are directed strongly downwards towards the cerebral peduncle, as ex-pected. NODDI-NIT gives a strong connection between the superior pre-central gyrus and peduncle. The connection is weaker for MRtrix tractography, suggesting that MR-trix streamlines spread significantly in this direction.