Table 1.
The ICE-T Framework as Pseudo-Code.
Figure 1.
The impact of the ICE-Tstreams parameter.
Curves show the parameter's effect upon the size of ICE-T_ROII for various choices of the ICE-Tthreshold parameter and seed regions (detailed in the legend). Seed regions are labelled as MC (motor cortex), SC (somatosensory cortex) or PFC (prefrontal cortex).
Figure 2.
The impact of the ICE-Tthreshold parameter.
Curves show the parameter's effect upon the size of ICE-T_ROII with the ICE-Tstreams parameter fixed at 20 streamlines, shown for each of the three ROIs (MC, PFC and SC) in each of the three ex-vivo datasets (P1, P2 and P3).
Figure 3.
The impact of the number of ICE-T iterations performed.
Curves show the effect of iteration number upon the size of ICE-T_ROII for dataset P1, using the MC seed region, and for various values of the ICE-Tthreshold parameter. For clarity, a log scale has been employed for the vertical axis.
Figure 4.
Illustration of the ability of ICE-T to penetrate through a complex region.
The figure shows how ICE-T successfully propagates through a known crossing-fibre region (centrum semi-ovale, light blue dotted circle, upper left panel) when seeded from the SC region (green region, upper left panel) of dataset P1. The dark blue region (upper left panel) shows the results using ICE-Tthreshold of 0.02, and the red region (upper left panel) for ICE-Tthreshold of 0.015. The graph (lower panel) shows the spatial extent of the ICE-T_ROII, sampled along the canonical streamline, from the seed region (defined as Distance = 0) as a function of both distance from the seed region, and of the value of the ICE-Tthreshold parameter. Here a coloured voxel represents that the segmented ROI was present at the given threshold and distance from the seed. Each threshold level is coloured differently for clarity. Once the ICE-Tthreshold parameter falls to 0.015 and below (lower three rows), the region-growing penetrates past the complexity and continues on to extract the distal portion of the tract. The 3D rendering (upper right panel) shows the ICE-T results at the same two thresholds (0.02 in blue and 0.015 in orange). The seed region is located at the site of the green arrow (upper right panel).
Figure 5.
Comparison of tractography with and without ICE-T.
Tractography is seeded from both the MC & PFC seeds (shown in green) of dataset P1.(Left Panel) Tractography without ICE-T (i.e. directly with the seed ROI) using N = 64000 streamlines per voxel and then visualised using the following thresholds (from top) = 0.050, 0.020, 0.010, 0.005, 0.002.(Right Panel) Tractography with ICE-T ROII (generated using ICE-Tstreams = 20, ICE-Tthreshold = 0.01) used as seed, and then visualised at the following thresholds (from top) = 0.015, 0.012, 0.010, 0.005, 0.002.Green arrows indicate areas demonstrating “near-seed flare”. Red arrows indicate premature termination of the tract ROI due to the PLD effect causing the PICo values to fall below the applied threshold.
Figure 6.
Tractography with ICE-T from each of the three ROIs (MC, PFC and SC), for each pig brain (P1, P2, P3).
Parameters: ICE-Tstreams = 20, ICE-Tthreshold = 0.005, results rendered at 0.005. Data show the glass brain of the unweighted diffusion image as anatomical reference.
Figure 7.
Comparison of along-tract profiles after tractography, with and without ICE-T, versus linear compensation.
Curves show the variation of PICo values (scaled to [0,1]) along the canonical streamline from the MC seed region (red dashed vertical line) on dataset P2 for tractography with ICE-T (green), without ICE-T (blue), and tractography with linear compensation (orange). Right panel shows a zoomed portion of the main graph, delineated by the purple dashed border, where both the tractography without ICE-T (blue) and tractography with linear compensation (orange) are now drawn according to the scale on the right axis. Dataset details: ICE-T performed using ICE-Tstreams = 20, ICE-Tthreshold = 0.005, number of ICE-T iterations until stability = 41. Tractography without ICE-T was generated using 5000 streams per voxel. A drastic fall-off in PICo values can be observed (green arrow) for tractography results without ICE-T soon after the streamlines exit the seed region due to their encountering a complex region. No such effect is seen for the results with ICE-T. For a tract reference, see the 3D render of this tract in Figure 6, P2, MC seed region.
Figure 8.
Comparison of thresholding tractography results, obtained with and without ICE-T, in a human in vivo subject.
Both results are generated from a cubic seed (dark green) placed approximately in the left MC region. Tractography without ICE-T used the original cubic seed ROI as the seed (25,000 streamlines, blue, top row). Tractography with ICE-T used the ICE-T ROII as seed (ICE-Tthreshold 0.01, ICE-Tstreams 20, purple, bottom row), shown here at various rendering thresholds (0.02, 0.01, 0.005, 0.001).The path-length dependency is very pronounced in the tractography results without ICE-T (top row), evidenced by the movement of the end-of-tract point (green arrows) as a function of the applied threshold. Probable false-positives are seen in tractography both with and without ICE-T around the descending portion of the contralateral CST (red arrows). These can be addressed in the conventional manner by the introduction of exclusion masks (dark green box and plane) that terminate and remove any streamlines that propagate through them. Here two are shown for both methods (last column) - one along the mid-sagittal plane and one in the contralateral CST. The former is to prevent streamlines crossing between the hemispheres at the cortical level dorsal to the corpus callosum due to the high partial volume effect. The latter is to prevent segmentation of a known false-positive branch of the contralateral CST.
Figure 9.
Tractography in a human in-vivo dataset, with and without ICE-T, showing dependency upon the size of the seed ROI.
Top row ((a), (b), (c)) shows the same results as for Figure 8, but from a posterior viewpoint. From this angle it is also clear how the tractography without ICE-T using the cubic seed also generates a lateral cortical branch ((a), (b): green arrow). Inset on (a) shows lateral view from the right side, highlighting the posteriorly-directed angle of the branch.Bottom row ((d), (e), (f)): tractography results from a single voxel seed within the left MC, using the same parameters as for the cubic seed. As for the cubic seed, the rendering thresholds have been selected so as to generate comparable propagation of the tractography into the contralateral ascending portion of the CST. In the tractography results without ICE-T ((a), (d)), the ipsilateral descending portion follows a more medial route than the results using ICE-T ((c), (f)), as can be seen on the merged views ((b), (e)). Further inspection of these results indicates that the streamlines diverge from the CST around the level of the ventricles and seem to instead pick-up a periventricular route through the medial thalamic nuclei ((d), (e): yellow arrows). The streams then diverge, following a descending route close to the CST ((d), (e): orange arrows), and a medial route along the anterior thalamic radiation ((d), (e): red arrows). The ICE-T results correctly follow the CST from both seed areas.