Figure 1.
The focus of this study is on the non-linear registration step which is investigated using four different approaches. Registration steps are described in Section.
Figure 2.
Most-Representative-Subject TBSS (RS-TBSS) and Study-Specific-Template TBSS (SS-TBS) pipeline.
The remainder of this paper is organised as follows: In Section, different registration approaches for the TBSS pipeline are reviewed and a modification to the pipeline is introduced to incorporate a group-wise atlas. In Section, a misalignment between two groups (patients and controls) is modelled using a simulation study. In Section, results are presented on the simulation study and on a dataset of AD (n = 20) and age-matched controls (n = 21).
Table 1.
Anatomical locations reported to show reduced FA in AD patients in the literature using TBSS to date.
Table 2.
Demographic and clinical data for the Alzheimer's disease (AD) patients and healthy control subjects whose scans were used in this study.
Figure 3.
Flowchart of the simulation study.
Flowchart of creating the misalignment used in the specificity evaluation. Ten control images (CON) images were individually warped to 10 AD images to produce ten warped control images using ITK-based affine registration method and Demons.
Figure 4.
Modelling typical atrophy in AD using Demons registration algorithm.
A control image, CON, an AD subject as the target and the deformed control image, , after applying the registration. Ventricular expansion in AD is well modelled in the control subject using the Demons algorithm.
Figure 5.
WM tract masks used in the true-positive experiment.
CB: Cingulum bundle; ILF: Inferior Longitudinal fasciculus (including the Inferior fronto-occipital fasciculus); SLF: Superior longitudinal fasciculus; UF: Uncinate fasciculus; PTR: Posterior thalamic radiation.
Table 3.
Summary of the results obtained with different TBSS pipelines in the literature and specificity evaluation study on FA.
Figure 6.
TBSS contrasts between two control groups (CON and) using different registration schemes.
The contrasts are overlaid on the mean FA map of each approach and the mean FA skeleton (in green, FA 0.2). The results are thresholded at , corrected for multiple comparisons. The yellow-red color indicate the areas with significantly decreased FA values in deformed control images compared with the original controls.
Table 4.
Results obtained with Group-wise TBSS on sensitivity evaluation study when reducing FA virtually.
Figure 7.
Bland-Altman plot showing differences in projected FA between GW-TBSS and the established methods of registration.
GW-TBSS has a higher projected FA across the mean skeleton compared to ST-TBSS, RS-TBSS and SS-TBSS. Median difference in FA are shown with horizontal lines for each comparison.
Figure 8.
Standard deviation in FA across the group after registration to the FA template (FMRIB58_FA) in ST-TBSS, RS-TBSS, SS-TBSS, GW-TBSS.
Standard deviation maps indicate standard deviation was greater when using ST-TBSS and RS-TBSS. Colour bar indicates standard deviation.
Figure 9.
Mean variance of difference between average image and subjects in each iteration when using GW-TBSS.
The mean and standard deviation of variance reduces in each iteration, r: rigid registration; a: affine registration; n: non-linear registration. Inlays of the linear (top right) and non-linear (bottom-right) iterations are shown separately to better illustrate the improvement of the group wise registration with each iteration.
Figure 10.
The voxel-wise statistical map between 20 patients with AD and 21 controls using different TBSS approaches.
FA results showing the contrast ADCON; Statistical threshold: p0.05 (corrected); In green the mean FA skeleton is shown and the statistical maps are overlaid on the mean FA image of each approach.