Out of all 6240 AFIDs placed, 21 were deemed outliers using a threshold AFLE of greater than 10 mm (0.33%). None of the outliers were placed by expert raters; therefore, 0.55% fiducials placed by novice raters were outliers. All outliers involved placements at some components of the lateral ventricles, and were as follows (number of outliers for this structure in brackets): right lateral ventricle at AC (5), left lateral ventricle at AC (6), right lateral ventricle at PC (2), right anterolateral temporal horn (1), right superior anteromedial horn (1), left superior anteromedial horn (1), right inferior anteromedial horn (2), left inferior anteromedial horn (2) and right ventral occipital horn (1).
Analysis in subject space
Online Resource 2 depicts mean distance from the mid-commissural point by each rater for the 32 AFIDs. The mean overall AFLE across all AFIDs was 1.57 mm ± 1.16 mm. The mean AFLE across all raters for each AFID can be seen in Fig. 2. AFID 25 and 26 (left and right lateral ventricle at AC, respectively) had the highest AFLE at 2.63 mm ± 1.75 mm and 2.79 mm ± 1.95 mm, respectively. The AFIDs with the lowest overall AFLE were AFID 01–02 (anterior commissure and posterior commissure respectively), with AFLEs of 0.70 mm ± 0.78 mm and 0.55 mm ± 0.34 mm, respectively.
Table 1 represents the mean AFLE obtained by expert and novice raters. Expert raters overall had a lower mean AFLE (1.33 mm ± 0.79 mm), compared to novice raters (1.73 mm ± 1.30 mm). Wilcoxon rank-sum tests for AFLE between expert and novice raters with Bonferroni correction for multiple comparisons are shown in Table 1. Expert raters had a lower AFLE in 29 of the 32 AFIDs. 6 AFIDs had significantly different AFLEs between raters, 5 of which were higher in the novice raters. The superior interpeduncular fossa (AFID05), however, had a greater AFLE obtained by expert raters compared to novice raters.
To illustrate the differences in AFLE obtained across all 39 subjects and 32 AFIDs, the mean AFLE across the 5 raters was obtained. This produced a 39 by 32 matrix which is represented as a colormap in Fig. 2. Each cell in the matrix represents the mean AFLE across the 5 raters for that subject and AFID. This figure illustrates the distribution of errors across the 39 subjects. Some fiducials with a high AFLE, such as AFIDs 25 and 26, demonstrate a consistently higher error across most subjects. However, other fiducials such as AFIDs 15 and 16 only demonstrate a higher AFLE in a subset of subjects.
Intraclass correlation coefficient (ICC) was calculated for each AFID between all raters, expert raters and novice raters, summarized in Table 2. The mean ICC across all AFIDs was 0.814 between all raters, 0.912 between expert raters and 0.777 between novice raters. The superior interpeduncular fossa (AFID 05) had the lowest ICC among both expert and novice raters (0.708 and 0.544, respectively). Otherwise, novice raters also had a lower inter-rater agreement when placing AFIDs associated with the temporal horns (AFIDs 23–26). The left anteromedial temporal horn (AFID26) had the second lowest ICC calculated at 0.567 between novice raters, but had an ICC of 0.963 between expert raters.
Analysis in MNI space
To demonstrate the use of AFIDs in determining registration error, subject scans were linearly and non-linearly transformed to the MNI152NLin2009cAsym brain template. The mean real-world and consensus AFREs were calculated. Linear and non-linear real-world AFREs for each AFID are presented in Online Resource 3. The mean non-linear real-world AFRE is 3.34 mm ± 1.94 mm, and the linear AFRE is 4.15 mm ± 2.03 mm. Wilcoxon rank-sum tests for real-world AFREs between linear and non-linear registration with Bonferroni correction for multiple comparisons are shown in Online Resource 3. 15 of the 32 AFIDs had a significantly greater AFRE when using linear registration compared to non-linear registration. Additionally, the consensus AFRE is presented in Online Resource 3, with a mean consensus AFRE of 2.82 mm ± 2.01 mm. 6 AFIDs had a significantly higher non-linear real-world AFRE compared to consensus AFRE (AFIDs 1, 5, 8, 9, 31 and 32).
Figure 3 demonstrates the mean non-linear real-world AFRE across all subjects and raters for each AFIDs. The anterior commissure (AFID01) had the smallest AFRE calculated at 1.11 mm ± 1.06 mm. The right and left ventral occipital horns (AFIDs 29 and 30) had the largest AFRE at 6.81 mm ± 2.94 mm and 7.36 mm ± 3.41 mm, respectively. A colormap of non-linear AFRE across the 5 raters for each subject and AFID is illustrated in Fig. 3. This figure demonstrates that the AFIDs with the smallest registration error (AFIDs 1, 2, 11, 12, 13, 31 and 32) were robustly decreased across most subjects. Alternatively, AFIDs 29 and 30 had large registration errors across multiple subjects.
Distance between AFIDs as a biomarker of disease
An unpaired two-tailed t test was performed to compare age across the two groups, demonstrating no statistical difference in age (p = 0.48). Additionally, a chi-square test demonstrated no difference in sex distribution among the two groups (χ2 (1, N = 69 = 3.76, p = 0.053).
496 unique Euclidean pairwise distances were calculated between AFIDs for our PD patients (n = 39) and OASIS-1 subjects (n = 30). Figure 4 represents the differences between the mean of each pairwise distance, calculated by subtracting the mean distances in the OASIS-1 dataset from the mean distances in the PD subject dataset (therefore, a positive value indicates a greater pairwise distance in the OASIS-1 subjects). Wilcoxon rank-sum tests were used, and statistically significant differences are indicated in Fig. 4. Significance was determined after Bonferroni correction (i.e. by obtaining a p value < 0.05/496).
Between our PD and OASIS-1 datasets, 40 pairwise distances were statistically significantly different (Online Resource 4). The largest distances as a percentage of the distance in the PD dataset were in the left anterolateral to left inferior anteromedial temporal horn (AFID 22 and 26; 2.82 mm, 29.6%), the pontomesencephalic junction to the superior interpeduncular fossa (AFIDs 4 and 5; 2.47 mm, 26.5%) and the infracollicular sulcus to the pineal gland (AFIDs 3 and 14; 1.90 mm, 18.7%).