Article Figures & Data
Figures
- Fig 1.
Sample patient with a right temporal lobe tumor. The upper row illustrates the challenge in functional-to-anatomic multimodal alignment between T2* EPI (A) and T1-weighted MPRAGE images (B). Note the inherently poor resolution of EPI. These images also show signal drop-out from susceptibility artifacts near the sinuses and craniotomy (asterisks). The lower row highlights the great difficulty in judging accurate alignment simply by visual inspection. The fusion image of the average EPI mask over the MPRAGE image (C) shows signal drop-out at the skull base (asterisks), which can potentially confound registration algorithms. The fusion image of edge-enhanced EPI (D) delineates structural boundaries in EPI data and reveals slight misalignment (arrows) due to patient motion between MPRAGE and EPI, primarily in the inferior-superior direction along with some rotation.
- Fig 2.
Graph plotting the average transformation parameters for translation and rotation as calculated by several metrics: Hellinger, mutual information, normalized mutual information, correlation ratio, and local Pearson correlation. The transformation effects of LPC are in the opposite direction (asterisks) for most translations and rotations, most notably the inferior translation and rotation in pitch. R-L indicates right-left; A-P, anterior-posterior; and S-I, superior-inferior.
- Fig 3.
Histogram summarizing the objective assessment of coregistration using edge-enhanced images for both HEL and LPC compared with the initial nonregistered condition. In the case of the HEL metric, the accuracy of function-to-structural alignment actually worsens in 59% of cases and improves in 11% of cases. However, in the case of the LPC metric, none of the cases worsened and 41% improved.
- Fig 4.
The upper row displays activation areas from the antonym task superimposed on coronal MPRAGE images (A–C). The high-contrast window-level setting helps demarcate the tumor in the left parietal lobe (asterisks). Lesion-to-activation distance between the tumor and the nearest activation area (large yellow arrows) is quite different: 2.6 mm in NR (A), 12.9 mm in HEL (B), and 7.4 mm in LPC (C). Notice that the activation area in the ventral temporo-occipital junction (green arrowheads) is displaced in the cerebellum in HEL (B). The lower row consists of edge-enhanced EPI superimposed on MPRAGE images (D–F). Visual inspection reveals that edge-enhanced EPI (small blue arrows) is too high in the initial NR (D), too low in HEL (E), but just right in LPC (F).
- Fig 5.
Histogram displaying the maximum differences in LAD among LPC, HEL, and NR. Data are also sorted by imaging plane. Notice that the LAD difference is <3 mm in most cases, and this is especially true in the axial plane. However, a nontrivial number of cases have a >3 mm discrepancy, particularly in the coronal and sagittal planes.
- Fig 6.
Histogram of the maximum differences in LAD between LPC and NR. Data are also sorted by imaging plane. Because edge-enhanced analysis has shown LPC to produce accurate alignment, the contrast between LPC and NR reveals the degree of misalignment when registration is not performed. The most pronounced LAD differences are found in the coronal and sagittal planes.
- Fig 7.
Histogram of the maximum differences in LAD between LPC and HEL. Data are also sorted by imaging plane. Because it has been shown that LPC produces accurate alignment and that HEL systematically introduces misregistration error, the contrast between LPC and HEL serves to illustrate the degree of misalignment that may result if a poor registration metric is chosen. The most pronounced LAD differences are found in the coronal and sagittal planes.
