intrapatient; rigid; localized mutual information
Figure 1: T1-weighted images acquired at a range of inversion times.
First, registration of the T1-weighted images was performed to improve the T1 map estimation. The scans were rigidly registered in 3D to the T1-weighted scan with the highest contrast (FSPGR(TI=2100)).
Second, registration of the FSPGR(TI=2100) scans was performed to align scan and rescan. Based on this registration, rescan T1 maps were transformed to the scan maps.
Both registration steps were performed with the same registration settings. Femoral and tibial regions were registered separately based on subvolumes containing only the specific bone and surrounding tissue to allow correction for motion of the knee joint (Figure 2). The registration was optimized over 1000 iterations with localized mutual information (LMI) as similarity measure. Per iteration, LMI was calculated using 2048 random samples obtained from a sample region of size 50x50x50 mm. Cubic B-spline interpolation was used when applying the deformation to the moving image.
Figure 2: Subvolumes for registration on femur and tibia separately.
For parameter files see the Elastix Model Zoo repository on GitHub.
elastix
version: 4.400
Command line calls:
For the registration of the T1-weighted images:
elastix -f scan_2100.nii -m scan_TI.nii -out outdir -p params.txt -fMask subvolume_scan.nii -mMask subvolume_scan.nii
For the registration of the scan and rescan FSPGRTI=2100:
elastix -f scan_2100.nii -m rescan_2100.nii -out outdir -p params.txt -fMask subvolume_scan.nii -mMask subvolume_rescan.nii
These registrations are described in the publication:
E.E. Bron, J. van Tiel, H. Smit, D.H.J. Poot, W.J. Niessen, G. P. Krestin, H. Weinans, E. Oei, G. Kotek and S. Klein, Image registration improves human knee cartilage T1 mapping with delayed gadolinium-enhanced MRI of cartilage (dGEMRIC), European Radiology, in press
© 2020 Viktor van der Valk