Skip to main content
. Author manuscript; available in PMC: 2018 Sep 1.
Published in final edited form as: Neuroimage. 2017 Jul 11;158:378–396. doi: 10.1016/j.neuroimage.2017.07.008

Table 5.

Deformation ranges and mean+standard deviation of the deformation errors between the prediction models (LP, LPC, LPP, LPC2, LPC3) and the optimization model (LO) for the image-to-image registration case.

Dataset: LPBA40
Voxel% 0%-10% 10%-20% 20%-30% 30%-40% 40%-50% 50%-60% 60%-70% 70%-80%
# of Test Cases 1560(100%) 1560(100%) 1560(100%) 1560(100%) 1560(100%) 1560(100%) 1560(100%) 1560(100%)
Imain 40 40 40 40 40 40 40 40
Deform (mm) 0.001-0.999 0.999-1.400 1.400-1.751 1.751-2.093 2.093-2.450 2.450-2.840 2.840-3.295 3.295-3.873
LP 0.478 ± 0.323 0.564 ± 0.361 0.623 ± 0.393 0.675 ± 0.425 0.728 ± 0.461 0.786 ± 0.503 0.853 ± 0.556 0.941 ± 0.630
LPC 0.415 ± 0.287 0.469 ± 0.327 0.509 ± 0.359 0.546 ± 0.390 0.584 ± 0.424 0.626 ± 0.464 0.676 ± 0.514 0.742 ± 0.583
LPP 0.543 ± 0.554 0.605 ± 0.613 0.650 ± 0.684 0.691 ± 0.823 0.733 ± 0.783 0.778 ± 0.852 0.830 ± 0.922 0.898 ± 1.055
LPC2 0.510 ± 0.339 0.551 ± 0.378 0.582 ± 0.409 0.611 ± 0.439 0.642 ± 0.471 0.676 ± 0.508 0.716 ± 0.553 0.770 ± 0.617
LPC3 0.637 ± 0.410 0.674 ± 0.451 0.703 ± 0.486 0.732 ± 0.516 0.762 ± 0.550 0.795 ± 0.588 0.834 ± 0.638 0.886 ± 0.713
Voxel% 80%-90% 90%-99% 99%-100% 99.9%-100% 99.99%-100% 99.999%-100% 99.9999%-100% 99.99999%-100%
# of Test Cases 1560(100%) 1560(100%) 1560(100%) 1474(94.5%) 417(26.7%) 72(4.6%) 30(1.9%) 8(0.5%)
Imain 40 40 40 38 31 8 2 1
Deform (mm) 3.873-4.757 4.757-7.317 7.317-23.393 9.866-23.393 12.435-23.393 14.734-23.393 16.835-23.393 19.090-23.393
LP 1.079 ± 0.752 1.418 ± 1.056 2.579 ± 1.869 4.395 ± 2.667 6.863 ± 3.657 9.220 ± 4.829 11.568 ± 6.340 14.475 ± 7.145
LPC 0.847 ± 0.696 1.101 ± 0.976 1.961 ± 1.761 3.431 ± 2.656 5.855 ± 3.711 8.408 ± 4.780 10.484 ± 6.148 14.041 ± 6.879
LPP 1.001 ± 1.396 1.229 ± 1.558 1.760 ± 2.965 2.514 ± 5.845 4.127 ± 7.988 7.097 ± 3.566 10.509 ± 5.093 11.436 ± 5.688
LPC2 0.856 ± 0.721 1.064 ± 0.979 1.765 ± 1.726 3.012 ± 2.644 5.351 ± 3.698 8.362 ± 4.765 11.310 ± 6.117 15.957 ± 6.458
LPC3 0.968 ± 0.835 1.166 ± 1.176 1.829 ± 2.880 3.019 ± 4.688 5.680 ± 10.450 11.434 ± 23.727 18.939 ± 33.953 24.152 ± 38.466
Dataset: IBSR18
Voxel% 0%-10% 10%-20% 20%-30% 30%-40% 40%-50% 50%-60% 60%-70% 70%-80%
# of Test Cases 306(100%) 306(100%) 306(100%) 306(100%) 306(100%) 306(100%) 306(100%) 306(100%)
Imain 18 18 18 18 18 18 18 18
Deform (mm) 0.003-1.221 1.221-1.691 1.691-2.101 2.101-2.501 2.501-2.915 2.915-3.370 3.370-3.901 3.901-4.580
LP 0.573 ± 0.401 0.670 ± 0.448 0.742 ± 0.489 0.810 ± 0.532 0.879 ± 0.580 0.957 ± 0.637 1.053 ± 0.711 1.182 ± 0.815
LPC 0.450 ± 0.343 0.521 ± 0.393 0.577 ± 0.435 0.631 ± 0.479 0.687 ± 0.526 0.751 ± 0.582 0.829 ± 0.655 0.936 ± 0.756
LPP 0.624 ± 0.553 0.698 ± 0.623 0.755 ± 0.678 0.809 ± 0.729 0.863 ± 0.782 0.923 ± 0.843 0.996 ± 0.918 1.094 ± 1.022
LPC2 0.505 ± 0.383 0.566 ± 0.437 0.614 ± 0.480 0.660 ± 0.522 0.707 ± 0.568 0.761 ± 0.622 0.827 ± 0.691 0.917 ± 0.788
LPC3 0.606 ± 0.446 0.666 ± 0.504 0.713 ± 0.549 0.757 ± 0.594 0.803 ± 0.640 0.854 ± 0.694 0.915 ± 0.763 1.001 ± 0.860
Voxel% 80%-90% 90%-99% 99%-100% 99.9%-100% 99.99%-100% 99.999%-100% 99.9999%-100% 99.99999%-100%
# of Test Cases 306(100%) 306(100%) 306(100%) 125(40.8%) 46(15.0%) 12(3.9%) 3(1.0%) 3(1.0%)
Imain 18 18 18 10 3 2 1 1
Deform (mm) 4.580-5.629 5.629-9.026 9.026-36.263 14.306-36.263 19.527-36.263 23.725-36.263 27.533-36.263 29.154-36.263
LP 1.397 ± 0.988 2.007 ± 1.451 5.343 ± 3.868 13.103 ± 4.171 20.302 ± 2.287 24.666 ± 1.688 28.081 ± 1.190 30.436 ± 1.807
LPC 1.114 ± 0.924 1.609 ± 1.367 4.485 ± 3.862 11.928 ± 4.800 19.939 ± 2.549 24.457 ± 1.877 27.983 ± 1.218 30.000 ± 1.796
LPP 1.249 ± 1.184 1.621 ± 1.572 2.894 ± 2.798 5.054 ± 4.387 9.631 ± 6.406 13.541 ± 7.744 17.600 ± 7.317 15.553 ± 6.339
LPC2 1.068 ± 0.951 1.488 ± 1.375 3.917 ± 3.730 10.437 ± 5.395 19.345 ± 3.061 24.154 ± 2.173 27.834 ± 1.403 29.654 ± 1.909
LPC3 1.141 ± 1.022 1.521 ± 1.432 3.596 ± 3.555 8.962 ± 5.690 18.252 ± 4.077 23.673 ± 2.583 27.681 ± 1.641 29.460 ± 2.227
Dataset: CUMC12
Voxel% 0%-10% 10%-20% 20%-30% 30%-40% 40%-50% 50%-60% 60%-70% 70%-80%
# of Test Cases 132(100%) 132(100%) 132(100%) 132(100%) 132(100%) 132(100%) 132(100%) 132(100%)
Imain 12 12 12 12 12 12 12 12
Deform (mm) 0.004-1.169 1.169-1.602 1.602-1.977 1.977-2.341 2.341-2.717 2.717-3.126 3.126-3.597 3.597-4.189
LP 0.617 ± 0.433 0.709 ± 0.480 0.784 ± 0.524 0.856 ± 0.570 0.929 ± 0.621 1.010 ± 0.680 1.109 ± 0.758 1.239 ± 0.868
LPC 0.525 ± 0.391 0.587 ± 0.441 0.640 ± 0.486 0.694 ± 0.534 0.750 ± 0.586 0.813 ± 0.646 0.890 ± 0.724 0.995 ± 0.834
LPP 0.653 ± 0.538 0.717 ± 0.607 0.772 ± 0.667 0.829 ± 0.727 0.888 ± 0.789 0.953 ± 0.859 1.032 ± 0.948 1.138 ± 1.068
LPC2 0.605 ± 0.444 0.653 ± 0.496 0.696 ± 0.543 0.739 ± 0.591 0.787 ± 0.645 0.839 ± 0.704 0.905 ± 0.782 0.996 ± 0.891
LPC3 0.730 ± 0.519 0.775 ± 0.575 0.815 ± 0.625 0.857 ± 0.675 0.903 ± 0.732 0.954 ± 0.793 1.019 ± 0.872 1.107 ± 0.984
Voxel% 80%-90% 90%-99% 99%-100% 99.9%-100% 99.99%-100% 99.999%-100% 99.9999%-100% 99.99999%-100%
# of Test Cases 132(100%) 132(100%) 132(100%) 132(100%) 75(56.8%) 18(13.6%) 1(0.8%) 1(0.8%)
Imain 12 12 12 12 9 6 1 1
Deform (mm) 4.189-5.070 5.070-7.443 7.443-18.753 9.581-18.753 12.115-18.753 14.383-18.753 16.651-18.753 18.297-18.753
LP 1.448 ± 1.050 1.955 ± 1.490 3.340 ± 2.412 4.882 ± 3.106 7.281 ± 3.378 9.978 ± 3.211 14.113 ± 1.211 15.868 ± 0.315
LPC 1.163 ± 1.017 1.571 ± 1.451 2.676 ± 2.386 3.930 ± 3.125 5.976 ± 3.646 8.527 ± 3.889 13.009 ± 1.584 14.908 ± 0.437
LPP 1.305 ± 1.258 1.683 ± 1.686 2.484 ± 2.503 3.047 ± 3.048 3.511 ± 3.287 2.999 ± 3.150 1.196 ± 0.327 1.277 ± 0.178
LPC2 1.142 ± 1.072 1.494 ± 1.494 2.405 ± 2.388 3.392 ± 3.055 5.003 ± 3.702 7.176 ± 4.218 11.155 ± 2.470 13.456 ± 0.812
LPC3 1.248 ± 1.165 1.581 ± 1.581 2.383 ± 2.434 3.165 ± 3.015 4.422 ± 3.629 6.283 ± 4.370 9.607 ± 3.120 11.757 ± 1.492
Dataset: MGH10
Voxel% 0%-10% 10%-20% 20%-30% 30%-40% 40%-50% 50%-60% 60%-70% 70%-80%
# of Test Cases 90(100%) 90(100%) 90(100%) 90(100%) 90(100%) 90(100%) 90(100%) 90(100%)
Imain 10 10 10 10 10 10 10 10
Deform (mm) 0.003-1.122 1.122-1.553 1.553-1.929 1.929-2.294 2.294-2.674 2.674-3.089 3.089-3.567 3.567-4.163
LP 0.578 ± 0.422 0.680 ± 0.471 0.757 ± 0.514 0.829 ± 0.559 0.904 ± 0.610 0.986 ± 0.671 1.082 ± 0.748 1.207 ± 0.858
LPC 0.486 ± 0.382 0.558 ± 0.436 0.615 ± 0.481 0.669 ± 0.528 0.726 ± 0.580 0.790 ± 0.640 0.865 ± 0.715 0.963 ± 0.820
LPP 0.624 ± 0.556 0.701 ± 0.624 0.761 ± 0.680 0.819 ± 0.737 0.881 ± 0.796 0.948 ± 0.865 1.026 ± 0.948 1.127 ± 1.061
LPC2 0.562 ± 0.436 0.623 ± 0.493 0.670 ± 0.541 0.716 ± 0.589 0.766 ± 0.643 0.821 ± 0.703 0.886 ± 0.778 0.971 ± 0.881
LPC3 0.684 ± 0.512 0.745 ± 0.574 0.792 ± 0.625 0.838 ± 0.677 0.887 ± 0.734 0.942 ± 0.797 1.007 ± 0.874 1.091 ± 0.979
Voxel% 80%-90% 90%-99% 99%-100% 99.9%-100% 99.99%-100% 99.999%-100% 99.9999%-100% 99.99999%-100%
# of Test Cases 90(100%) 90(100%) 90(100%) 89(98.9%) 37(41.1%) 7(7.8%) 3(3.3%) 2(2.2%)
Imain 10 10 10 9 6 3 1 1
Deform (mm) 4.163-5.047 5.047-7.462 7.462-18.727 9.833-18.727 12.607-18.727 15.564-18.727 17.684-18.727 18.534-18.727
LP 1.408 ± 1.041 1.924 ± 1.499 3.348 ± 2.341 4.873 ± 2.704 7.299 ± 3.256 10.503 ± 4.049 11.764 ± 3.005 13.041 ± 1.691
LPC 1.120 ± 0.998 1.526 ± 1.444 2.627 ± 2.275 3.674 ± 2.621 5.283 ± 3.272 8.492 ± 4.088 9.336 ± 3.738 10.499 ± 2.692
LPP 1.289 ± 1.250 1.676 ± 1.705 2.430 ± 2.415 2.707 ± 2.403 2.998 ± 2.214 2.695 ± 1.419 2.377 ± 1.063 2.226 ± 0.365
LPC2 1.109 ± 1.055 1.455 ± 1.482 2.361 ± 2.270 3.138 ± 2.585 4.267 ± 3.133 6.603 ± 3.760 7.444 ± 3.657 8.415 ± 3.053
LPC3 1.225 1.153 17551 17570— 2.358 ± 2.306 3.023 ± 2.588 3.755 ± 2.957 5.214 ± 3.486 6.240 ± 3.357 7.569 ± 3.283

All measures are evaluated within the brain mask only. All deformation values and deformation errors are evaluated in millimeters (mm). Voxel%: percentile range of voxels that fall in a particular deformation range based on the optimization model (LO). # of Test cases: The number of registration cases that contain voxels within a given percentile range. Imain: Minimum number of distinct images required to cover all registration test cases in a particular deformation range either as the moving or the target image. This measure is meant to quantify the influence of few images on very large deformations. For example, for the four registration cases A-B, B-C, B-D and E-A, Imain = 2 as it is sufficient to select images A and B to cover all four registrations. The results show that a comparatively small subset of images is responsible for most of the very large deformations. Of course, all images of a particular dataset are involved in the small deformation ranges. Deform: range of deformations within a given percentile range. The cells with the lowest mean deformation errors for every deformation range are highlighted. Best-viewed in color.