Registration motivation
Mads Nielsen
270312 Mads Nielsen
Statistical Network
2Registration in Medical Imaging
Find correspondences – intrapatient:
inhale phase to exhale phase
Castillo, R., Castillo, E., Guerra, R., Johnson, V.E., McPhail, T., Garg, A.K., Guerrero, T. 2009 “A framework for
evaluation of deformable image registration spatial accuracy using large landmark point sets” Phys Med Biol 54 1849-1870.
1
1
270312 Mads Nielsen Statistical Network
Monitoring subtle changes Local is better!
Problem: which pixel goes where?
Baseline Follow-up
270312 Mads Nielsen Statistical Network
Optimizatio Transfor n
m
F ix e d Im a g e
M o v in g Im a g e
Cost Functio
n
Transform Coefficients
Incorporate model of density change
Image registration
270312 Mads Nielsen Statistical Network
Disease monitoring using image registration
Possible to indicate which locations change Consistent in time
Local changes predict decline better
270312 Mads Nielsen
Statistical Network
6Registration in Medical Imaging
Find correspondences – intrapatient:
Marcus, DS, Fotenos, AF, Csernansky, JG, Morris, JC, Buckner, RL, 2009. Open Access Series of Imaging Studies
(OASIS): Longitudinal MRI Data in Nondemented and Demented Older Adults. Journal of Cognitive Neuroscience, in press.
disease progression
2
2
270312 Mads Nielsen Statistical Network
Readings: Atrophy Assessment
7
Step 1: Whole Brain Segmentation on baseline
Step 2: Un-biased Multi-scale Nonlinear Deformation from Baseline to Follow-up
Step 3: Estimate the volume change per voxel from the deformation field
Step 4: Estimate the structure change by
summing up values in Step 3
270312 Mads Nielsen Statistical Network
Readings: Atrophy Accuracy Test
8
Group Hippocampus baseline volume FreeSurfer (mm3)
Cross- Sectional FreeSurfer (%)
Longitudinal
FreeSurfer (%) Hippocampus baseline volume
Synarc (mm3)
Cross- Sectional Synarc (%)
Registration Based(%)
Normal
(N = 27) 3410 ± 329 (L)
3430 ± 416 (R) -0.9 ± 2.6
0.9 ± 6.6 -0.8 ± 1.4
-0.6 ± 2.4 2333 ± 330 (L)
2268 ± 299 (R) -1.9 ± 2.0 -1.4 ± 2.9
-0.8 ± 2.0 -0.6 ± 1.9
MCI
(N = 37) 2992 ± 281(L)
3070 ± 420(R) -0.9 ± 2.9
-0.5 ± 7.0 -1.2 ± 1.5
-1.4 ± 3.0 2081 ± 312 (L)
2075 ± 363 (R) -2.5 ± 3.1 -3.1 ± 3.3
-1.1 ± 1.2 -2.0 ± 2.0
AD (N = 57)
2542 ± 349(L)
2728 ± 621(R) -4.3 ± 3.1
-4.0 ± 7.1 -4.2 ± 2.4
-4.4 ± 4.1 1787 ± 482 (L)
1773 ± 484 (R) -4.0 ± 4.4 -5.2 ± 4.3
-4.2 ± 1.6 -4.4 ± 2.4
101 Subjects From ADNI. Changes from BLM12
270312 Mads Nielsen
Statistical Network
9Registration at Large and Small Scale
Inter-patient variation at large scale Atrophy at
smaller scale
Data from: Marcus, DS, Fotenos, AF, Csernansky, JG, Morris, JC, Buckner, RL, 2009.
Open Access Series of Imaging Studies (OASIS):
“Longitudinal MRI Data in Nondemented and Demented Older Adults.“
Journal of Cognitive Neuroscience, in press.
disease progression
2
270312 Mads Nielsen
Statistical Network
10LDDMM/LDDKBM registration
Large Deformation Diffeomorphic Metric Mapping
Domain , deformations
Find minimizing
regularization/smoothness term
matching term Images
Ω ⊆ℝ d ϕ∈ G V , ϕ : Ω → Ω ϕ
E (ϕ)= E 1 ( ϕ)+ U (ϕ)
E 1 (ϕ)
U (ϕ)= U (ϕ . I m , I f )
I m I f
270312 Mads Nielsen Statistical Network
in LDDMM, regularization is the length of minimal paths
If Sobolev-norm <Lv,v>
on v, then diffeomorphism L is the momentum
Operator: Lv = a
V t = ∫K(:,x)a t (x)dx
11