Abstract
Materials and methods
Why? Accurately quantifying atrophy in brain. How?
Step 1: Whole Brain Segmentation on baseline
Rigid: Translation and rotation Freesurfer
Step 2: Multi-scale Nonlinear Deformation from Baseline to Follow-up
Non-Rigid: Cubic B-Splines Similarity Measure: Normalized Mutual Information (NMI) B-spline grid
Step 3: Estimate the volume change per voxel from the deformation field
And? We apply these methods on publically available subjects from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) data and present statistics on several ROI's
Alzhiemer’s: Alzheimer’s disease is an irreversible, progressive brain disease that slowly destroys memory and thinking skills. Data: Baseline and 12 month follow-up 1.5 T T1-weighted MRI volumes of a subset of 101 ADNI database were analyzed: 24 normal controls (NC), 29 MCI, and 48 AD.
Evaluation points: Every 3rd voxel, every voxel in ROI in the last level. Regularization: A simple gradient of motion in neighborhood control points is used. S(ϕ)=|∇ϕ|2 Why Regularize?
Cubic B-spline formulation
ill-posed problem, unique solution Without Regularization
Atrophy computations: Each vertex of a cubic voxel is pushed through the transformation and the volume of the deformed triangulated cube is computed.
With Regularization, lambda = 0.1
In 3 dimensions, the computation take the form, AD disease
Normal PET
AD PET
!" !" !" !!! !, !. ! = ! . ×! !" !" !"
Bias Estimates
Efficacy tests Comparison of registration based and freesurfer static pipeline based methods in separating the cognitive groups using non-parametric testing (Wilcoxon’s ranksum). Hippocampus Registration based NC=0.2±2.06 MC1=1.3±2.11 AD=2.9±2.9 MCI>No=0.069 AD>MCI=0.0095 AD>No=0.000114 Freesurfer NC=1.00±7.04 MC1=0.9±5.6 AD=4.7±7.3 MCI>No=0.356 AD>MCI=0.018259 AD>No=0.003552
Transformation of each voxel cube
Swap Analysis Hippocampus
Ventricles
Ventricles NC=-4.6±4.0 MCI=-6.7±13.04 AD=10.9±6.6 MCI>No=0.45783 AD>MCI=0.06949 AD>No=0.000101 NC=-4.2±5.2 MCI=-8.5±15.8 AD=13.2±6.7 MCI>No=0.44103 AD>MCI=0.07158 AD>No=0.00015
References [1] D. Rueckert, L. I. Sonoda, C. Hayes, D. L. G. Hill, M. O. Leach, and D. J. Hawkes, “Nonrigid registration using freeform deformations: Application to breast MR images,” IEEE Trans. Med. Imag., vol. 18, pp. 712–721, Aug. 1999. [2] Fischl, B., Salat, D.H., Busa, E., Albert, M., Dieterich, M., Haselgrove, C., van der Kouwe, A., Killiany, R., Kennedy, D., Klaveness, S., Montillo, A., Makris, N., Rosen, B., Dale, A.M., 2002. Whole brain segmentation: automated labeling of neuroanatomical structures in the human brain. Neuron 33 (January), 341–355.
R>0.96
R>0.99
Conclusions: The algorithm developed has been able to successfully separate the control and AD groups. The algorithm is fairly consistent in backward registration.