Premium
IC‐P‐089: Monitoring the Brain's Longitudinal Changes in Clinical Trials for Alzheimer's Disease: A Robust and Reliable Nonrigid Registration Framework
Author(s) -
Lorenzi Marco,
Pennec Xavier,
Frisoni Giovanni
Publication year - 2011
Publication title -
alzheimer's and dementia
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 6.713
H-Index - 118
eISSN - 1552-5279
pISSN - 1552-5260
DOI - 10.1016/j.jalz.2011.05.055
Subject(s) - segmentation , robustness (evolution) , artificial intelligence , neuroimaging , atrophy , computer science , pattern recognition (psychology) , medicine , psychology , pathology , neuroscience , biology , biochemistry , gene
Background: The reliable detection of brain structural changes in Alzheimer’s disease (AD) is of primary importance in clinical trials. The statistical power of the measurements is potentially affected by the variability introduced by manual tracings of brain regions or the consistency of the segmentation over multiple time points. Differently from the purely segmentation based methods, non-rigid registration allows the quantification of the topographic changes over the brain, which might support evidence of disease-specific treatment effects. However, registration algorithms are usually sensitive to the image biases. We propose an automated, efficient and robust pipeline based on non-rigid registration, to quantify the longitudinal brain changes in patients with AD. Methods: We selected the baseline and 1-year follow-up scans from 100 AD patients and 100 healthy controls from the ADNI dataset. For each subject, the longitudinal changes were detected by registration of the baseline scan to the follow-up with a modified version of the Demons algorithm [Vercauteren, 2008]. The robustness to the intensity bias was achieved by adding to the framework a similarity measure based on a local intensity scaling [Cachier, 2002]. The volume changes were measured as the flux of the deformation field [Lorenzi ,2010] across selected regions, and the results for the whole brain atrophy were validated against the manual tracings and the KNBSI algorithm. Results: The measured percentage whole-brain volume change for the healthy controls and the AD group is illustrated in Table1. The related sample size per arm required from an hypothetical clinical trial to detect a 25% change in the AD progression (80% power) was: 64 (flux), 89 (KNBSI), 62 (manual). Conclusions: The proposed pipeline: provides statistically powered measurements of longitudinal brain atrophy; robustly estimates the changes at different spatial scales and over different regions by local intensity scaling and integration of the flux on probabilistic masks; coupled with specific registration frameworks for time series of images of multiple time points [Lorenzi, 2010]; and provides a consistent measure of the longitudinal changes along both spatial and temporal dimension.