Premium
IC‐P‐030: High‐throughput and accurate segmentation of lateral ventricles using combined single atlas propagation and tissue classification: Application to ADNI and IBSR data
Author(s) -
Belaroussi Boubakeur,
Bracoud Luc,
Noever Klaus,
Pachai Chahin
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.2011.05.046
Subject(s) - segmentation , artificial intelligence , computer science , atlas (anatomy) , pattern recognition (psychology) , markov random field , sørensen–dice coefficient , image segmentation , medicine , anatomy
use of a Jacobian Integration (JI) technique for the measurement of BP and LV volume changes in multicenter studies was investigated and its performance expressed in terms of reproducibility, significant group separation and sample size. Methods: Ninety subjects (Table 1) from the ADNI database, with baseline and 1-year MRI scans, each scan including two 3DT1 acquisitions, were analyzed. Each baseline scan underwent the following automated pre-processing steps: skull extraction, BP and LV segmentation. Repeated and follow-up MRI scans were affinely registered and their intensity normalized with respect to the baseline scan. JI was performed using the non-linear registration technique described in (Vemuri et al, 2003). Volume changes were computed by averaging Jacobian values inside the baseline BP and LV masks, respectively. Test-retest reproducibility was measured as the mean absolute volume change between the two 3DT1 acquisitions at each scan. A power analysis was also performed to estimate the sample size required to detect a 25% reduction in the mean annual rate of volume change with a power of 80%, using a two-sided test, and standard significance level (a1⁄40.05) for a hypothetical two-arm study. Results: Test-retest reproducibility forBPandLVwas0.1%and0.2%respectively (Table1).Mean annualized volume changes in BP and LV were (-0.5%; 1.8%) for HC, (-0.7%; 2.6%) for MCI and (-1.0 %; 5.0%) for AD, respectively. Both BP and LV volume changes were statistically different between groups (p < 0.05 forMCI, andp<0.0001 forAD,with respect toHC)andsimilar to those reported in (Jack et al., 2004). Estimated sample size per arm, without controllingforageing,was(BP:124-LV:178) forADand(BP:129-LV:205) for MCI. Sample size estimations obtained with JI were better that the one obtained with the best clinical scores (CDR).Conclusions: We have demonstrated that Jacobian integration can be applied to the measurement of whole brain and lateral ventricles volume changes in multicenter clinical trials.