TY - JOUR
T1 - Quantitative comparisons of three automated methods for estimating intracranial volume
T2 - A study of 270 longitudinal magnetic resonance images
AU - Shang, Xiaoyan
AU - Carlson, Michelle C.
AU - Tang, Xiaoying
N1 - Funding Information:
This study was supported by the National Natural Science Foundation of China (Grant no. 81501546 ), the SYSU-CMU Shunde International Joint Research Institute Start-up Grant (Grant no. 20150306 ), and the Johns Hopkins Neurobehavioral Research Unit and a supplement from the National Institute on Aging (Grant no. P01 AG027735-03 ). The authors would like to acknowledge the generous contribution of all Baltimore Experience Corps Trial and Brain Health Study participants who gave their time to be involved in this study. Without their service and contributions, this research would not be possible.
Funding Information:
This study was supported by the National Natural Science Foundation of China (Grant no. 81501546), the SYSU-CMU Shunde International Joint Research Institute Start-up Grant (Grant no. 20150306), and the Johns Hopkins Neurobehavioral Research Unit and a supplement from the National Institute on Aging (Grant no. P01 AG027735-03). The authors would like to acknowledge the generous contribution of all Baltimore Experience Corps Trial and Brain Health Study participants who gave their time to be involved in this study. Without their service and contributions, this research would not be possible.
Publisher Copyright:
© 2018 Elsevier B.V.
PY - 2018/4/30
Y1 - 2018/4/30
N2 - Total intracranial volume (TIV) is often used as a measure of brain size to correct for individual variability in magnetic resonance imaging (MRI) based morphometric studies. An adjustment of TIV can greatly increase the statistical power of brain morphometry methods. As such, an accurate and precise TIV estimation is of great importance in MRI studies. In this paper, we compared three automated TIV estimation methods (multi-atlas likelihood fusion (MALF), Statistical Parametric Mapping 8 (SPM8) and FreeSurfer (FS)) using longitudinal T1-weighted MR images in a cohort of 70 older participants at elevated sociodemographic risk for Alzheimer's disease. Statistical group comparisons in terms of four different metrics were performed. Furthermore, sex, education level, and intervention status were investigated separately for their impacts on the TIV estimation performance of each method. According to our experimental results, MALF was the least susceptible to atrophy, while SPM8 and FS suffered a loss in precision. In group-wise analysis, MALF was the least sensitive method to group variation, whereas SPM8 was particularly sensitive to sex and FS was unstable with respect to education level. In terms of effectiveness, both MALF and SPM8 delivered a user-friendly performance, while FS was relatively computationally intensive.
AB - Total intracranial volume (TIV) is often used as a measure of brain size to correct for individual variability in magnetic resonance imaging (MRI) based morphometric studies. An adjustment of TIV can greatly increase the statistical power of brain morphometry methods. As such, an accurate and precise TIV estimation is of great importance in MRI studies. In this paper, we compared three automated TIV estimation methods (multi-atlas likelihood fusion (MALF), Statistical Parametric Mapping 8 (SPM8) and FreeSurfer (FS)) using longitudinal T1-weighted MR images in a cohort of 70 older participants at elevated sociodemographic risk for Alzheimer's disease. Statistical group comparisons in terms of four different metrics were performed. Furthermore, sex, education level, and intervention status were investigated separately for their impacts on the TIV estimation performance of each method. According to our experimental results, MALF was the least susceptible to atrophy, while SPM8 and FS suffered a loss in precision. In group-wise analysis, MALF was the least sensitive method to group variation, whereas SPM8 was particularly sensitive to sex and FS was unstable with respect to education level. In terms of effectiveness, both MALF and SPM8 delivered a user-friendly performance, while FS was relatively computationally intensive.
KW - Automated estimation
KW - Freesurfer
KW - Magnetic resonance imaging
KW - Multi-atlas likelihood fusion
KW - Statistical Parametric Mapping
KW - Total intracranial volume
UR - http://www.scopus.com/inward/record.url?scp=85042073879&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85042073879&partnerID=8YFLogxK
U2 - 10.1016/j.pscychresns.2018.02.005
DO - 10.1016/j.pscychresns.2018.02.005
M3 - Article
C2 - 29458996
AN - SCOPUS:85042073879
SN - 0925-4927
VL - 274
SP - 23
EP - 30
JO - Psychiatry Research - Neuroimaging
JF - Psychiatry Research - Neuroimaging
ER -