Premium
P4‐091: Neuroimaging findings in patients with Alzheimer's disease and mild cognitive impairment
Author(s) -
Ahn Jin-Young,
Kim Hee-Tae,
Heo Jae-Hyeok
Publication year - 2015
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.2015.06.1796
Subject(s) - hyperintensity , medicine , dementia , magnetic resonance imaging , population , memory clinic , cardiology , neuroimaging , atrophy , vascular dementia , stroke (engine) , cognitive decline , radiology , disease , psychiatry , mechanical engineering , environmental health , engineering
Changes in HVs were calculated using the extended Boundary Shift Integral method. Table 1 summarizes the predictors and nuisance variables used in the modelling. Two different models were tested: least-mean-squares (LMS) regression with elastic net regularization and least absolute deviation (LAD) regression. The regularized regression includes penalty terms that constrain some of the regression coefficients to zero, leading to feature selection. To account for the effect of baseline disease severity, we added a disease state index (DSI) variable; this scalar captures the similarity of the subject’s data to healthy and diseased populations. Performance of the models was evaluated using the 10fold cross-validation stratified according to the four subject groups above. Results: Table 2 presents root-mean-square errors (RMSE) and correlation coefficients between the predicted and observed changes. The LMS L1 and LAD L1 regression models were selected as the best, based on these parameters. Table 3 shows the variables most often selected into the models. Correlations between the observed and predicted changes were high (Figure 1) and best for LAD L1 regression. Conclusions:Once validated in additional cohorts, the multivariate LAD L1model to predict change in HV may be useful to predict AD progression. In addition, such models may control for the variance in AD progression related to known risk factors, facilitating the discovery of novel biomarkers and disease mechanisms that underlie the extreme phenotype, namely those subjects whose disease progresses faster or slower than predicted.