skip to content
LeeLab
[WIP] Multiple imputation for neuroimagin data
Literature Review
Multiple imputation for high-dimensional data
- Evaluate MI methods based on Bayesian Lasso regression and regualized regression
- Conclusion: Bayesian lasso regression and its extensions are better suited for multiple imputation in the presence of high-dimensional data than the other regression methods
- For high-dimensional data, MICE is not feasible.
- MICE-DURR: using regularized regression in each MICE iteraction
- MICE-IURR: approximate distribution of the mnissing values based on regularized regression (kind of Bayesian idea)
- Application1 : Georgia registry data
- n=86322 and 203 variables; 13 variables are of interest, and only 3 variables had completely observed.
- Well, the application is not high-dimensional..
- Application2: Prostsate cancer data
- n=99, 20,000 genomic biomarkers are avaialble. For illustration 3 variables were imputed using 2107 biomarkers do not have missing values.
- Still not imputing all high-dimensional data.
- Multiple Imputation Random Lasso (MIRL)
- Application: Project EAT
- n=2793, p=62
- simulation showed p=100 case. Not sure it works for neuroimaging.
- MI diagnostic tool
- We need to address this aspect as well.
@link
@link
References