There are many active research projects accessing and applying shared ADNI data. Use the search above to find specific research focuses on the active ADNI investigations. This information is requested annually as a requirement for data access.
| Principal Investigator | |
| Principal Investigator's Name: | RONG ZHU |
| Institution: | Newcastle University |
| Department: | School of Mathematics, Statistics & Physics |
| Country: | |
| Proposed Analysis: | Because of some reasons, e.g. the high cost of PET scans, the dropout of the patients, part of the data are block missing. The missing data mechanism (MDM) is very complicated in this case, and the most complex one is the non-ignorable missingness, also known as missing not at random (MNAR). We are going to deal with the block missing data with MNAR. As for the methods to analyse block missing data, Yuan et al. (2012) proposed an incomplete Multi-Source Feature (iMSF) learning method, which is a multi-task learning algorithm that is able to deal with missing feature values, and compare with some imputation method. Xiang et al. (2014) proposed the Incomplete Source-Feature Selection (ISFS) method, which can do the feature-level and source-level analysis without imputing missing data, and present efficient optimization algorithms. There are also some methods to deal with block missing data, but none of them considers the MDM, especially MNAR. Since the parameter in MNAR is not estimable, we consider a range of candidate sensitivity parameters in MNAR. After given the sensitivity parameter, we can solve the parameters in the model and impute the missing data, which are used to compare with the observed data by some distance qualities. After comparison, we can obtain an interval estimation of the sensitivity parameters in MNAR. Then, based on this interval estimation, the model can be predicted in some sense. The difficulty is that the dimension of the data is usually high, thus, it is hardly to obtain the premium data by imputation with such high dimensional variables. There are some descending dimension method occurs in our minds, which might be used to solve the imputation with high dimensional data. Then, we plan to use factor analysis to extract some important factors, and the MDM is constructed by the important factors instead of the high dimensional variables, which may help to improve the computation efficiency. Our idea is to construct the factor regression model, where the variables turn into some important factors, and use the important factors to impute the missing factors after given the sensitivity parameters in MNAR. Then we compare the imputed data with the observed data to test which range of sensitivity parameters should be better to be selected. The data that we plan to access is the Mini-Mental State Examination (MMSE) scores, which will be regarded as the response variable in our regression model; the CSF, PET, MRI and GENE-gene expression, which are the four sources block missing explaining variables/ covariates; the sex, age, education, which are the auxiliary variables. References Xiang, S., Yuan, L., Fan, W., Wang, Y., Thompson, P. M., Ye, J., Initiative, A. D. N., et al. (2014), “Bi-level multi-source learning for heterogeneous block-wise missing data,” NeuroImage, 102, 192–206. Yuan, L., Wang, Y., Thompson, P. M., Narayan, V. A., Ye, J., Initiative, A. D. N., et al. (2012), “Multi-source feature learning for joint analysis of incomplete multiple heterogeneous neuroimaging data,” NeuroImage, 61, 622–632. |
| Additional Investigators | |
| Investigator's Name: | JIAN QING SHI |
| Proposed Analysis: | Because of some reasons, e.g. the high cost of PET scans, the dropout of the patients, part of the data are block missing. The missing data mechanism (MDM) is very complicated in this case, and the most complex one is the non-ignorable missingness, also known as missing not at random (MNAR). We are going to deal with the block missing data with MNAR. As for the methods to analyse block missing data, Yuan et al. (2012) proposed an incomplete Multi-Source Feature (iMSF) learning method, which is a multi-task learning algorithm that is able to deal with missing feature values, and compare with some imputation method. Xiang et al. (2014) proposed the Incomplete Source-Feature Selection (ISFS) method, which can do the feature-level and source-level analysis without imputing missing data, and present efficient optimization algorithms. There are also some methods to deal with block missing data, but none of them considers the MDM, especially MNAR. Since the parameter in MNAR is not estimable, we consider a range of candidate sensitivity parameters in MNAR. After given the sensitivity parameter, we can solve the parameters in the model and impute the missing data, which are used to compare with the observed data by some distance qualities. After comparison, we can obtain an interval estimation of the sensitivity parameters in MNAR. Then, based on this interval estimation, the model can be predicted in some sense. The difficulty is that the dimension of the data is usually high, thus, it is hardly to obtain the premium data by imputation with such high dimensional variables. There are some descending dimension method occurs in our minds, which might be used to solve the imputation with high dimensional data. Then, we plan to use factor analysis to extract some important factors, and the MDM is constructed by the important factors instead of the high dimensional variables, which may help to improve the computation efficiency. Our idea is to construct the factor regression model, where the variables turn into some important factors, and use the important factors to impute the missing factors after given the sensitivity parameters in MNAR. Then we compare the imputed data with the observed data to test which range of sensitivity parameters should be better to be selected. The data that we plan to access is the Mini-Mental State Examination (MMSE) scores, which will be regarded as the response variable in our regression model; the CSF, PET, MRI and GENE-gene expression, which are the four sources block missing explaining variables/ covariates; the sex, age, education, which are the auxiliary variables. References Xiang, S., Yuan, L., Fan, W., Wang, Y., Thompson, P. M., Ye, J., Initiative, A. D. N., et al. (2014), “Bi-level multi-source learning for heterogeneous block-wise missing data,” NeuroImage, 102, 192–206. Yuan, L., Wang, Y., Thompson, P. M., Narayan, V. A., Ye, J., Initiative, A. D. N., et al. (2012), “Multi-source feature learning for joint analysis of incomplete multiple heterogeneous neuroimaging data,” NeuroImage, 61, 622–632. |
| Investigator's Name: | JIAN QING SHI |
| Proposed Analysis: | Because of some reasons, e.g. the high cost of PET scans, the dropout of the patients, part of the data are block missing. The missing data mechanism (MDM) is very complicated in this case, and the most complex one is the non-ignorable missingness, also known as missing not at random (MNAR). We are going to deal with the block missing data with MNAR. As for the methods to analyse block missing data, Yuan et al. (2012) proposed an incomplete Multi-Source Feature (iMSF) learning method, which is a multi-task learning algorithm that is able to deal with missing feature values, and compare with some imputation method. Xiang et al. (2014) proposed the Incomplete Source-Feature Selection (ISFS) method, which can do the feature-level and source-level analysis without imputing missing data, and present efficient optimization algorithms. There are also some methods to deal with block missing data, but none of them considers the MDM, especially MNAR. Since the parameter in MNAR is not estimable, we consider a range of candidate sensitivity parameters in MNAR. After given the sensitivity parameter, we can solve the parameters in the model and impute the missing data, which are used to compare with the observed data by some distance qualities. After comparison, we can obtain an interval estimation of the sensitivity parameters in MNAR. Then, based on this interval estimation, the model can be predicted in some sense. The difficulty is that the dimension of the data is usually high, thus, it is hardly to obtain the premium data by imputation with such high dimensional variables. There are some descending dimension method occurs in our minds, which might be used to solve the imputation with high dimensional data. Then, we plan to use factor analysis to extract some important factors, and the MDM is constructed by the important factors instead of the high dimensional variables, which may help to improve the computation efficiency. Our idea is to construct the factor regression model, where the variables turn into some important factors, and use the important factors to impute the missing factors after given the sensitivity parameters in MNAR. Then we compare the imputed data with the observed data to test which range of sensitivity parameters should be better to be selected. The data that we plan to access is the Mini-Mental State Examination (MMSE) scores, which will be regarded as the response variable in our regression model; the CSF, PET, MRI and GENE-gene expression, which are the four sources block missing explaining variables/ covariates; the sex, age, education, which are the auxiliary variables. References Xiang, S., Yuan, L., Fan, W., Wang, Y., Thompson, P. M., Ye, J., Initiative, A. D. N., et al. (2014), “Bi-level multi-source learning for heterogeneous block-wise missing data,” NeuroImage, 102, 192–206. Yuan, L., Wang, Y., Thompson, P. M., Narayan, V. A., Ye, J., Initiative, A. D. N., et al. (2012), “Multi-source feature learning for joint analysis of incomplete multiple heterogeneous neuroimaging data,” NeuroImage, 61, 622–632. |
