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: | Michael Motes |
| Institution: | Universtiy of Texas at Dallas |
| Department: | School of Behavioral and Brain Sciences |
| Country: | |
| Proposed Analysis: | We will investigate agglomerative clustering methods and parameters that use individual features (i.e., data from the different modalities) and group features into candidate “phenotypes” of diagnostic status by minimizing within-phenotype dissimilarity metrics. From the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset, we would like to use neuropsychological, neuroimaging, genetic, and cerebrospinal fluid data on normal elderly controls, patients with significant memory concerns, patients with MCI, and patients with Alzheimer's disease. More specifically we would llike to use the MPRAGE anatomical neuroimages, resting-state BOLD images, cerebrospinal fluid tau and amyloid biomarkers, PET analyloid images, APOE genetic status, demographic data (i.e., age, education, gender, race, and ethnicity), and neuropsychological data (i.e., Diagnostic Symptoms and Summary, Alzheimer's Disease Assessment Scale, Mini-Mental State Exam, Montreal Cognitive Assessment, Hechinski Ischemia Scale, Neuropsychiatric Inventory Questionnaire, Assessment of Everyday Cognition, Cogstate Battery, Cognitive Change Index, Financial Capacity Instrument, Functional Activities Questionnaire, Geriatric Depression Scale, Family History, Neuropsychological & Physical Exam Summary, and Vital Signs). In Stage 1, using 50% of the ADNI data, we will screen attributes in the data as discriminatory features of diagnostic status phenotyping through an agglomerative clustering method. This will involve first deriving dissimilarity measures based on defined attribute dissimilarity for both quantitative and categorical attributes, and then calculating a weighted average over the individual attribute dissimilarities. The obtained weights then will be evaluated based on their relative influence in the contributions of some features to defining a phenotype. In Stage 2, we will use the unanalyzed portion of the data ADNI data. The clusters obtained from Stage 1 will then be validated in Stage 2 using a supervised learning approach, with classification models from Stage 1 being used to guide learning. We will focus on adaptive, nonlinear learning models that will utilize parameter constraints and basis expansions of input features. These will include penalized discriminant functions, nonparametric logistic models, multi-layered neural networks, and support vector machines. As an illustration, our general model-building approach will focus on a nonparametric logistic model, which can reduce to a linear logistic model as a special case, while the notation is broad enough to include all levels of complexity. We will model the log-odds of a “diagnostic phenotype” as a generic, smooth function of input features, allowing for complexity through the spline basis expansion of the input features but also penalizing complexity in the minimization of the log-likelihood function. We will use 10-fold cross-validation to evaluate the prediction models. |
| Additional Investigators | |
| Investigator's Name: | Jeffrey Spence |
| Proposed Analysis: | We will investigate agglomerative clustering methods and parameters that use individual features (i.e., data from the different modalities) and group features into candidate “phenotypes” of diagnostic status by minimizing within-phenotype dissimilarity metrics. From the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset, we would like to use neuropsychological, neuroimaging, genetic, and cerebrospinal fluid data on normal elderly controls, patients with significant memory concerns, patients with MCI, and patients with Alzheimer's disease. More specifically we would llike to use the MPRAGE anatomical neuroimages, resting-state BOLD images, cerebrospinal fluid tau and amyloid biomarkers, PET analyloid images, APOE genetic status, demographic data (i.e., age, education, gender, race, and ethnicity), and neuropsychological data (i.e., Diagnostic Symptoms and Summary, Alzheimer's Disease Assessment Scale, Mini-Mental State Exam, Montreal Cognitive Assessment, Hechinski Ischemia Scale, Neuropsychiatric Inventory Questionnaire, Assessment of Everyday Cognition, Cogstate Battery, Cognitive Change Index, Financial Capacity Instrument, Functional Activities Questionnaire, Geriatric Depression Scale, Family History, Neuropsychological & Physical Exam Summary, and Vital Signs). In Stage 1, using 50% of the ADNI data, we will screen attributes in the data as discriminatory features of diagnostic status phenotyping through an agglomerative clustering method. This will involve first deriving dissimilarity measures based on defined attribute dissimilarity for both quantitative and categorical attributes, and then calculating a weighted average over the individual attribute dissimilarities. The obtained weights then will be evaluated based on their relative influence in the contributions of some features to defining a phenotype. In Stage 2, we will use the unanalyzed portion of the data ADNI data. The clusters obtained from Stage 1 will then be validated in Stage 2 using a supervised learning approach, with classification models from Stage 1 being used to guide learning. We will focus on adaptive, nonlinear learning models that will utilize parameter constraints and basis expansions of input features. These will include penalized discriminant functions, nonparametric logistic models, multi-layered neural networks, and support vector machines. As an illustration, our general model-building approach will focus on a nonparametric logistic model, which can reduce to a linear logistic model as a special case, while the notation is broad enough to include all levels of complexity. We will model the log-odds of a “diagnostic phenotype” as a generic, smooth function of input features, allowing for complexity through the spline basis expansion of the input features but also penalizing complexity in the minimization of the log-likelihood function. We will use 10-fold cross-validation to evaluate the prediction models. |
