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: | Sarah Brueningk |
| Institution: | ETH Zurich |
| Department: | Biosystems |
| Country: | |
| Proposed Analysis: | Aim. In the interim of developing novel therapies, immense public health gains could be appreciated if Alzheimer’s disease becomes preventable. The overall objective is to develop a machine learning pipeline to identify risk factors that are most associated with Alzheimer’s disease and its progression. Inclusion criteria. Patients in the ADNI will be selected for analysis based on the following criteria: 1) diagnosed with Alzheimer’s disease according to ICD10 - WHO International Classification of Diseases (ICD Code G30 [G30.1-G30.9]) and 2) complete cognitive examination at baseline and follow-up. In addition, data from a healthy age- and sex- matched cohort will be analyzed. Predictors (Features). A battery of potential predictors will be used with the algorithms. Predictors will include: 1) subject and disease characteristics (demographics, disease severity); 2) available biospecimens (blood markers, genetic data); 3) imaging (structural brain magnet resonance imaging (MRI) and CT scan); 4) medical history (diagnostic features); 5) concomitant medications; 6) comorbidities (additional diseases co-occurring with Alzheimer’s disease), and functional measures (e.g., cognitive function). For predictors variables where data are missing on >20% of the sample, this variable will be excluded from the primary analysis, and included in sensitivity analyses only. Specific predictors will be determined via consultation with content experts (i.e., clinicians), as well as from established measures from published literature. Analysis. Building on state-of-the-art machine learning methods, we will develop an algorithm to identify the features that are most associated with cognitive impairments (e.g., Alzheimer’s disease) and its progression. We will perform a feature extraction in order to reduce the amount of resources required to describe our large data sets. Feature extraction is integral to an effective model (algorithm) construction. A variety of feature extraction methods will be applied, including Principal Component Analysis (PCA) and Independent Component Analysis (ICA). We will combine and extend those techniques to deal with multi-modal classification and learn interpretable predictive features across modalities (imaging, genetic, biospecimens, clinical data). Binary classifiers (yes/no for disease state) will be estimated using conventional parametric methods (logistic regression), non-parametric machine learning methods (Random Forest model, Gradient boosting trees), and an ensemble of classifiers based on gradient boosted trees. The latter outputs from several individual classifiers to capture different properties from different methods to create a single classifier to further improve overall predictive performance. Precision-recall curves will be employed to test the model performances. |
| Additional Investigators | |
| Investigator's Name: | Catherine Juetzler |
| Proposed Analysis: | Aim. In the interim of developing novel therapies, immense public health gains could be appreciated if Alzheimer’s disease becomes preventable. The overall objective is to develop a machine learning pipeline to identify risk factors that are most associated with Alzheimer’s disease and its progression. Inclusion criteria. Patients in the ADNI will be selected for analysis based on the following criteria: 1) diagnosed with Alzheimer’s disease according to ICD10 - WHO International Classification of Diseases (ICD Code G30 [G30.1-G30.9]) and 2) complete cognitive examination at baseline and follow-up. In addition, data from a healthy age- and sex- matched cohort will be analyzed. Predictors (Features). A battery of potential predictors will be used with the algorithms. Predictors will include: 1) subject and disease characteristics (demographics, disease severity); 2) available biospecimens (blood markers, genetic data); 3) imaging (structural brain magnet resonance imaging (MRI) and CT scan); 4) medical history (diagnostic features); 5) concomitant medications; 6) comorbidities (additional diseases co-occurring with Alzheimer’s disease), and functional measures (e.g., cognitive function). For predictors variables where data are missing on >20% of the sample, this variable will be excluded from the primary analysis, and included in sensitivity analyses only. Specific predictors will be determined via consultation with content experts (i.e., clinicians), as well as from established measures from published literature. Analysis. Building on state-of-the-art machine learning methods, we will develop an algorithm to identify the features that are most associated with cognitive impairments (e.g., Alzheimer’s disease) and its progression. We will perform a feature extraction in order to reduce the amount of resources required to describe our large data sets. Feature extraction is integral to an effective model (algorithm) construction. A variety of feature extraction methods will be applied, including Principal Component Analysis (PCA) and Independent Component Analysis (ICA). We will combine and extend those techniques to deal with multi-modal classification and learn interpretable predictive features across modalities (imaging, genetic, biospecimens, clinical data). Binary classifiers (yes/no for disease state) will be estimated using conventional parametric methods (logistic regression), non-parametric machine learning methods (Random Forest model, Gradient boosting trees), and an ensemble of classifiers based on gradient boosted trees. The latter outputs from several individual classifiers to capture different properties from different methods to create a single classifier to further improve overall predictive performance. Precision-recall curves will be employed to test the model performances. |
