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: | Yibo Xi |
| Institution: | ShanXi Medical University |
| Department: | ShanXi Medical University |
| Country: | |
| Proposed Analysis: | At present, most studies on Alzheimer's disease (AD) and Mild cognitive impairment (MCI) use a single data model to predict, such as the stage of AD. The fusion of multiple data patterns can provide an overall view of AD staging analysis. Therefore, we use deep learning to comprehensively analyze imaging (magnetic resonance imaging (MRI)), genes (Single-nucleotide polymorphism (SNPs)) and clinical test data, and divide patients into AD, MCI and control group (CN). We use stacked noise reduction Autoencoder to extract features from clinical and genetic data, and use 3D Convolutional neural network (CNN) to process imaging data. AD multimodal analysis has been combined with various imaging modes, such as structural MRI (T1 weighted, T2 weighted), functional MRI, Positron emission tomography (PET) and image genetics. In addition, genetics has been combined with clinical data to increase data labels and phenotypes. In addition to shallow learning, deep learning models such as Autoencoder and deep belief network (Table A1) have been used for fusion of PET and MRI image data, and prediction has been improved. Although the use of multiple types of data has shown excellent performance in clinical decision support, a major drawback of widely adopting deep learning models for clinical decision-making is the lack of methods to explain the definition of deep models. We address this challenge by developing new perturbation algorithms and clustering based methods to identify the most important features that contribute to decision-making. we elucidate the main contributions to AD staging prediction as follows: The new deep learning architecture is superior to shallow learning models; Multimodal data analysis using deep learning is superior to single modal deep learning models; New interpretable deep learning methods can extract the best performing features. |
| Additional Investigators |
