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: | Morteza Amini |
| Institution: | Institute for Cognitive Science Studies |
| Department: | Cognitive Science |
| Country: | |
| Proposed Analysis: | Alzheimer’s disease is the most prevalent aging-associated neurodegenerative disease, affecting two percent of the elderly population (increasing), and is an unrecoverable disease. Besides the clinical diagnostic criteria, certain biomarkers help the physician or researcher to identify potential AD patients in a case of insufficient cognitive impairment, causing diagnosis to fail. In steps of mild cognitive impairment (MCI), for instance, accurate utilization of these biomarkers calls for greater verification and validation. The physician’s diagnosis accuracy and decision-making capability for prescription and selection of drug dose can be enhanced through PET-fMRI fusion. The main purpose of the present study is to investigate how data fusion is used for early diagnosis of potential aMCI patients and to differentiate them from a healthy control group of the same age. Prediction capability is examined here for several fMRI features in large scale, and their combination is assessed as the patients are performing episodic memory tasks. An integrated aMCI index is proposed based on the analysis results for establishment of an efficient classification system for early diagnosis of aMCI. The eventual purpose of the research is to present a highly efficient method and the relevant visualization tool that can contribute to early diagnosis of aMCI. Use of the deep learning method causes the multimodal fusion image to be effective. We use a CNN-based fusion method for fMRI and PET medical images for early diagnosis of Alzheimer’s disease. In this research, we first present the architecture of the Siamese convolutional neural network trained with natural data. We then train the proposed CNN with medical images using the transfer learning method. The training dataset is composed of positive and negative patch pairs of Shearlet coefficients. We feed the samples in a two- stream-deep CNN to extract the feature map. We then apply similarity metric learning based on cross- correlation to learn the mapping among the features. We use the stochastic gradient descent (SGD) algorithm to minimize the logistic loss objective function. The fusion process flow is converted into several images through decomposition of the source fMRI and PET mages through non-subsampled Shearlet transform. The high-frequency subbands are fused based on a weighted normalized cross-correlation among the feature maps given by the extraction part of the CNN, while the low-frequency coefficients are combined using the local energy. The training and test datasets contain pre-registered fMRI and PET pairs available in the database. Alzheimer’s disease can be delayed through an increase in the accuracy of its early diagnosis with pharmacotherapy and more accurate specification of drug dose. |
| Additional Investigators |
