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: | Saima Hilal |
| Institution: | National University of Singapore |
| Department: | Epidemiology |
| Country: | |
| Proposed Analysis: | Objectives and aims: Among the emerging challenges of an aging society, with its increasing prevalence of age-associated diseases, inadequate treatment for mild cognitive impairment (MCI) and dementia is one of the largest unmet healthcare needs globally. Early detection of this devastating disease involves complex and heterogeneous mechanisms. Deep learning techniques identifies complex patterns in high-dimensional MRI data, which can make clinical predictions feasible. Hence, our objective is to develop MRI-based deep learning algorithms for prediction of cognitive impairment at its early stages. Our specific aim is to develop AI method using brain MRI, which can facilitate quick, accurate and consistent prediction of MCI and dementia for clinical and research purposes. We hypothesize that brain MRI-based deep learning approaches harmonize complex patterns and identifies ‘at-risk’ individuals for cognitive impairment and dementia. Deep learning algorithm: We will employ the Autoencoder (AE) algorithm in this project consisting of ADNI subjects, which is an unsupervised artificial neural network. It attempts to produce outputs identical to its input. During the training process, the AE model is purely trained on No Cognitive Impairment (NCI) images. The encoder portion receives the input NCI image and compresses it into a smaller dimension z, and then the encoded data goes into the decoder. Ideally, the output of the decoder, that is, the reconstructed image would be identical to the input NCI image. We call the difference between the input data and reconstructed data as reconstruction error, and the model is trained by minimizing the reconstruction error. Based on this rationale, during the inference process, for any new image, if the output from the trained model is very different from the input, or if the reconstruct error is large, we can detect it as an MCI image. Outcomes and Impact: This proposal will provide a unique opportunity in the following aspects: 1) To generate pilot data and translate research to other population-based cohorts. 2) Results from this study may lead to earlier detection and more accurate diagnosis, which can in turn result in more effective use of resources through early adoption of treatments or appropriate interventions. 3) The data from this proposal can be incorporated into multimodal algorithms integrating multiple indicators into a predictive platform taking into account lifestyle, blood biomarkers, clinical assessments and MRI information. 4) To build up a user-friendly platform for future clinical and research uses, which will only require input in terms of MRI data, and then the platform would be able to train different models and show performance of these models automatically. |
| Additional Investigators | |
| Investigator's Name: | M. Daiman Maskur |
| Proposed Analysis: | Same as provided for the PI |
| Investigator's Name: | M. Rusdy Taib |
| Proposed Analysis: | Same as provided for the PI |
