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: | Sung-Ho Woo |
| Institution: | Interdisciplinary Brain Research Institute, Dongguk University |
| Department: | Department of medicine |
| Country: | |
| Proposed Analysis: | Research title: Development of synthetic MRI generation and evaluation methods for the inference of brain aging and disease progress Research Objectives: To build a highly reliable medical AI system, a large amount of high-quality data is required. However, in the medical domain, large-scale data collection requires an enormous amount of time and effort. Also, because medical data is often unstructured, the construction of a large-scale dataset can be irregular and inaccurate. In particular, in the case of brain imaging data related to aging and degenerative diseases, it is extremely difficult to construct a longitudinal dataset, so the development of AI for the purpose of inferring and predicting diseases is very limited. Additionally, since medical data contains sensitive individual information, there are many limitations in using real data as learning data. To solve this problem, synthetic data generation technology that imitates the statistical distribution and correlation of the real data has recently emerged. Therefore, our study aims to develop technology to generate synthetic data on the occurrence and progress of dementia using methods such as knowledge encoding, objective reinforcement, data assimilation, and correlation modeling. Furthermore, evaluation techniques to verify the reliability and validity of synthetic data to which the developed technique is applied are also being developed. Through this, we aim to contribute to the development of state-of-the-art AI in the brain healthcare fields where training data is scarce by developing the technologies that synthesize, generate, and evaluate MRI data on brain aging and brain diseases for which longitudinal data is difficult to obtain. Additionally, by developing a new clinical and brain imaging data synthesis method based on a large amount of clinical and physiological information, we aim to verify the proposed neurophysiological mechanisms of cognitive decline and dementia development and progression through the synthesized data Method: We plan to utilize the ADNI dataset and the dataset from Dongguk University Ilsan Hospital, with which our institute is collaborating. The dataset from Dongguk University Ilsan Hospital consists of clinical information and MRIs of approximately 2000 normal elderly people, mild cognitive impairment (MCI) patients, and Alzheimer's disease (AD) patients. We will conduct research on these datasets in the following ways. FSL, FreeSurfer, and SPM will be utilized to perform preprocessing, brain tissue segmentation, and parcellation of anatomical regions. Brain regions segmented from 3D T1-w MRI include subcortical areas such as the hippocampus, anatomically defined cortical areas such as the prefrontal cortex and orbitofrontal cortex, and brain tissues such as gray matter, white matter, and ventricles. The correlation between the volume and shape information of the segmented brain regions from the 3D T1-w MRI and the clinical information will be estimated. From T1- and T2-w MRIs, myelination indices in brain anatomical regions and white matter tracts will be calculated. White matter hyperintensity (WMH) will be extracted from FLAIR MRIs. We will build a correlation model of risk factors for MCI, AD onset and progression, and structural changes such as brain atrophy, and use it to develop synthetic data generation methods. Additionally, we will extract knowledge about the correlation between WMH and various clinical information and their effects on brain structural changes, and use it to generate synthetic MRI. We will develop techniques for longitudinal synthetic MRI generation through methods such as knowledge encoding, correlation modeling, and data assimilation. We will also develop evaluation techniques for statistical similarity between synthesized data and real datasets and for the validity of AI development. |
| Additional Investigators |
