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: | Helmet Karim |
| Institution: | University of Pittsburgh |
| Department: | Psychiatry |
| Country: | |
| Proposed Analysis: | Current methods of detection for Alzheimer's Disease (AD) have largely focused on predicting current diagnosis of AD, however there is significant evidence that early neurodegeneration is a strong predictor of future cognitive decline and diagnosis of AD. Few analyses have focused primarily on predicting future AD diagnosis - we propose to use a deep neural network to predict future AD. This is important as there is mounting evidence that treating in the earliest stages or altogether focusing on prevention may be an optimal strategy for AD. Several recent studies have shown that buildup of preclinical markers of AD (amyloid and tau) are associated with neurodegeneration - which may be the critical factor in the last stages of the AD cascade (i.e., AD biomarker curves indicate atrophy as the final stage of disease prior to clinical onset). Furthermore, there is sufficient evidence that current levels of gray matter are a marker of reserve - individuals with greater volume/surface area may be able to tolerate a longer course of disease (i.e., brain reserve). Additionally, other diseases can also affect gray matter neurodegeneration or cognitive function - many models fail to take this into account. We aim to use all ADNI data to predict future cognitive dysfunction and years to onset. Additionally, we propose to develop a model that highlights critical features of the imaging data to further help evaluate why a particular patient is at greater risk. This model will use an unprocessed T1-weighted MRI and predict the following: Aim 1. We propose to train a deep neural network model that predicts whether an individual is stable cognitive function (i.e., currently healthy and stays cognitively unimpaired, stable HC or sHC), is cognitively impaired (i.e., current MCI or AD, current cognitive impairment or CI), or is cognitively healthy and develops AD/MCI (i.e., future cognitive impairment or fCI). Aim 2. We propose to optimize a second model in the fCI group to predict years to onset (i.e., how many years prior to a diagnosis of AD from the current scan). Aim 3. We will then develop an approach that outputs a class activation map (CAM) that highlights individualized features on the T1-weighted MRI that are most predictive of future cognitive impairment. This model will help not only identify patients at risk but also develop an approach that informs clinical guidance by identifying critical features of AD-risk in individualized patients. Using such approaches will allow for serial tracking of neurodegenerative health in individual patients and identify changes over time that may be indicative of future risk of impairment. This approach will allow for healthcare providers to identify and track patients at risk early on and potentially evaluate clinical progression and management of preventative care. In order to help reduce Alzheimer’s Disease progression, it is important to intervene early in the process. In this way, we are able to better detect early biomarkers for AD and develop patient-specific treatment plans. It is for this reason that an efficient, automated, and accurate deep learning model may help identify patients at the earliest stages of risk and allow for preventative care and treatment. |
| Additional Investigators |
