Early prediction of Alzheimer's Disease from MRI Images

 
   Lead Investigator:    Narges Razavian
   Institution      :    New York University School of Medicine
   E-Mail           :    narges.razavian@nyumc.org
   Proposal ID      :    1138



   Proposal Description:
   Personalized medicine requires the ability to quantify the 'change' of the physiology of the body over time. This quantification impacts decisions involved in diagnosis, staging, and treatment of the disease. Quantifying this change is challenging for complex and high-dimensional signals like brain MRI.    Currently, there has been great progress in AI and deep learning to quantify changes over time on images, particularly in the context of video classification and stereo-vision. None of these techniques have been translated to medical questions. We are interested in:    1)Looking at a sequence of MRI scans (Brain) over time, classify the trajectory into diseases of the brain. Our aim is to forecast/detect conditions early.        1.1) We are particularly interested in development of dementia and Alzheimer?s disease(AD), therefore we will focus on Normal, Mild-Cognitive-Impairment(MCI) and AD.                                                                            1.2) There?s supporting evidence that mechanisms of brain degeneration may be related to vascular conditions, glucose metabolism, hormonal changes, etc. Therefore in addition to clinical classes of MCI and AD, we will also focus on classifying the sequences of scans into all comorbidities, and will use advanced AI techniques to disentangle the anatomical changes that are attributed to each of the comorbid conditions.    We have developed our deep learning model to identify AD/MCI/CN directly from T1 MRIs. We have also brought in ADNI data to improve our model and evaluation over multiple large cohorts. For publication purposes, we will be comparing our model to a Freesurfer based model, and we need the brain regional volumes for that.