A Machine Learning Framework for Interpretable Linked Mechanisms between
Alzheimer's Disease and TBI
 
   Lead Investigator:    Ruogu Fang
   Institution      :    University of Florida
   E-Mail           :    ruogu.fang@bme.ufl.edu
   Proposal ID      :    1460



   Proposal Description:
   The goal of this project is to design a machine learning network in relation with neurodegeneration mechanisms for the diagnosis of Alzheimer's disease. It has been studied that patients with TBI are at higher risk of Alzheimer's than its counterpart. Our hypothesis is that machine learning methods can be used to identify specific cortical degeneration regions and mechanisms in Alzheimer's disease that can be further induced by TBI. To this end, we look to explore alterations in Alzheimer's patients with and without TBI, with respect to their similarities and differences in neurodegeneration.    The objectives of this study include   1) To design and validate an imaging-data-driven machine learning classifier for Alzheimer's with and without TBI, that reveals indicative regions of the brain related to automated diagnosis and its connection with known neuroscience mechanisms.  2) To design and validate a biomarker-driven machine learning network for the classification of Alzheimer's with and without TBI, that reveals indicative mechanisms at which Alzheimer's is induced and accelerated by TBI.  3) Evaluating the importance of such biomarkers in combination with the machine learning classifier in an end-to-end framework by analyzing the performance differences between the individual counterparts. This final network is intended to be specialized towards Alzheimer's and TBI for the purpose of early diagnosis and to reveal new neurodegeneration mechanisms.     Expected Outcome: The results from each network should reveal interpretable features from the brain image and from a biomarker-based model that reveals a shared correlation in cognitive impairment. A machine learning feature selection method for biomarkers is intended to perform as equivalently to the imaging-network which would act as a proper biomarker for early diagnosis. By studying TBI induced mechanisms, such results would indirectly lead to newly realized mechanisms for the progression of Alzheimer's disea