Disentangle heterogeneity of dementia patterns using deep learning model with
structural MR imaging 


   Note: An Official Proposal has not been Submitted to NACC.


   Lead Investigator:    Habes Mohamad
   Institution      :    University of Texas Health Science Center at San A
   E-Mail           :    habes@uthscsa.edu
   Proposal ID      :    1738



   Proposal Description:
   Aims:    (1)     Develop advanced deep and transfer learning based methods to differentiate  dementia subtypes using MRI data, we aim to differentiate Alzheimer???s  disease, Frontotemporal Dementia, Lewy Body Dementia and Vascular Dementia.  We  will develop transfer learning techniques to improve learning abilities and  increase robustness in the models.    (2)     Develop explainable AI (XAI) method to visualize how deep learning model  made the decision and discover specific dementia subtypes patterns. The output  of this method is relevance maps reveal which brain regions contributed most to  the classification decisions.    Hypotheses:    (1)     Deep learning model would accurately classify different dementias using MRI  data. Transfer learning would greatly improve accuracy for small dataset  classification.    (2)     Relevance maps generated from XAI method would show distinct patterns among  different types of dementias.