Samantha Smiley, Brigham Young University
Mathematical Sciences
Dementia is a clinical syndrome characterized by an overall loss of cognitive ability. There are multiple forms of dementia with various causes and various impacts on the suffering individuals. Accurate diagnosis is essential to effective intervention and treatment. Currently, clinicians lack a biological marker that definitively distinguishes the different forms of dementia. Hence, they rely on physical exams, neuropsychological tests, and patient report to provide a diagnosis. Recent advances in brain imaging make it possible to obtain detailed maps of brain activity, which in turn may offer insight into many conditions such as dementia. Developing a predictive model from patient data, including brain scans, would greatly enhance the ability of clinicians to provide accurate diagnosis, and hence appropriate treatment, to their patients. Doing so, however, is not trivial as patient data is heterogeneously and non-uniformly distributed across sites, where some sites have far more data than others and calibration varies among scanners used. We report on the development of novel predictive models based on distributed learning for the effective diagnosis of dementia.