Bayesian Network Modelling for the Clinical Diagnosis of Alzheimer’s Disease

Abstract

AbstractAlzheimer’s disease (AD) is a progressively debilitating disease commonly affecting the elderly. Correct diagnosis is important for patients to access suitable therapies and support that can help improve or manage symptoms of the condition. Reports of misdiagnosis and difficulty diagnosing AD highlight existing clinical challenges. Here we propose a Bayesian network as a preliminary model for a complementary clinical diagnostic tool for dementia due to AD and mild cognitive impairment due to AD. The model structure was built based on medical reasoning patterns which help bridge the gap between clinical professionals and algorithmic decision making. The parameters of the model were specified from a combination of learning from data (using the NACC Uniform Data Set), extracting data from literature, and knowledge-based judgment. The resulting model includes variables laid out in NIA-AA diagnostic criteria and differentiates actual AD cases from formal AD diagnoses. The model is validated against a range of real-world data. Unlike machine-learnt (black box) AI models, this model provides a visible and auditable justification for its predictions and can be used for multiple types of ‘what if analysis’. An easy-to-use web accessible version of the model has been made available.