ADiag: Graph Neural Network Based Diagnosis of Alzheimer's Disease
This work offers a new quantitative diagnostic tool for clinicians to more accurately identify Alzheimer's Disease, potentially improving patient quality of life.
This paper addresses the challenge of Alzheimer's Disease diagnosis by proposing ADiag, a novel quantitative method utilizing GraphSAGE Network and Dense Differentiable Pooling (DDP) to analyze cortical thickness differences. Preliminary tests show ADiag achieves a robust accuracy of 83%, outperforming existing diagnostic techniques.
Alzheimer's Disease (AD) is the most widespread neurodegenerative disease, affecting over 50 million people across the world. While its progression cannot be stopped, early and accurate diagnostic testing can drastically improve quality of life in patients. Currently, only qualitative means of testing are employed in the form of scoring performance on a battery of cognitive tests. The inherent disadvantage of this method is that the burden of an accurate diagnosis falls on the clinician's competence. Quantitative methods like MRI scan assessment are inaccurate at best,due to the elusive nature of visually observable changes in the brain. In lieu of these disadvantages to extant methods of AD diagnosis, we have developed ADiag, a novel quantitative method to diagnose AD through GraphSAGE Network and Dense Differentiable Pooling (DDP) analysis of large graphs based on thickness difference between different structural regions of the cortex. Preliminary tests of ADiag have revealed a robust accuracy of 83%, vastly outperforming other qualitative and quantitative diagnostic techniques.