.png){kind=icon}
.png){kind=icon}
Return
Exponential random graph-based eXplainable Artificial Intelligence for Alzheimer disease
.png){kind=icon}
PRE
AI
DOI：10.1103/dkn2-9b2t.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
The use of statistical physics models to investigate real-world networks and reveal their underlying dynamics has shown promising results and acquired increasing attention. Here, we show how exponential random-graph (ERG) models can be suitably adopted to characterize how Alzheimer's disease (AD) affects brain connectivity. Magnetic-resonance imaging (MRI) of the brain was used to define a brain connectivity network whose nodes are the different brain regions, and the links indicate the pairwise structural relationships. Based on T1-weighted MRI brain scans of 126 normal controls (NC) and 92 AD patients, ERGs were able to outline both global and local disease patterns. Our findings demonstrate that ERGs accurately highlight how AD affects brain connectivity reaching an overall classification accuracy of 0.82 +/- 0.08. Besides, ERGs outline which regions of the brain are the most affected by the disease, thus proving to be a formidable instrument also to investigate the disease pathological mechanisms; more importantly, as these effects are evaluated at patient level, they can be exploited to design innovative diagnosis support systems or to provide a novel explainable framework for decision support systems. Finally, thanks to its generality, the approach proposed in this study paves the way for further applications and investigations inquiring into the use of ERGs for other diseases and different data sources or the use of alternative models.
Keywords:
RANDOM FOREST
STATISTICAL-MECHANICS
HIPPOCAMPAL ATROPHY
NETWORKS
CLASSIFICATION
SELECTION
PHYSICS
Journal
IF:
2.4
Papers:
1.3K
Citations:
10.2W