Early prediction of dementia using fMRI data with a graph convolutional network approach

Area of Science:

• Neuroscience
• Medical Imaging
• Artificial Intelligence

Background:

• Alzheimer's disease (AD) is a progressive neurodegenerative dementia with significant global health implications.
• Early detection of mild cognitive impairment (MCI) is crucial for timely intervention and preventing dementia progression.
• Magnetic Resonance Imaging (MRI) offers valuable data for analyzing brain structure and function.

Purpose of the Study:

• To develop and evaluate a Graph Convolutional Network (GCN) framework for early dementia prediction using resting-state fMRI data.
• To assess the GCN model's performance in classifying MCI from normal controls (NCs) and predicting dementia risk in NCs.
• To explore the impact of different functional connectivity (FC) types and processing methods on classification accuracy.

Main Methods:

• A functional connectivity (FC) based Graph Convolutional Network (GCN) framework was developed for binary classification tasks.
• Resting-state fMRI data from the OASIS-3 dataset were utilized.
• The GCN model was compared against a baseline GCN and a Support Vector Machine (SVM) using various FC approaches.

Main Results:

• The proposed GCN model achieved superior performance, with the highest accuracy reaching 91.2% and an average accuracy of 80.3%.
• The GCN framework demonstrated significantly better performance than both the baseline GCN and SVM.
• Analysis indicated that individual FC generally performed better than global FC, though specific global graph connectivities showed competitive results, highlighting the importance of appropriate connectivity.

Conclusions:

• Graph Convolutional Networks show significant potential for the early diagnosis of dementia and MCI using MRI data.
• The study provides valuable insights into brain network connectivity patterns associated with cognitive decline.
• The developed GCN framework has substantial potential for clinical applications in early dementia detection and intervention strategies.