A Graph Gaussian Embedding Method for Predicting Alzheimer's Disease Progression with MEG Brain Networks
This work addresses early diagnosis and prediction of Alzheimer's disease progression for patients with mild cognitive impairment, representing an incremental improvement in domain-specific methods.
The authors tackled the problem of predicting Alzheimer's disease progression by developing a deep learning method called MG2G to map high-dimensional brain networks into a low-dimensional latent space, enabling early stage prediction and identification of brain region alterations.
Characterizing the subtle changes of functional brain networks associated with the pathological cascade of Alzheimer's disease (AD) is important for early diagnosis and prediction of disease progression prior to clinical symptoms. We developed a new deep learning method, termed multiple graph Gaussian embedding model (MG2G), which can learn highly informative network features by mapping high-dimensional resting-state brain networks into a low-dimensional latent space. These latent distribution-based embeddings enable a quantitative characterization of subtle and heterogeneous brain connectivity patterns at different regions and can be used as input to traditional classifiers for various downstream graph analytic tasks, such as AD early stage prediction, and statistical evaluation of between-group significant alterations across brain regions. We used MG2G to detect the intrinsic latent dimensionality of MEG brain networks, predict the progression of patients with mild cognitive impairment (MCI) to AD, and identify brain regions with network alterations related to MCI.