Gaussian embedding-based functional brain connectomic analysis for amnestic mild cognitive impairment patients with cognitive training

Abstract

Identifying heterogeneous cognitive impairment markers at an early stage is vital for Alzheimer's disease diagnosis. However, due to complex and uncertain brain connectivity features in the cognitive domains, it remains challenging to quantify functional brain connectomic changes during non-pharmacological interventions for amnestic mild cognitive impairment (aMCI) patients. We present a new quantitative functional brain network analysis of fMRI data based on the multi-graph unsupervised Gaussian embedding method (MG2G). This neural network-based model can effectively learn low-dimensional Gaussian distributions from the original high-dimensional sparse functional brain networks, quantify uncertainties in link prediction, and discover the intrinsic dimensionality of brain networks. Using the Wasserstein distance to measure probabilistic changes, we discovered that brain regions in the default mode network and somatosensory/somatomotor hand, fronto-parietal task control, memory retrieval, and visual and dorsal attention systems had relatively large variations during non-pharmacological training, which might provide distinct biomarkers for fine-grained monitoring of aMCI cognitive alteration. zheimer's disease (AD) is a neurodegenerative brain disorder and the most common form of dementia 1 . However, there is still no cure and no effective drug treatment for AD 2 . Hence, non-pharmacological cognitive intervention for patients at early stages of AD has received a lot of attention due to its non-invasive manner, safety and scalability. Recent studies show that non-pharmacological cognitive intervention can play a positive role in delaying the process or even reducing the cognitive decline for both healthy controls 3 and amnestic mild cognitive impairment (aMCI) patients 4 . In particular, aMCI is a vital prodromal state of AD harboring memory impairment and has a high risk to develop AD 5 . Multi-domain interventions A

A Graph Gaussian Embedding Method for Predicting Alzheimer's Disease Progression With MEG Brain Networks

“…In order to automatically learn multi-scale, nonlinear MEG brain network embeddings (or "patterns") in latent space from original high-dimensional MEG brain networks, yet maximally preserving the structure properties for accurate AD progression prediction, we proposed to apply a deep learning-based embedding method called multiple graph Gaussian embedding model (MG2G) [37]. This model is a generalization of the Graph2Gauss architecture [31] to multiple graphs.…”

“…The architecture of the deep learning MG2G model is shown in Fig. 1D, and detailed information is available in [37]. Briefly, the MG2G model learns non-linear node embeddings from original highdimensional brain networks into a stochastic latent space.…”

“…, where E pos and E neg refer to the Kullback-Leibler (KL) divergence energy of the generated node triplets [37] involving positive node pairs and negative node pairs, respectively. In particular, we computed the shortest distances between node pairs based on the edge weights and generated different hops.…”

“…In particular, we computed the shortest distances between node pairs based on the edge weights and generated different hops. In order to capture high-order proximity, we sampled valid triplet sets based on the obtained hops [37].…”

“…First, we constructed subject-specific MEG brain networks based on mutual information in alpha band for brain regions extracted from the Desikan-Killiany atlas [36]. We then employed the multiple graph Gaussian embedding model (MG2G) [37] to learn highly informative brain network embeddings (patterns) from the original high-dimensional MEG brain networks. Thus, each brain region was represented as a multivariate Gaussian distribution (or Gaussian embeddings) via two vectors, the mean and variance.…”