Deep learning models reveal the link between dynamic brain connectivity patterns and states of consciousness

Abstract

Decoding states of consciousness from brain activity is a central challenge in neuroscience. Dynamic functional connectivity (dFC) allows the study of short-term temporal changes in functional connectivity (FC) between distributed brain areas. By clustering dFC matrices from resting-state fMRI, we previously described "brain patterns" that underlie different functional configurations of the brain at rest. The networks associated with these patterns have been extensively analyzed. However, the overall dynamic organization and how it relates to consciousness remains unclear. We hypothesized that deep learning networks would help to model this relationship. Recent studies have used low-dimensional variational autoencoders (VAE) to learn meaningful representations that can help explaining consciousness. Here, we investigated the complexity of selecting such a generative model to study brain dynamics, and extended the available methods for latent space characterization and modeling. Therefore, our contributions are threefold. First, compared with probabilistic principal component analysis and sparse VAE, we showed that the selected low-dimensional VAE exhibits balanced performance in reconstructing dFCs and classifying brain patterns. We then explored the organization of the obtained low-dimensional dFC latent representations. We showed how these representations stratify the dynamic organization of the brain patterns as well as the experimental conditions. Finally, we proposed to delve into the proposed brain computational model. We first applied a receptive field analysis to identify preferred directions in the latent space to move from one brain pattern to another. Then, an ablation study was achieved where we virtually inactivated specific brain areas. We demonstrated the model's efficiency in summarizing consciousness-specific information encoded in key inter-areal connections, as described in the global neuronal workspace theory of consciousness. The proposed framework advocates the possibility of developing an interpretable computational brain model of interest for disorders of consciousness, paving the way for a dynamic diagnostic support tool.

Fig. 1

Illustration of the proposed VAE-VIENT framework. A VAE learns 2D latent representations $z=(z{1}_i,z{2}_i)$ from dynamic functional connectivity matrices (dFCs), leading to (1) evaluation of the proposed model against other generative models implementing different latent dimensions, (2) exploration of latent space with the ability to view discrete or continuous representations (here we observe how brain patterns are organized in latent space), and (3) two simulation paradigms, including a receptive field analysis that generates tensor representations to study the effect of perturbing input dFCs, and an ablation study of Global Neuronal Workspace (GNW) connections to study the transition from wakefulness to unconsciousness.

Fig. 2

Illustration of the two steps involved in the receptive field analysis. (A) From a dFC matrix $X^{(i)}$, or equivalently its upper terms $x^{(i)}$, two perturbations are performed on connections b (blue) and r (red) by swapping one connection with the following three correlations $[-1,0,1]$ ($p=3$). (B) Corresponding $z_b^{(i)}$ and $z_r^{(i)}$ latent representations ($N=2$) follow lines and are summarized by their inclination angles with the x-axis ${\theta }_b$ and ${\theta }_r$, respectively.

Fig. 3

Brain pattern (BP) classification/reconstruction using VAE, PPCA, and sVAE models: (A) the Pearson correlation coefficient of BP-wise averaged dFCs with respect to the model parameters, (B) the balanced accuracy (BAcc) between the ground truth and the matched predicted label, (C) the proposed consensus metric $\mathcal{M}$, and (D) the Pearson correlation coefficient recorded for each BP. In plots A, B, and C, the selected ${\text{VAE}}_2$ is highlighted by a red bounding box. The dashed lines represent the trends obtained for each latent space dimension across the considered models.

Fig. 4

The brain patterns (BPs) from the k-means clustering and after reconstruction with the selected ${\beta }_{20}-VA{E}_2$ The distance between the original and reconstructed centroids is displayed, for various metrics. SSIM distance = 1 − SSIM; Correlation distance = 1 − Pearson correlation.

Fig. 5

Discrete stratification of the latent space of the selected VAE into a base of (A) Brain Patterns (BPs)—the centroids from a seven-class k-means clustering on the dFCs and (B) lifetimes - the time spent continuously in the corresponding brain pattern. Note that the obtained BP stratification mainly shows non-overlapping clusters as quantified in Appendix S3. For the lifetimes, we discretize the values into three categories: the 25% longest (in red), the 25% shortest (in blue), and all others medium (in pink).

Fig. 6

Continuous stratification of the latent space of the selected VAE and corresponding confidence and reliability maps: (A) continuous representation of the Brain Patterns (BPs), (B) decoded dFCs sampled using a regular $19\times 19$ grid in the latent space, (C) estimated confidence map $\mathcal{C}\mathcal{M}$, and (D) estimated reliability map $\mathcal{R}\mathcal{M}$. A, C, and D used a regular $200\times 200$ grid in the latent space.

Fig. 7

Results of RF analysis of the seven brain patterns and associated connections with a high potential for action. Using the proposed connection-wise RF analysis, a local perturbation model computed as an ellipse is derived at each encoded latent space location. Note that to improve readability, we scaled each ellipse. The associated mean diffusivity (MD) is calculated. For each ellipse, the twenty connections that cause the most displacement in the latent space are displayed using a circular layout. The connectivity plots were generated with MNE-Python: 1.6.0.

Fig. 8

Ablation study performed from the GNW nodes. We evaluate the performance of a trained SVM classifier in predicting the awake state using balance accuracy (BAcc). As input, we take only the raw or perturbed awake dFCs. We denote the corresponding prediction scores as BAcc (vertical red dot line) and $\widetilde{\mathit{BAcc}}$ (vertical blue dot line), respectively. We also display the histogram of ${\overline{\mathit{BAcc}}}^i$ when random connections are removed.