Towards understanding the nature of direct functional connectivity in visual brain network

TL;DR

The paper investigates how visual information is represented in the human brain by constructing visual brain networks (VBN) from fMRI time series using both marginal and partial correlations, and analyzing these networks with graph-theoretic features. By incorporating both positive and negative functional connectivity and evaluating image complexity–specific networks (COCO, ImageNet, SUN) via time-series segments, the authors demonstrate distinct global and local network properties and show that anti-correlations carry substantial, robust information. They apply XGBoost to classify VBNs by image dataset and achieve high accuracies, highlighting the discriminative power of local graph features and the central role of LOC in visual processing. The findings support the importance of including anti-correlated connectivity and direct connectivity estimates in understanding vision representation in the brain, providing a baseline framework for future, larger-scale neuroimaging studies.

Abstract

Recent advances in neuroimaging have enabled studies in functional connectivity (FC) of human brain, alongside investigation of the neuronal basis of cognition. One important FC study is the representation of vision in human brain. The release of publicly available dataset BOLD5000 has made it possible to study the brain dynamics during visual tasks in greater detail. In this paper, a comprehensive analysis of fMRI time series (TS) has been performed to explore different types of visual brain networks (VBN). The novelty of this work lies in (1) constructing VBN with consistently significant direct connectivity using both marginal and partial correlation, which is further analyzed using graph theoretic measures, (2) classification of VBNs as formed by image complexity-specific TS, using graphical features. In image complexity-specific VBN classification, XGBoost yields average accuracy in the range of 86.5% to 91.5% for positively correlated VBN, which is 2% greater than that using negative correlation. This result not only reflects the distinguishing graphical characteristics of each image complexity-specific VBN, but also highlights the importance of studying both positively correlated and negatively correlated VBN to understand the how differently brain functions while viewing different complexities of real-world images.

Figures (14)

• Figure 1: Block schematic of the proposed study.
• Figure 2: Sample images, taken from the three computer vision datasets having different complexities: Left: COCO (Red box)- contains multiple objects and actions, Middle: ImageNet (Green box)- contains single-focused object and Right: SUN (Blue box)- contains indoor and outdoor scenes.
• Figure 3: The MRI images in the figure (left) illustrates voxel activation within brain while viewing images, for all four subjects. Right: FMRI time series which is extracted from one active voxel is shown in Black for one representative subject. Image complexity-specific time series is obtained by splitting the whole time series according to the BOLD intensity values of distinct image complexities (COCO, ImageNet and SUN), as seen by the participants at each time stamp. The three time series representing COCO, ImageNet and SUN as obtained from the whole time series are shown in Red, Green and Blue color respectively.
• Figure 4: Construction of VBN from whole TS for one representative subject is shown. Figure (A) and (C) show graphical representation of networks with positive and negative edge strengths respectively. The respective mapping of network graph on fMRI image is shown in Figure (B) and (D).
• Figure 5: Marginal (left) and Partial correlation (right) as obtained for one representative subject. A lot of positive connectivity was seen to be either removed or lessened in strength in PC, as expected, since PC captures the direct linear association between network nodes after eliminating the spurious effects from all confounders.