^{1}

^{2}

^{*}

^{3}

^{4}

^{1}

^{2}

^{3}

^{4}

Edited by: Xiaolin Zhou, Peking University, China

Reviewed by: Gennady Knyazev, Institute of Physiology and Basic Medicine, Russia; Xiaoang Wan, Tsinghua University, China

This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

This study investigates the dynamics of attention during continuous, naturalistic interactions in a video game. Specifically, the effect of repeated distraction on a continuous primary task is related to a functional model of network connectivity. We introduce the Non-linear Attentional Saturation Hypothesis (NASH), which predicts that effective connectivity within attentional networks increases non-linearly with decreasing distraction over time, and exhibits dampening at critical parameter values. Functional magnetic resonance imaging (fMRI) data collected using a naturalistic behavioral paradigm coupled with an interactive video game is used to test the hypothesis. As predicted, connectivity in pre-defined regions corresponding to attentional networks increases as distraction decreases. Moreover, the functional relationship between connectivity and distraction is convex, that is, network connectivity somewhat increases as distraction decreases during the continuous primary task, however, connectivity increases considerably as distraction falls below critical levels. This result characterizes the non-linear pattern of connectivity within attentional networks, particularly with respect to their dynamics during behavior. These results are also summarized in the form of a network structure analysis, which underscores the role of various nodes in regulating the global network state. In conclusion, we situate the implications of this research in the context of cognitive complexity and an emerging theory of flow during media exposure.

All visual and auditory stimuli are mediated in some way by attentional processing. Biologically, attention serves both as a general alertness mechanism and as a specific resource allocation mechanism (

We investigated the neural dynamics of network connectivity for attention while participants are engaged in a continuous activity while undergoing functional magnetic resonance imaging (fMRI). Participants played a first-person shooter video game (

Our experimental task can be broken down into two sub-components, each with unique properties. One stimulus (the video game) requires continuous attention, while the other stimulus (the laser light distractor) serves as a means to disrupt attention at random intervals. Within a limited capacity of attention framework (

Relationship between Connectivity and Distraction. Hypothesized relationship between connectivity in attentional networks and distraction from a primary task.

In this manuscript, we focus on connectivity patterns within attentional networks (

The first premise is that cognitive functions are regulated by interconnected brain structures (

In general, FBNs require interaction between multiple distributed brain regions (

In opposition to linear models of attentional capacity, the second premise argues that connectivity exhibits a robust response to distraction (laser light presentations) in the form of large distracter parameter values. One common feature of robust phenomena across various types of networks (for examples, see

Attentional networks operate in a dynamic fashion, with functional regulation occurring on multiple time scales. This facilitates adaptation to new conditions, produces non-linear connectivity patterns between network structures, and allows brain networks that operate one way under normal conditions to acquire a new (or modified) functional state during disruption (

In terms of experimental design, we expect that the network components will shift from a linear response to a non-linear response with increasing amounts of distraction. Expertise in so-called “action” video games results from training, which has several augmentative effects on attentional capacity. These include rapid switching between tasks, enhanced acuity with respect to the visual field perimeter, increased speed of processing, and greater cognitive control (

In our study, we assume that continuous brain dynamics can be better understood using a continuous stimulus and a naturalistic experimental paradigm (

For this study, temporally-specific information regarding a single attentive episode will be inferred from the attention models originally presented by

To test for the hypothesized non-linearity in the relationship between distraction and connectivity, we use a non-linear, quadratic form of a general linear model to explain attentional functional connectivity. The NASH is expressed as three related statistical hypotheses related to Eq. (3). The first of these predicts that connectivity between pre-defined regions depends on the level of distraction as defined by the laser light stimulus:

H_{1}: Connectivity depends on distraction, c ≠ 0

For the attentional network components unrelated to sensorimotor coordination, our general hypothesis not only predicts the existence of a dependency, but also specifically a reduction of connectivity. However, considering that the distractor task was a left-handed button-press task and involved sensorimotor coordination (sensing, planning, and execution of a button-press) we do not expect this reduction to occur in networks related to sensorimotor coordination (networks functionally connected to the cerebellum). We therefore expect that connectivity should decrease with increasing distraction for networks unrelated to sensorimotor coordination: c < 0 Analogously, we expect for networks that also contribute to sensorimotor coordination a positive relationship between connectivity and distraction, so that connectivity should increase with increasing distraction for networks related to sensorimotor coordination: c > 0.

Moreover, we can make predictions on non-linear behavior of the distraction-connectivity relationship. The NASH implies that increases in connectivity accelerate when distraction falls below a threshold

H_{2}: The curvature of an “increasing distraction-decreasing connectivity” relation is convex, i.e. if c < 0 then d > 0.

Correspondingly, for attentional network components related to sensorimotor coordination, i.e. for networks for which we expect an increase of connectivity with increasing distraction, we predict:

H_{3}: The curvature of an “increasing distraction-increasing connectivity” relation is concave, i.e. if c > 0 then d < 0.

As we can see from H_{2} and H3, it is predicted that linear and non-linear coefficients will exhibit opposite signs, sign(_{1}, H_{2}, and H_{3}) were tested with second level statistics across the group. Calculations were performed in Matlab after a standard preprocessing pipeline for fMRI data (

Different aspects of the data had been evaluated previously in (

Thirteen male volunteers (age 18–26, median 23) were recruited on the basis of previous experience with video games (15.1 ± 9.0 h/week) with ads posted at the local university and in video game stores. Inclusion criteria were: male, age between 18 and 26 years, playing at least 5 h weekly of video games, and right-handedness. Individuals with contraindication against MR investigations, acute or anamnesis of major neurological, psychiatric, or ophthalmologic disorders were excluded. All participants gave their written informed consent and the local ethics committee approved the study protocol. The study protocol was approved by the ethics committee of the University of Tübingen, Germany.

The video game used in this experiment is a first person shooter simulation called

As subjects interacted with the continuous primary stimulus, a red laser projected a light point into the periphery of the visual field (right upper quadrant) at random time intervals until participants responded. Delays in laser light presentations followed a Poisson distribution with an average time delay of 10 s (I_{p}) after the last button press. The Poisson distribution was chosen to ensure equal probabilities for laser light presentations at any moment. The time intervals were independent from changes in the primary task, but required the subject to respond in the fastest possible time by pressing a button with the left hand. Pressing the response button reset the timer on the laser light and initiated another trial. As the secondary distractor task was performed inside the MR scanner during ongoing game play, it served as an incongruent stimulus relative to the main action in the simulation. The action in the video game itself is the primary task, and required regulation by the executive attention network. The mean time interval between laser light presentations (I_{p}) and the mean response time to each presentation of the laser light (I_{r}) was used to calculate the distraction parameter

The distraction parameter (_{p}) multiplied by the mean response time to each presentation of the laser light (_{r}):

As such, distraction is defined as the various demands on attentional capacity throughout the course of the task. In other words, distraction is measured as the inverse of the time between events multiplied by the time needed for a response. The more laser light presentations (_{p}_{r}

For this study, fMRI was conducted at a magnetic field strength of 3 Tesla (Magnetom TRIO, Siemens, Erlangen, Germany). Multi-echo single-shot echo-planar imaging (EPI; echo times = 23, 40, and 62 ms) with dynamic distortion correction (

The ROI analyses that are reported here rely on

In the simplest of terms, connectivity can be understood as the correlation (or statistical dependency) between two neural time-series. This is known as functional connectivity (

We use a model of Psycho-Physiological Interactions (PPI) to characterize effective connectivity. The PPI model (

where β_{1} represents the parameter estimate for the psychological variable main effect, β_{2} is the parameter estimate for the physiological variable main effect, β_{3} represents a parameter estimate for the interaction term, _{sy} is the psychological variable of interest (e.g., attention), and _{hy} is the physiological variable of interest (e.g., neural time series within a given ROI), and

For our analysis we choose ROIs _{T}) with that of the source region (ROI_{S}), the distractor parameter

The General Linear Model in our analysis is similar to the general PPI model (Eq. 3):

Since this is a non-linear equation, we assume theicients from the model were used as a measure of connectivity between brain ROIs.

The data featured in this study is publically available in the Open Science Framework (OSF) in the form of two CSV files containing all extracted ROIs and our distraction measures (one file for each hemisphere). This data has been used to estimate the model parameters reported in this article. The URL of this repository is

The success of our study rests on participants treating the video game as a primary task, and the distraction measure as a secondary task (

The behavioral distractor task was a light point projected by a red laser requiring a speeded button-press response of the left hand. The mean response time (_{r}) was 1158.3 ms (90% interval, 5th to 95th percentile: [434.7, 16739.6]) with slower responses after longer Inter-Stimulus Intervals (ISIs – ^{-2} (90% interval, 5th to 95th percentile: [2.2, 212.8] ms^{-2}). A 90% interval for the _{r} and derived distraction measures provides a better insight into the hyperbolic nature of its distribution with respect to response times within and between participants.

Distribution of the distractor parameter

Linear attentional networks. Nine regions of interest representing

Non-linear attentional networks. Nine regions of interest representing

In Hypothesis 1 (H_{1}) we predicted reduced connectivity with increasing distraction for attentional network components unrelated to sensorimotor coordination. This pattern did indeed emerge for most local network connections, particularly between frontal areas and thalamo-frontal projections (red arrows in _{1} (unrelated to sensorimotor coordination for local connections and related to sensorimotor networks for long-range connections).

Group means of linear connectivity coefficients (

Thal | SFG | IFGs | FFGm | IFGi | Cere | MFG | FFGl | ACC | |
---|---|---|---|---|---|---|---|---|---|

Thal | –1.14** | –0.33 | 0.01 | –0.05 | –0.57 | –2.34** | 0.59 | –0.20 | |

SFG | –0.51 | –0.26 | 0.13 | –0.38 | 0.50 | 1.14* | –0.95 | –0.22 | |

IFGs | –0.55 | –0.06 | –0.28 | –0.64** | 1.03*** | –0.38 | 0.73 | –0.41* | |

FFGm | 0.31 | –0.81** | –0.38 | 0.83* | –0.35 | –1.65 | 1.11 | –0.28 | |

IFGi | 0.06 | –0.89* | –1.38*** | 0.37 | 0.55 | –2.51* | 0.97 | –0.33 | |

Cere | –0.26 | 1.04*** | 1.79*** | –0.22 | 0.67 | 2.41*** | –1.1 | 0.41 | |

MFG | –0.09 | 0.24 | 0.22 | –0.01 | –0.22 | 0.23 | –0.38 | –0.06 | |

FFGl | 0.24 | –0.51 | 0.66 | 0.02 | 0.49 | –0.71* | –0.40 | 0.64** | |

ACC | 0.03 | –0.73 | –1.66** | –0.12 | –0.13 | 0.67* | –2.20** | 1.67 |

^{∗}D + d

^{∗}D

^{2})

^{∗}Thal =

^{∗}D + 2.64

^{∗}D

^{2})

^{∗}Thal (for d coefficient, see

_{2}: sign(d)

^{∗}

^{∗∗}

^{∗∗∗}

We confirm that H_{2} predicts that the curvature of a convex “increasing distraction-decreasing connectivity” relation (_{3} predicts that the curvature of an “increasing distraction-increasing connectivity” relation is concave (

Estimations of the distraction-connectivity function. Inter-frontal connectivity falls off but levels out with increasing distraction

We also tested our results for consistency in the left hemisphere. The analysis replicated the right-hemispheric pattern with generally reduced effect sizes. Both the linear (

Another way to further quantify network connectivity for the significant linear and non-linear connections is shown in ^{2}-value close to 0, while hierarchically-structured sets of connection should yield a ^{2}-value approaching 1.

Group means of non-linear connectivity coefficients (

RThal | RSFG | RIFGs | RFFGm | RIFGi | RCere | RMFG | RFFGl | RACC | |
---|---|---|---|---|---|---|---|---|---|

RThal | 2.64** | 0.62 | 0.62 | 0.62 | 2.11 | 7.61** | 0.61 | 0.73 | |

RSFG | 0.41 | –0.89 | –1.38 | 0.19 | –3.51 | –5.50* | 2.89 | 0.37 | |

RIFGs | 0.50 | –0.36 | –0.21 | 1.41* | –2.43** | 5.17 | –2.61 | 0.54 | |

RFFGm | –1.14 | 2.60* | 0.17 | –2.05* | 3.86 | 5.21* | –2.25 | 0.33 | |

RIFGi | –0.54 | 2.96 | 4.81*** | 0.07 | 1.33 | 11.59* | –3.47 | 0.66 | |

RCere | –0.05 | –3.53*** | –5.57*** | 1.32 | –2.61 | –10.30*** | 1.94 | –0.93 | |

RMFG | –0.29 | –0.20 | –0.12 | –0.24 | 0.94 | –1.57 | 0.60 | 0.27 | |

RFFGl | –1.14 | 2.52** | –1.99 | –0.65 | –3.30 | 2.61* | 1.79 | –3.04** | |

RACC | 0.54 | 2.26 | 4.09*** | 0.37 | 0.48 | 1.43 | 11.57** | –4.66 |

^{∗}D + d

^{∗}D

^{2})

^{∗}Thal =

^{∗}D +2.64

^{∗}D

^{2})

^{∗}Thal (for c coefficient, see

_{2}: sign(d)

^{∗}

^{∗∗}

^{∗∗∗}

Results of the network structure analysis.

Linear |
Non-linear |
||||||
---|---|---|---|---|---|---|---|

Inbound | Outbound | Total | Inbound | Outbound | Total | ||

Cere | 3 | 3 | 6 | RCere | 2 | 3 | 5 |

FFGl | 0 | 2 | 2 | RFFGl | 0 | 3 | 3 |

FFGm | 0 | 2 | 2 | RFFGm | 0 | 3 | 3 |

MFG | 5 | 0 | 5 | RMFG | 6 | 0 | 6 |

Thal | 0 | 2 | 2 | RThal | 0 | 2 | 2 |

SFG | 4 | 1 | 5 | RSFG | 4 | 1 | 5 |

ACC | 2 | 3 | 5 | RACC | 1 | 2 | 3 |

IFGs | 3 | 2 | 5 | RIFGs | 3 | 2 | 5 |

IFGi | 2 | 3 | 5 | RIFGi | 2 | 2 | 4 |

Edges | – | – | 19 | Edges | – | – | 18 |

^{2} |
0.74 | 0.42 | 0.72 | R^{2} |
0.87 | 0.70 | 0.88 |

– | – | 3 | – | – | 4 | ||

– | – | 0.21 | – | – | 0.23 |

^{2}are located in

For the directed linear graph (shown in ^{2}-value of 0.72 (^{2}-value is 0.74 (^{2}-value is 0.42 (^{2}-value of 0.88 (^{2}-value for inbound nodes only is 0.87 (^{2}-value for outbound nodes only is 0.70 (

A comparison of linear components (

The local frontal and thalamo-frontal connections which showed a decrease in connectivity (

Importantly, all commonly emerging pathways showed opposite signs for linear increase versus curvature. In other words, the increase or decrease of connectivity due to distraction was limited by the non-linear term, and was thus dependent on the level of distraction. The independent graphs in _{2}), while _{3}).

In this study we have presented both a rationale and technique that permits us to investigate the dynamics of attention in a complex, immersive environment (a video game) by mapping psychophysiological responses to an attentional network (

As predicted by the NASH, the relationship between network connectivity and distraction is non-linear and convex. While we concede that the primary experimental task (playing a video game) might also be activating alerting and spatial orientation networks to some degree, one should keep in mind that first-person video games are designed to fully capture alertness and orientation in any moment (especially when played in an experimental setting under continuous observation). This means that either there is insufficient variation of activity in attentional networks given the nature of our primary task, or that even a slight distraction from the primary task might lead to complete disruption within alerting and orienting networks.

Other secondary distractor tasks, such as asking participants to execute simple repetitive actions simultaneously to the primary task, might be more suitable for studying those networks (

The convex and concave relationships in _{2}), _{3}), respectively, provide a dynamic view of how distraction can affect attentional networks. We can also understand the effect of distraction on attentional processing in neuropsychological terms. In particular, distraction tends to play a much more complex modulatory role with respect to the attentional network. While previous studies have not accounted for the effects of varying degrees of distraction on attention, the broader mechanisms have been identified. In studies of pain perception (

When distractor processing has an effect on attentional processing, the effects are heterogeneous with respect to various parts of the attentional network. For example, high cognitive load experienced in the frontal regions of the attentional network can increase distractor processing, while high amounts of cognitive load in other regions can decrease distractor processing (_{2} means that a breakdown of attention is equivalent to increased behavioral distraction. While this relationship is linear for normal levels of distraction, H_{3} predicts that a robustness mechanism may also contribute to limited attentional resources for very high levels of distraction. Thus,

Given this context, we can say that _{2} (increasing distraction, decreasing connectivity), the ROIs demonstrate an ability to work independently. These centers tend to be in the frontal areas of the brain, which is consistent with the notion of distraction processing. Connections between ROIs consistent with H_{3} (increasing distraction, increasing connectivity) involve centers that require interdependence as cognitive processing is assisted through offloading. While network statistics suggest that this effect is small, changes in the demands of cognitive processing result in regions with a greater number of connections in the linear case becoming slightly more connected. Meanwhile, regions with fewer connections in the non-linear case become relatively less connected with an emphasis on retaining outbound (directed) connections (

Having found support for NASH, it is worth recalling that the hypothesis is based on two central premises. The first premise suggested that cognitive functions are regulated by connected brain structures. Since we have shown that connectivity in attentional networks decreases non-linearly for a certain range of increasing

The second premise suggested that the relationship between distraction and attentional network connectivity exhibits a non-linearity that demonstrates a robust response at a critical threshold value. Our combination of naturalistic behavior, short repetition time, and presence of noise in the form of our distractor task (for use in perceptual systems, see

There is still much to learn about brain dynamics and complex cognition in real-world environments (

The approach presented here contributes to this inquiry by simulating real-world behaviors in an interactive virtual environment and developing advanced metrics for the analysis of cognitive dynamics. This research can also inform emerging communication and media theories, which at their core are dependent upon advances in understanding attentional network dynamics. As media such as video games and virtual reality become increasingly immersive, ubiquitous, and continually stimulative, we require an understanding of attentional networks at the level of first-principles. The Synchronization Theory of Flow (

RW and KM: experimental design and execution. RW, BA, and KM: concept and analysis. RW, BA, and RH: manuscript writing. RW and RH: figures and open dataset. RW, BA, RH, and KM: methods and supplemental materials.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

The Supplementary Material for this article can be found online at:

Supporting Methods. Methodological detail, including mathematical and descriptive details for measures and statistical models (PDF File).