Functional mechanisms of probabilistic inference in feature- and space-based attentional systems

Abstract

Humans flexibly attend to features or locations and these processes are influenced by the probability of sensory events. We combined computational modeling of response times with fMRI to compare the functional correlates of (re-)orienting, and the modulation by probabilistic inference in spatial and feature-based attention systems. Twenty-four volunteers performed two task versions with spatial or color cues. Percentage of cue validity changed unpredictably. A hierarchical Bayesian model was used to derive trial-wise estimates of probability-dependent attention, entering the fMRI analysis as parametric regressors. Attentional orienting activated a dorsal frontoparietal network in both tasks, without significant parametric modulation. Spatially invalid trials activated a bilateral frontoparietal network and the precuneus, while invalid feature trials activated the left intraparietal sulcus (IPS). Probability-dependent attention modulated activity in the precuneus, left posterior IPS, middle occipital gyrus, and right temporoparietal junction for spatial attention, and in the left anterior IPS for feature-based and spatial attention. These findings provide novel insights into the generality and specificity of the functional basis of attentional control. They suggest that probabilistic inference can distinctively affect each attentional subsystem, but that there is an overlap in the left IPS, which responds to both spatial and feature-based expectancy violations.

Introduction

Prior information about the location or features of a stimulus facilitates its detection and speeds up response times (RTs). Conversely, violations of spatial or feature-based expectancies result in RT costs. These effects, as well as their neural underpinnings, can be investigated with probabilistic cueing paradigms in which a spatial or feature cue is presented prior to a behaviorally relevant target (Posner, 1980).

Orienting of attention in response to spatial or feature cues engages a shared frontoparietal network including the bilateral frontal eye fields (FEF), intraparietal sulcus (IPS), and inferior frontal cortex (IFC) (Egner et al., 2008, Giesbrecht et al., 2003, Liu et al., 2003, Schenkluhn et al., 2008, Slagter et al., 2007, Vandenberghe et al., 2001b, Wojciulik and Kanwisher, 1999). However, only spatial cues lead to a lateralized biasing of activity of visual areas (Egner et al., 2008).

Reorienting of attention to unexpected events in cueing paradigms is investigated by contrasting invalidly with validly cued targets. Spatially invalidly cued targets increase activity in ventral frontoparietal regions such as the temporoparietal junction (TPJ) and IFC, but also in dorsal frontoparietal regions such as the FEF and IPS (Corbetta et al., 2008, Corbetta and Shulman, 2011). Reorienting of attention to invalidity cued targets in the feature-based domain has rarely been investigated as yet, but one study has reported that activation in the left supramarginal gyrus (SMG), bilateral inferior frontal gyrus (IFG), medial frontal areas, and the cerebellum is enhanced during dimensional reorienting, i.e., when the target-defining dimension (orientation or color) in a visual search task is invalidly cued (Weidner et al., 2009). Additionally, it has been shown that shifts from color to motion, or vice versa, activate the left IPS, left precentral gyrus, the precuneus, and visual areas (Liu et al., 2003). A left hemispheric dominance has also been reported for object-based as compared to location-based spatial attention orienting (Arrington et al., 2000). However, the functional correlates of spatial and feature-based reorienting have so far not been directly compared within the same paradigm.

Importantly, the behavioral effects of orienting and reorienting in both attentional systems scale with the percentage of cue validity (%CV), i.e., the probability that the information provided by the cue is correct (Dombert et al., 2015, Egner et al., 2008, Vossel et al., 2006, Vossel et al., 2012). More specifically, response time differences between invalidly and validly cued targets increase with higher %CV. Most studies explicitly informed the subjects about the %CV; however, more recent work has shown that even without this explicit information, RTs are highly sensitive to unsignaled changes in %CV, suggesting that the subjects continuously infer the probability of the cue-target outcome in a given trial on the basis of observations in prior trials. This inference process can plausibly be described by approximate Bayes-optimal learning rules (Mathys et al., 2011, Vossel et al., 2014a). Though the update equations of this Bayesian model bear structural similarity to reinforcement learning models such as the Rescorla–Wagner rule (Rescorla and Wagner, 1972) in that the update of the probability estimate is the product of a learning rate and a prediction error, the learning rate in the Bayesian model is not fixed but governed by higher hierarchical levels. In our specific case, the update of the probability that the cue will be valid in a given trial depends on the trial-wise belief about the stability/volatility of the environment (highest hierarchical level) and on a subject-specific parameter. In other words, updating will be faster if the subject has learned that the environment is not stable. Such flexible models have been shown to provide a more plausible account of behavior than the Rescorla–Wagner rule, particularly in volatile environments where a fixed learning rate is suboptimal (Behrens et al., 2007, den Ouden et al., 2010, Vossel et al., 2014a). Another advantage of the current modeling approach is that it represents an individualized Bayes-optimality, allowing for a quantification and comparison of updating behavior in the two attentional systems. These parameters can be estimated on the basis of trial-wise RTs. This computational modeling of behavioral responses has been combined with functional magnetic resonance imaging (fMRI) (Vossel et al., 2015). It was observed that activity in the right FEF, TPJ, and the putamen during reorienting responses in a spatial cueing paradigm with saccadic responses to the targets was modulated by the trial-wise belief about cue validity. No significant effect of the belief about cue validity was found for orienting of attention.

Taken together, the functional mechanisms of orienting and reorienting of attention and the modulation of attentional deployment by inferred percentage of cue validity are well-characterized in the spatial attention system, but have so far not been studied for the cueing of target features such as color. Hence, it remains to be established whether reorienting, probabilistic inference, and their physiological implementations are universal across the two visual attentional systems—or whether they are domain-specific with differential functional correlates for spatial and feature-based attention. First behavioral observations from patients with right-hemispheric brain damage may point to differential functional mechanisms of the processing of statistical regularities (repetition priming) for locations and features: Shaqiri and Anderson (2012) reported that the speeding of RTs to the repeated presentation of a stimulus at the same location is disrupted after right-hemispheric stroke, while the RT benefit is still present for the repetition of stimulus color in these patients.

To address these outstanding issues, we applied the combined computational modeling—fMRI approach outlined above to two different versions of a probabilistic cueing task with spatial or color cues. While we also aimed at replicating the effects of attentional orienting in both domains with our modified paradigm, our specific emphasis was the characterization of reorienting after invalid cues, as well as its modulation by probability-dependent attention in the two systems. Based on Weidner et al. (2009) and Liu et al. (2003), we expected a stronger involvement of left parietal areas in feature-based reorienting. Due to the differential effects of spatial and non-spatial statistical regularities in stroke patients (Shaqiri and Anderson, 2012) and the results by Vossel et al. (2015), we hypothesized that attentional deployment by probabilistic inference involves right-hemispheric frontoparietal structures in the spatial attentional domain.