Causal Inference of Body Ownership in the Posterior Parietal Cortex

Abstract

How do we come to sense that a hand in view belongs to our own body or not? Previous studies have suggested that the integration of vision and somatosensation in the frontoparietal areas plays a critical role in the sense of body ownership (i.e., the multisensory perception of limbs and body parts as our own). However, little is known about how these areas implement the multisensory integration process at the computational level and whether activity predicts illusion elicitation in individual participants on a trial-by-trial basis. To address these questions, we used functional magnetic resonance imaging and a rubber hand illusion-detection task and fitted the registered neural responses to a Bayesian causal inference model of body ownership. Thirty healthy human participants (male and female) performed 12 s trials with varying degrees of asynchronously delivered visual and tactile stimuli of a rubber hand (in view) and a (hidden) real hand. After the 12 s period, participants had to judge whether the rubber hand felt like their own. As hypothesized, activity in the premotor and posterior parietal cortices was related to illusion elicitation at the level of individual participants and trials. Importantly, activity in the posterior parietal cortex fit the predicted probability of illusion emergence of the Bayesian causal inference model based on each participant's behavioral response profile. Our findings suggest an important role for the posterior parietal cortex in implementing Bayesian causal inference of body ownership and reveal how trial-by-trial variations in neural signatures of multisensory integration relate to the elicitation of the rubber hand illusion.SIGNIFICANCE STATEMENT How does the brain create a coherent perceptual experience of one's own body based on information from the different senses? We examined how the likelihood of eliciting a classical bodily illusion that depends on vision and touch-the rubber hand illusion-is related to neural activity measured by functional magnetic resonance imaging. We found that trial-by-trial variations in the neural signal in the posterior parietal cortex, a well known center for sensory integration, fitted a statistical function that describes how likely it is that the brain infers that a rubber hand is one's own given the available visual and tactile evidence. Thus, probabilistic analysis of sensory information in the parietal lobe underlies our unitary sense of bodily self.

Keywords: Bayesian causal inference; body perception; functional MRI; multisensory integration; neuroimaging; psychophysics; rubber hand illusion.

Figure 1.

Experimental setup and procedure. A, The participant's right arm and hand lay next to their body in a relaxed position, palm down, on a flat support, tilted upward (∼20°), and a robot arm applied tactile stimuli (taps) to the index fingers of the rubber hand in the video and to the participant's real hand during the experiment. B, A 3D video that was prerecorded was used as a visual stimulus in the experiment showing the rubber hand (∼40°) being touched by the same MR-compatible robot that was used to touch the participants' real hands during the experiment. C, All trials followed the same sequence: after the presentation of a fixation cross for 1 s, participants saw the rubber hand being touched in the head-mounted display while their hand was touched by the robot, synchronously or not. This 12 s visuotactile stimulation was followed by a 2 s display of the question “[did the rubber hand] feel like [it was] your [own] hand?” to which participants answered “yes” or “no” by pressing the corresponding key with their left hands. This schematic representation of the procedure shows an example of a sequence of five consecutive trials with five of the seven different asynchrony conditions (−300, −500, +150, 0, and +300 ms; the + 500 and –150 ms conditions are not shown in this example). D, The collected yes/no judgments (here, a theoretical example) were used in an fMRI analysis to define a regressor at the first level for the trials eliciting the RHI (represented by the top boxcar) and one for the trials not eliciting any illusion (represented by the bottom boxcar). E, These participants' answers were also fitted individually in our BCI model to estimate the probability of emergence of the RHI for a given asynchrony for each participant. These estimates were used to build a parametric modulation regressor at the first level to test for brain regions showing a relationship between the predictions of the BCI model and the strength of neural response across the different asynchronies tested.

Figure 2.

RHI elicited under different levels of asynchrony. A, The black dots represent the reported proportion of rubber hand illusion detection (i.e., responding “yes” to the statement “[did the rubber hand] felt like [it was] your [own] hand”; mean ± SEM) for each of the seven asynchrony conditions (−500, −300, 0, +150, +300, and +500 ms). In the synchronous condition, the participants reported perceiving the rubber hand like their own hand in 84 ± 4% (mean ± SEM) of the 20 trials when the visual and tactile stimulations were presented simultaneously (no asynchrony). Moreover, for every participant, increasing the asynchrony between the seen and felt touches decreased the prevalence of the illusion in a graded fashion: when the rubber hand was touched 500 ms before or after the real hand was touched, the illusion was reported only in 22 ± 5% and 16 ± 5% of the 20 trials, respectively. B, Repartition of the trials in which the RHI was detected by asynchrony conditions (color coded). For example, synchronous visuotactile stimulation (0 ms condition) accounted for 23% of illusion detections, and consequently, 77% of the “yes” trial responses occurred following stimulation with varying degrees of asynchrony. C, Repartition of the “no” trials when the participants judged that the RHI had not been experienced (responding “no” to the statement above). Synchronous visuotactile stimulation (0 ms condition) accounted for 5% of the unsuccessful RHI fixations across all trials, while trials with maximum asynchrony (±500 ms) accounted for 48% of the total number of “no” trials across all conditions.

Figure 3.

Observed and predicted probability of the emergence of the RHI. Top left corner, Mean observed probability of the emergence of the RHI (percentage “yes” judgments; x-axis) plotted against the probability of the emergence of the RHI predicted by the BCI model (y-axis) for the seven different asynchrony conditions (black dots; color-coded conditions). S1–S30, The proportion of RHI elicitations reported by each of the 30 individual participants (x-axes) for each level of asynchrony (y-axes) is shown (black dots), as is the distribution predicted by the BCI model (red curve; Extended Data Table 1-1, information about the estimated model parameters).

Figure 4.

Activations related to RHI detection. Increased BOLD signal when contrasting trials in which visuotactile stimulation led to participants answering “yes” to the illusion question (did the rubber hand felt like it was your own hand?) compared with stimulations for which participants answered “no” to this question across all levels of asynchrony/synchrony. For display purposes only, the activation map is displayed at a threshold of p < 0.001 (uncorrected for multiple comparisons, cluster threshold: 10 voxels), projected on a single-subject T1 MNI template (Fig. 6, presentation on the participants' mean structural MRI). The six highlighted activations were all significant (p < 0.05) after correction for multiple comparisons. Areas circled in orange survived whole-brain correction, and areas circled in blue survived small-volume correction based on a priori anatomic hypotheses (Table 1). The right panels show the BOLD signal (contrast estimates extracted from a sphere with a 5 mm radius center on the peak activation) from the six regions in question for the “yes” trials (red) and “no” trials (gray; compared with the baseline) to illustrate the effect sizes for purely descriptive purposes. LOC, Lateral occipital cortex.

Figure 5.

Activity in the PPC reflects individual BCI model predictions. The level of activity in the PPC is positively linearly related to the probability of the emergence of the rubber hand illusion as predicted by our BCI model, as observed in the parametrical modulation analysis. A, B, Two significant peaks of activation are displayed (p < 0.05, WBcorr), one located in the left angular gyrus (x = −40, y = −76, z = 24; A) and one in the left IPS (x = −18, y = −66, z = 50; B). For display purposes only, the activation map is displayed at a threshold of p < 0.001 (uncorrected for multiple comparisons; cluster threshold, 10 voxels), projected on a single-subject T1 MNI template (Fig. 6, presentation on the participants' mean structural MRI). The plots display the mean BOLD signal level (±SEM; blue dots and axis) in the respective region (left angular gyrus, left plot; left IPS, right plot) and mean BCI model prediction (orange shape and axis) as a function of visuotactile asynchrony. (Note that these mean BCI model plots across the whole sample are for illustration purposes only; the analysis was conducted at the first level with a parametric modulator specific to each participant's individual behavioral profile. See Materials and Methods for details.)

Figure 6.

A, B, Activations related to RHI detection (A) and predictions of the BCI model (B) are presented on a mean T1-weighted MRI from the current group of participants for more precise anatomic localization. For information about the contrasts and the statistical thresholds used for the activation maps, see Figure 4 for A and Figure 5 for B.