Single-Trial fMRI Decoding of 3D Motion with Stereoscopic and Perspective Cues

Abstract

How does the brain process 3D motion? Here, we focused on the human motion complex (hMT+), extending insights from monkey studies. Using 3D-motion stimuli containing perspective and/or stereoscopic cues, we investigated the hierarchy within the motion complex in humans of both sexes to understand the neural mechanisms underlying motion perception. On each trial we decoded 3D motion direction (toward/away) based on the BOLD response in primary visual cortex (V1), and areas MT, MST, and FST within hMT+. We found that 3D-motion direction could be reliably decoded from all four areas, but with distinct patterns of cue preference. MT showed greatest accuracy with perspective cues, whereas FST showed greatest accuracy with stereoscopic cues. While motion direction could be decoded in V1 and MST, these results could be explained by retinotopic variation in the BOLD response that depended on motion direction. In contrast, MT and FST were less impacted by retinotopic biases in the BOLD response. We also identified significant behavioral differences between participants: some were proficient at using stereoscopic cues and others performed near chance. Good behavioral performance with stereoscopic cues was accompanied by better decoding performance in FST but not in MT. A control experiment that eliminated 3D-motion percepts for stereoscopic stimuli, but not perspective stimuli, revealed that unlike MT, decoding accuracy in FST was influenced by perceptual components of 3D motion. Our findings support that MT and FST play distinct roles in the analysis of visual motion and are key in the transformation of retinal input into perceptual report.

Significance statement Visual motion representations are elaborated hierarchically across distinct regions of the primate brain. In humans, the hMT+ complex contains multiple subdivisions including homologues of non-human primate (NHP) motion areas MT and MST. Using fMRI localizers, hMT+ was recently found to include a third subdivision consistent with NHP area FST. Here, we show that human FST and MT, like their NHP counterparts, are functionally distinguishable based on the representation of 3D motion. Most notably, we find a perceptual representation of 3D motion in human FST, but not MT, that is distinct from the patterns of motion found on the retinae. Our findings reveal that the human visual motion-processing network extends crucially beyond MT to represent complex, perceptual motion signals.