Decoding Brain Patterns and Navigation: MVPA and Scene Perception in fMRI

This MIT OCW lecture surveys how multivoxel pattern analysis in fMRI reveals information contained in brain activity patterns beyond magnitude alone. The instructor revisits Haxby's challenge to the idea that regions like the FFA and PPA are strictly selective for faces and places, showing that distinct patterns can carry information about nonpreferred stimuli. The talk then shifts to navigation, introducing beaconing and cognitive maps, Tolman style experiments, and the neural underpinnings in the parahippocampal place area, retrosplenial cortex, and related regions. The session covers causal approaches such as transcranial magnetic stimulation and intracranial stimulation, and discusses the limitations of MVPA and the complementary method of fMRI adaptation. Key takeaways include the idea that multiple scene related brain regions collaborate to support scene perception and navigation and that pattern information does not always translate to behavior.

Introduction to pattern analysis in the brain

The lecture introduces MVPA as a tool to ask not just which regions respond to stimuli, but what information those regions encode. It uses the FFA and PPA as focal points for examining whether selective regions truly specialize in a category or carry information about nonpreferred stimuli as well.

Haxby critique and empirical tests

Haxby argued that selective regions may still contain discriminative information about nonpreferred stimuli. The instructor reviews data that address this claim, including data from TMS, intracranial stimulation, and lesion studies, which point toward a causal link between region activity and perception for certain categories, while pattern information can be epiphenomenal in other contexts.

Decoding patterns in the PPA and beyond

We localize the parahippocampal place area and ask whether the pattern of activity in this region can distinguish beach scenes from city scenes. A four-pattern design across runs demonstrates the stability and separability of the beach and city patterns, illustrating the core idea behind decoding approaches in neuroscience.

Navigation and the two big questions

The talk broadens to navigation, outlining two core problems: where am I and how do I get from here to there. Beaconing provides direct paths to goals, while cognitive maps enable flexible navigation when direct cues are unavailable. Tolman-like experiments with rats motivate the concept of cognitive maps and the idea that internal representations guide orientation and route planning.

Neural circuits for scenes and navigation

The PPA, retrosplenial cortex, and other scene selective regions are discussed as parts of a broader network supporting navigation. The talk covers how familiarity, heading direction, and spatial layout information contribute to the activity in these regions, with evidence from carefully controlled experiments and causal manipulations such as transcranial stimulation and intracranial recordings.

MVPA limits and adaptation

While MVPA is powerful, it does not always reveal the underlying neural code, especially when neuronal populations are intermingled. The lecture introduces event related fMRI adaptation as a method to probe whether a region encodes sameness vs difference across stimuli, offering a complementary window into the brain's coding strategies.

Takeaways and future directions

The session emphasizes that scene perception and navigation rely on a distributed system with multiple regions contributing distinct roles. It also highlights the importance of combining methods to infer causality and representation in the living brain.