Epstein Lab Research Publications People Links Center for Cognitive Neuroscience University of Pennsylvania


Russell A. Epstein, PhD
Professor of Psychology
University of Pennsylvania
t:  215.573.3532
f:  215.898.1982

Laboratory Address:
Center for Cognitive Neuroscience
Goddard Labs (3rd and 5th floors)
3710 Hamilton Walk
Philadelphia, PA 19104-6241

Lab Twitter

Research Questions

Our research focuses on two interrelated questions:
(1) How do people perceive and recognize real-world visual scenes?
(2) How do people build up representations of their spatial environment in order to navigate from place to place?

Scene Recognition
Previous work from our laboratory has identified specific regions of the brain that respond preferentially to real-world visual scenes, including the Parahippocampal Place Area (PPA) and Retrosplenial Complex (RSC). We are currently focusing on understanding how scenes are represented in these regions and how scene representations might be built up from simpler components.

Spatial Navigation
Like other animals, humans rely on representations of the large-scale spatial structure of the world in order to navigate successfully from place to place. We are exploring the neural systems that support representations of large-scale, navigable spaces, such as a city or a college campus. We are especially interested in understanding how people represent routes, landmarks, and cognitive maps.

Adapted from: Epstein, R.A. (2008). Trends in Cognitive Sciences, 12: 388-396.
Response properties of PPA and RSC.
The Parahippocampal Place Area (PPA) and Retrosplenial Complex (RSC) are shown on a reference brain with voxels showing greater response to scenes than to objects indicated in orange. Bar charts show the fMRI response in the PPA and RSC to six stimulus categories, plotted as percent signal change relative to a fixation (no-stimulation) baseline.
Adapted from: Epstein, R.A. (2008). Trends in Cognitive Sciences, 12: 388-396.
The PPA and scene encoding.
The Parahippocampal Place Area (PPA) responds referentially to images of places (like landscapes and cityscapes). The PPA is critically involved in the encoding and recognition of real-world visual scenes.
(A) A patient with right hemisphere parahippocampal damage was unable to recognize scenes unless they contained a single prominent object-like landmark, indicating that patients with PPA damage can recognize large objects but not scenes per se.
(B) Under the spatial layout hypothesis, the PPA encodes the overall spatial layout of the scene as defined by fixed topographical features. In contrast, information about the individual objects in the scene are encoded by other brain regions. When the PPA is damaged, patients report that they can see the objects in the scene but the overall organization of the scene is lost.
(C) The PPA responds strongly to layout-defining scene features such as walls and ground planes. Indeed, the PPA response to indoor scenes is not significantly changed by removing the objects from the scene, but is significantly reduced by removing the background elements. Furthermore, PPA response to the Lego 'scenes' is significantly greater than response to Lego 'objects', demonstrating that PPA responds strongly to artificial stimuli if they have scene-like geometry.
Adapted from: Epstein, R.A. (2008). Trends in Cognitive Sciences, 12: 388-396.
The RSC and retrieval of long-term spatial knowledge.
The Retrosplenial Complex (RSC) also responds preferentially to images of places (like landscapes and cityscapes). In contrast to the PPA, which is primarily involved in representing the local scene, RSC appears to represent spatial information that extends beyond the visible horizon.
(A) University of Pennsylvania students were scanned with fMRI while viewing photographs of their home campus (Penn), an unfamiliar campus, or nonscene objects. For the home campus photographs, they either retrieved spatial information (location or orientation) about each scene or made a simple familiarity judgment.
(B) RSC response was higher when viewing images of the familiar campus than when viewing images of the unfamiliar campus. Furthermore, RSC response was stronger when subjects made location or orientation judgments about the familiar locations than when they made simple familiarity judgments. Thus, RSC activity is enhanced when viewing scenes whose locations are known, and is even further enhanced when subjects explicitly retrieve spatial information about those locations. In contrast, PPA responded equally in all scene conditions, consistent with a role in local scene perception.
Adapted from: Epstein, R.A.& Morgan, L.K. (2012). Neuropsychologia, 50(4): 530-543.
Representation of scene categories and landmarks in the human brain.
Multi-voxel pattern analysis (MVPA) is a method of fMRI data analysis used to uncover the representational distinctions supported by a brain region (i.e. which items a region treats as identical and which it treats as distinct).
(A) Examples of the 10 outdoor scene categories and 10 Penn landmarks displayed during an experiment. MVPA was used to identify regions that carry information about the category of the scene or the identity of the landmark being viewed.
(B) MVPA searchlight analysis revealed a wide swath of territory in occipito-temporal-parietal cortex for which multi-voxel activity patterns conveyed information about scene category (left) or landmark identity (right).
Adapted from: MacEvoy, S.P. & Epstein, R.A. (2011). Nature Neuroscience, 14: 1323-1329.
Lateral occipital (LO) cortex represents objects within scenes.
Lateral occipital (LO) cortex represents objects within scenes. Unlike PPA and RSC, which respond maximally to scenes, lateral occipital (LO) cortex responds maximally to objects. While numerous lines of evidence implicate this region in object recognition, recent findings suggest that LO may play a role in scene recognition as well, by extracting information about the objects within the scene.
(A) Multi-voxel patterns evoked by bathrooms, kitchens, intersections, and playgrounds were compared to synthetic "scene" patterns that were the average of the patterns evoked by two single objects. Activity maps shown are scene-evoked patterns (top) and synthetic "scene" patterns (bottom) for one subject.
(B) Scene classification accuracy using the object-based predictors was significantly above chance in LO in three experiments (E1, E2 and E3, respectively). Accuracy was not above chance in any experiment in either pF or PPA. Thus, LO appears to construct scene patterns from the constituent object patterns.
Adapted from: Schinazi, V.R. & Epstein, R.A. (2010). NeuroImage, 53(2): 725-735.
Study area for real-world route learning experiment.
Learning a route through a large-scale environment involves learning landmarks located at critical points along the route. To investigate the neural mechanisms supporting this ability, participants were lead along a 3.8 km route around the Penn campus and surrounding neighborhood before being scanned with fMRI. A total of 180 buildings were located directly along the route, with 85 at navigational decision points (yellow) and 95 at non-decision point locations (blue).
Adapted from: Schinazi, V.R. & Epstein, R.A. (2010). NeuroImage, 53(2): 725-735.
Results for real-world route learning experiment.
Buildings located at navigational decision points (i.e. intersections) along a newly-learned route activated a wide network of regions more strongly than buildings located at other points along the route (top); these regions included the PPA and RSC (bottom).
Adapted from: Morgan, L.K., MacEvoy, S.P., Aguirre, G.K. & Epstein, R.A. (2011). Journal of Neuroscience, 31(4): 1238-1245.
Neural representation of landmarks in the human brain.
(A) Examples of stimuli and map showing the locations of the 10 landmarks on the University of Pennsylvania campus. Twenty-two distinct photographs were taken of each landmark.
(B) Whole-brain (searchlight) analysis. Regions in which landmark identity could be reliably decoded from multivoxel response patterns are plotted on an inflated version of the cortex. These regions included the PPA and RSC.
Adapted from: Morgan, L.K., MacEvoy, S.P., Aguirre, G.K. & Epstein, R.A. (2011). Journal of Neuroscience, 31(4): 1238-1245.
Distance-related adaptation in the human brain.
A key aspect of a cognitive map is that it preserved information about the distances between locations. To determine which areas of the brain represent this distance information, Penn students were scanned with fMRI while viewing images of familiar landmarks on campus.
(A) Colored voxels exhibit fMRI response that scales linearly with real-world distances between landmarks shown on successive trials. Distance-related adaptation was observed in the left inferior insula (ins), left aSTS, left anterior hippocampus (hipp), and right pITS.
(B) fMRI response in the anatomically defined left anterior hippocampus corresponded to the real-world distance between successively presented landmarks.
(C) The same plot for subjective distance.