When:
June 17, 2015 @ 10:30 AM – 11:30 AM
2015-06-17T10:30:00-04:00
2015-06-17T11:30:00-04:00
Where:
Baker Hall 336b

Presenter: Yuanning Li

Time: 10:30am on Wednesday, June 17th

Location: Baker Hall 336b

Title: Multivariate Discriminant Analysis of Functional Connectivity between Neural Populations

Advisor: Avniel Ghuman

Abstract:
How neural populations and interacting neural circuits encode information is a central question in neuroscience. Pattern classification methods from modern statistics and machine learning, such as multivariate pattern analysis (MVPA), have gained popularity in recent years for decoding the information content contained in neuroimaging and multiunit data. These methods allow one to go beyond examining the involvement of a population in a particular neural process and infer the representational content of the population activity. However, current MVPA methods do not allow one to assess the discriminant information encoded in the pattern of functional connections between different neural populations. Furthermore, traditional methods for assessing functional connectivity only allow one to examine differences in the degree of coupling across conditions and not the information carried by the pattern of interregional connections. Here we propose a method termed Multi-Connection Pattern Analysis (MCPA) to extract the discriminant information about cognitive states solely from the shared activity between neural populations from two brain areas. First, canonical correlation analysis (CCA) was used to learn the transformation of the activity between two populations. A classifier was then trained based on the patterns of interaction corresponding to different cognitive conditions. Simulation results showed that MCPA could successfully detect the information contained in the functional connectivity patterns across a wide range of simulated data types covering the normal range of neuroimaging and intracranial data from a variety of modalities. We then used MCPA to analyze intracranial EEG data recorded from human occipital face area (OFA) and fusiform face area (FFA) during a visual face processing task. The results demonstrated that the interaction between OFA and FFA contained information about which face the participant was viewing in a later window (around 200 – 500 ms after stim onset). The timecourse of this interaction is consistent with a previous neural decoding study based on event related potential signals within only FFA, suggesting that face recognition relies on reciprocal OFA-FFA interactions. Here we present a novel tool that uses multivariate interactions between neural populations to decode representational content contained in the coupled activity of distributed, interacting neural circuits.

Yuanning Li

Ph.D. Student
Program in Neural Computation
Center for the Neural Basis of Cognition
Carnegie Mellon University and the University of Pittsburgh

Department of Neurological Surgery
School of Medicine
University of Pittsburgh