Andreea-Maria Gui: Design and evaluation of categorical encoding models for fMRI analysis using simulated data

BCCN Berlin / Technische Universität Berlin

Abstract

Visual working memory (VWM) is a cognitive function that engages multiple neural representations, distributed across the cortex. Behavioural and fMRI studies show that the VWM representations can be categorical or continuous in nature. Here, we study this categorical and continuous nature of orientation neural codes using encoding models and cvCrossMANOVA.

First, we use simulations to develop analysis techniques for fMRI data that can distinguish between these two types of neural codes. For this we use a newly developed toolbox for simulating neural patterns in fMRI data and test using cvCrossMANOVA whether these simulated neural patterns are best explained by a categorical or a continuous encoding model. Simulation results demonstrate that encoding modelling using cvCrossMANOVA is a reliable and precise tool for model comparison between categorical and continuous neural representations.

Second, we apply these analysis techniques to real data from a delayed recall working memory experiment with orientations. The results show that posterior regions of the human brain (V1 and V3ab) prefer continuous encoding models whereas anterior regions (IPS and FEF) prefer categorical encoding models. This finding suggests that orientation representations are more continuous in nature in posterior areas and more categorical in anterior areas.

 

Additional Information

Master Thesis Defense

 

Organized by

Dr. Thomas B. Christophel   & Prof. Dr. Carsten Allefeld   / Lisa Velenosi