Decoding and Classification of Category‐Specific Visual Stimuli
in the Fusiform Gyrus Using fMRI Data and Machine Learning
This thesis examines the use of Functional Magnetic Resonance Imaging (fMRI) data from the Human Connectome Project (HCP) to analyze neural systems involved in the processing of category-specific visual stimuli. It focuses on the Working Memory (WM) task within a broad fMRI paradigm designed to explore various neural domains including visual, language, and decision-making systems.
Specifically, the study examines brain activation patterns in the Fusiform Gyrus (FG) related to different stimulus categories such as faces, places, tools, and body parts. Employing a pipeline of preprocessing steps followed by Multi-Variate Pattern Analysis (MVPA), the research assesses distributed patterns of voxel activation allowing for complex and detailed analyses of cognitive states. Data is processed through a Support Vector Machine (SVM) classification script developed for this study, aiming to differentiate between faces and other stimulus categories based solely on fMRI data.
The thesis also addresses classification methodologies,
with a particular focus on optimizing classifier accuracy by adjusting parameters
like the number of data chunks, fold counts for cross-validation, and subject counts. The
findings offer insights into the distinct brain activation patterns associated with different stimuli
in the FG, contributing to our understanding of neural mechanisms in cognitive processes and
practical applications for these insights.
The source code in this project, specifically within the "Scripts" folder, is licensed under the General Public License 3 (GPLv3).
All other media and files are licensed under Creative Commons Attribution-Sharealike 4.0 International license (CC BY-SA 4.0).