Abstract
Despite measuring neural activity indirectly, functional magnetic resonance imaging (fMRI) has become the most important tool in mapping the human visual system. However, current fMRI measurements may suffer from low signal-to-noise ratio (SNR) in detecting high-level neural responses in individual brains, leading to low test-retest reliability in spatial activation maps. Here we developed a novel fMRI approach to map visual categorization with fMRI. As in EEG frequency-tagging (Rossion et al., 2015), we presented a large variety of natural images at a fast rate (6Hz) throughout the entire experiment to stimulate the visual areas continuously. By introducing transient switches to a target category (faces) at a slow fixed frequency (1/54 stimuli, i.e., 0.111 Hz), we obtained a periodic differential neural response that directly reflects category selectivity. A model-free Fast Fourier Transform (FFT) of hemodynamic activity in this paradigm achieved a two-fold increase in sensitivity (peak SNR) in comparison to a conventional block design, allowing us to map comprehensive extended face-selective areas including the anterior temporal lobe (ATL) in individual brains. Using diverse natural images, we created contrasts at both lower-level visual properties and higher-level category properties in successive images. While the lower-level contrasts were at random time points, the category contrast would only happen at a fixed frequency by design. Therefore, we effectively eliminated the influence of low-level visual cues and increased the specificity of category-selective response. As a result of high sensitivity and specificity, we achieved high test-retest reliability, which reached the highest values (80-90%) ever reported in this area of research. The power of a model-free fast periodic visual stimulation (FPVS) approach with a slow temporal resolution method opens a real avenue for understanding brain mapping of visual categorization.
Meeting abstract presented at VSS 2017