Comparison of fMRI Experimental Paradigm for Decoding Color Constancy

Abstract

An effective paradigm is key to explore the neural mechanisms behind color constancy (CC) using fMRI. Based on the principle of hemodynamics, we design a different paradigm containing a resting block to compare with the original one without the resting block (Bannert and Bartels, 2017 NeuroImage). The stimulus blocks consist of scenes containing two different color patches illuminated by three different illumination conditions. Two classification tasks are used to decode the color signals from the recorded fMRI signals when the subject is watching the stimulation blocks under these stimulus conditions. For the classification task within the illuminants (e.g., a model is trained on the fMRI signals to discriminate the blue or yellow patches under two illuminants and tested to predict the color of the patches under the similar illuminants), we found that the classification accuracy of fMRI signals in the V1, V2, V3, and V4 areas in the paradigm with and without a resting block are both significantly higher than the random level. However, for the classification task across the different illuminants (e.g., a model is trained on the fMRI signals to discriminate the blue or yellow patches under two illuminants (blue and yellow) and then tested to predict the color of the patches under the third illuminant (e.g., neutral)), only the classification accuracy of the fMRI signals of V2 and V4 areas in the new paradigm are significantly higher than the random level, but not the case without a resting block. The major results are that both V2 and V4 regions have the function of decoding CC, which is highly relying on the experimental paradigm used. The resting blocks can effectively separate the BOLD signals between the stimulation blocks so that the signals do not overlap and interact, which is more conducive to decode the neural mechanism of CC.