Template-based and model-based decoding of movie clip identities from brain hemodynamics with high-density diffuse optical tomography

Functional magnetic resonance imaging has decoded complex information about naturalistic stimuli using brain responses, but other non-invasive technologies have not achieved similar decoding capabilities. To evaluate feasibility of naturalistic visual decoding with diffuse optical tomography (DOT), a 6.5-mm-spaced optode grid was employed to decode which of four naturalistic, 90-second movie clips was viewed by human subjects. Over 85% average decoding accuracy was achieved using a template-matching decoder. Average accuracy remained above 60% and above chance using a model-based decoder to identify 4 and 40 clips outside the decoder's training set, respectively. DOT therefore has potential for more-complex neural decoding tasks.