(10/10) Big thank you to @ycleong.bsky.social for all the guidance and support, @shannon47burns.bsky.social for sharing expertise in fNIRS, and the UChicago Neuroscience Institute Shared Equipment Award for the fNIRS device.

(9/x) To encourage broader adoption and further development of our method, we have made our model and preprocessed fNIRS movie-watching data publicly available at github.com/ycleong/fNIR..., along with example scripts that demonstrate how to load and apply the model.

(8/x) Our model extends the capabilities of fNIRS by allowing researchers to infer broader neural dynamics in naturalistic settings from limited fNIRS signals. We suggest using it as a hypothesis-generation tool for identifying regions and network interactions that warrant targeted fMRI studies.

(7/x) The fNIRS-fMRI signal prediction model generalized across stimuli. A model trained on Run 1 of Sherlock significantly predicted fMRI activity when watching Friday Night Lights in 66 ROIs, while a model trained on both runs significantly predicted fMRI activity in 84 ROIs.

(6/x) Do predicted neural dynamics retain moment-to-moment semantic information in the movie? To test it, we built an encoding model from time-resolve embeddings of movie annotations. Out of the 66 ROIs, 12 exhibited above-chance accuracy, including dmPFC and the language network (dlPFC and LTC).

(5/x) The predicted whole-brain neural dynamics also recapitulated ground-truth intersubject functional connectivity (ISFC) patterns.

(4/x) Our model significantly predicted the neural dynamics in 66 out of 122 ROIs, including more than 80% of ROIs in the default mode network and the control network, and including areas that were anatomically inaccessible by fNIRS, such as the precuneus and basal ganglia.

(3/x) In light of prior research on functional connectivity and linear mapping between fNIRS and fMRI signals, we adapted principal component regression (aPCR) to predict whole-brain fMRI signals from PFC fNIRS signals.

(2/x) We recorded neural activity at the prefrontal cortex (PFC) using fNIRS as participants watched a Sherlock episode. We also utilized a public fMRI dataset where different participants watched the same episode. Shared naturalistic stimuli allowed us to functionally align the two modalities.

Thinking about using fNIRS to study the brain when people chat, play, or explore the world, but can’t give up on fMRI’s whole brain coverage? We introduce a predictive model aiming to bring together portability and coverage in neuroimaging. (1/x)

(9/9) Big thank you to @ycleong.bsky.social for all the guidance and support, @shannon47burns.bsky.social for sharing expertise in fNIRS, and the UChicago Neuroscience Institute Shared Equipment Award for the fNIRS device.

(8/x) To encourage broader adoption and further development of our method, we have made our model and preprocessed fNIRS movie-watching data publicly available at github.com/ycleong/fNIR..., along with example scripts that demonstrate how to load and apply the model.

(7/x) Our model extends the capabilities of fNIRS by allowing researchers to infer broader neural dynamics in naturalistic settings from limited fNIRS signals. We suggest using it as a hypothesis-generation tool for identifying regions and network interactions that warrant targeted fMRI studies.

(6/x) Do the predicted neural dynamics retain moment-to-moment semantic information in the movie? To test it, we built an encoding model from time-resolve embeddings of movie annotations. Out of the 66 ROIs, 12 exhibited above-chance accuracy, including dmPFC and the language network (dlPFC and LTC)

(5/x) The predicted whole-brain neural dynamics also recapitulated ground-truth intersubject functional connectivity (ISFC) patterns.

(4/x) Our model significantly predicted the neural dynamics in 66 out of 122 ROIs, including more than 80% of ROIs in the default mode network and the control network, and including areas that were anatomically inaccessible by fNIRS, such as the precuneus and basal ganglia.

(3/x) In light of prior research on functional connectivity and linear mapping between fNIRS and fMRI signals, we adapted principal component regression (aPCR) to predict whole-brain fMRI signals from PFC fNIRS signals.

(2/x) We recorded neural activity at the prefrontal cortex (PFC) using fNIRS as participants watched a Sherlock episode. We also utilized a public fMRI dataset where different participants watched the same episode. Shared naturalistic stimuli allowed us to functionally align the two modalities.