UT Austin shows slow fMRI data boosts fast ECoG brain signal predictions

Neuroscientists at UT Austin (Vaidya, Antonello, Huth) have demonstrated that “slow” fMRI brain recordings can significantly improve prediction of “fast” ECoG (electrocorticography) signals. ECoG offers fine spatial and temporal resolution but is limited by its invasive nature—only certain patient populations can receive implants. The team used spoken language representations fine-tuned on fMRI to build encoding models of ECoG. Despite fMRI’s temporal resolution being roughly two orders of magnitude worse (~2 seconds vs ~10 milliseconds for ECoG), the fine-tuned models showed improved prediction performance in ECoG across frequency bands well beyond those directly measured in fMRI.

To test generalization, the researchers temporally downsampled fMRI responses by a factor of 2. Even with that additional loss in resolution, the models predicted fMRI and ECoG responses at levels comparable to the original fine-tuned models. Crucially, ECoG performance steadily scaled with the amount of fMRI tuning data, suggesting that non-invasive data can act as a valuable resource for building better brain decoding models. The results open the door to integrating multiple recording modalities to boost decoding accuracy in future neural interfaces.