←
Nick Card124 posts
Nick Card avatar

Nick Card

@NS_Card

Neuroengineer building speech BCIs | BWF CASI & A.P. Giannini Postdoctoral Fellow @UCDavis department of neurological surgery | Previously @PittBioE

◷ Joined June 2020▧ 24 Photos and videos
Nick Card avatar
Nick Card@NS_Card·9 Jul 2025

Our new preprint describes a multimodal intracortical brain-computer interface that a man with ALS has used at home, independently, almost every day for >19 months. It decodes both speech and cursor control to enable him to communicate and use his computer. Here’s a quick tour👇

0:31
Nick Card avatar
Nick Card@NS_Card·Jul 29

This was a hugely collaborative project amongst several BrainGate sites and participants! Thanks to all, but especially to my co-first author Zachery Fogg and co-senior authors/mentors @DrDavidBrandman and @SergeyStavisky.

Nick Card avatar
Nick Card@NS_Card·Jul 29

This work demonstrates that pooling intracortical speech data across users is effective for creating robust higher-accuracy speech decoders that generalize to new users with minimal calibration data. Read the preprint for more details: biorxiv.org/content/10.64898…

Nick Card avatar
Nick Card@NS_Card·Jul 29

Finally, MU models could be rapidly adapted to new users to enable high-accuracy, large-vocabulary decoding of attempted speech with limited calibration data, far outpacing SU models. This approach could enable new BCI users to communicate effectively without delay.

Post attachment
Nick Card avatar
Nick Card@NS_Card·Jul 29

MU decoding accuracy scaled with the number of pretraining users, consistently delivering higher accuracy decoding for held-out users as the number of pretraining users grew. Thus, MU models can generalize to unseen users, and generalization improves with more pretraining users.

Post attachment
Nick Card avatar
Nick Card@NS_Card·Jul 29

These closed-loop gains were mirrored in offline experiments with all six participants, regardless of sex, disease etiology, or attempted speaking strategy: MU models again decoded attempted speech more accurately (51.1% relative improvement in word error rate) than SU models.

Post attachment
Nick Card avatar
Nick Card@NS_Card·Jul 29

Multi-user (MU) models decoded attempted speech more accurately than single-user (SU) models in closed-loop experiments with three participants, resulting in a 55.9% relative improvement in word error rate.

Post attachment
Nick Card avatar
Nick Card@NS_Card·Jul 29

We used data from six BrainGate2 clinical trial participants to train a transformer-based decoder model. Participant-specific input networks mapped each users’ neural data to a shared latent space, which were then decoded into phoneme probabilities, and then into word sequences.

Post attachment
Nick Card avatar
Nick Card@NS_Card·Jul 29

Our new preprint describes an intracortical speech neuroprosthesis that generalizes across users, improving brain-to-text decoding accuracy for existing users by ~2x and enabling high-accuracy decoding for new users with minimal calibration data. biorxiv.org/content/10.64898…

Nick Card avatar
Nick Card@NS_Card·Jun 15

This is now published in @NatureMedicine! Check out this "behind the paper" blogpost, co-written by participant T15 and I, about how this paper came together and how he uses the BCI in his daily life: go.nature.com/3SgxMEZ

Our new preprint describes a multimodal intracortical brain-computer interface that a man with ALS has used at home, independently, almost every day for >19 months. It decodes both speech and cursor control to enable him to communicate and use his computer. Here’s a quick tour👇

0:31