FALCON Benchmark H2: Human Handwriting iBCI

The dataset contains neural activity from participant T5, who gave informed consent and was enrolled in the BrainGate2 Neural Interface System clinical trial (ClinicalTrials.gov Identifier: NCT00912041, registered June 3, 2009). This pilot clinical trial was approved under an Investigational Device Exemption (IDE) by the US Food and Drug Administration (Investigational Device Exemption #G090003). Permission was also granted by the Institutional Review Boards of Stanford University (protocol #20804). Data was recorded from two 96-channel intracortical electrode arrays in the hand "knob" area of of T5's left hemisphere precentral gyrus. On each trial, T5 was prompted to copy a cued sentence by writing individual characters, separated by a ">" space character. Decoded letters appeared on the screen below the prompt.

In each datafile, the `acquisitions` field contains `binned_spikes` (binned at 20ms) and the `eval_mask` used for determining valid FALCON evaluation periods. The `trials` field contains trial metadata including the prompt that T5 was asked to attempt to write (`cue`) and the experimental block that each trial belongs to (`block_num`).

As part of the FALCON challenge, dataset files are released in three ways. The `held-in-calib` data includes many trials of data intended for calibration of decoders in the FALCON challenge. The `held-in-minival` data is a small subset of the `held-in-calib` data intended for FALCON models to validate submission format. Lastly, `held-out-calib` data are separate sessions of data from `held-in-calib`, but have much fewer trials, intended to be used for few-shot recalibration. Please bear in mind these factors if making use of this dataset for purposes outside of the FALCON Benchmark. Each dataset split is considered one "subject" by the DANDI repository; only data from one iBCI participant is included.