Applications

An intracortical BCI that decodes attempted handwriting from motor cortex and types it in real time: 90 characters per minute at 94.1% raw online accuracy in a participant with a paralysed hand. Nature, May 2021.

Application — Intracortical

Handwriting BCI (Willett, 2021)

Source →

handwriting · intracortical · motor cortex · typing speed · recurrent neural network · spinal cord injury · academic · human

Device

No device entry covers this hardware yet. The spec card below lists what the source says.

Lab or organization

Handwriting BCI (Willett, 2021)

One-line verdict: A decoding idea more than a new electrode. Handwriting is a rich, temporally varied movement, so it is easier to tell apart than moving a cursor to targets.

Quick tags: Recording · Intracortical · Human · Published 12 May 2021 (Nature)

Overview

What it is: An intracortical BCI that decodes attempted handwriting movements from neural activity in the motor cortex and translates it to text in real time, using a recurrent neural network. The participant, T5, has a hand paralysed by spinal cord injury.

Results (as reported): 90 characters per minute with 94.1% raw accuracy online, and greater than 99% accuracy offline with a general-purpose autocorrect. The authors say these speeds exceed those reported for any other BCI and are comparable to typical smartphone typing for the participant’s age group (115 characters per minute).

Against the prior record: The previous intracortical typing record, from a 2017 study with the same participant (Pandarinath et al.), was 39 correct characters per minute using point-and-click typing. The figure caption in the paper calls the new interface more than twice as fast.

Why it matters beyond speed: The authors argue that temporally complex movements such as handwriting may be fundamentally easier to decode than point-to-point movements, and that accurate decoding of rapid, dexterous movement is feasible years after paralysis.

Limits: One participant. The abstract does not describe the array hardware, so none is described here.

Spec Card Grid

Identity

  • Paper: “High-performance brain-to-text communication via handwriting,” Nature, 12 May 2021
  • Participant: T5, spinal cord injury

Results

  • Online speed and accuracy: 90 characters per minute, 94.1% raw
  • Offline accuracy with autocorrect: above 99%
  • Prior intracortical typing record: 39 correct characters per minute

Evidence and limits

References