Abstract
Brain-computer interfaces (BCIs) can restore communication to people who have lost the ability to move or speak. To date, a major focus of BCI research has been on restoring gross motor skills, such as reaching and grasping or point-and-click typing with a 2D computer cursor. However, rapid sequences of highly dexterous behaviors, such as handwriting or touch typing, might enable faster communication rates. Here, we demonstrate an intracortical BCI that can decode imagined handwriting movements from neural activity in motor cortex and translate it to text in real-time, using a novel recurrent neural network decoding approach. With this BCI, our study participant (whose hand was paralyzed) achieved typing speeds that exceed those of any other BCI yet reported: 90 characters per minute at >99% accuracy with a general-purpose autocorrect. These speeds are comparable to able-bodied smartphone typing speeds in our participant’s age group (115 characters per minute) and significantly close the gap between BCI-enabled typing and able-bodied typing rates. Finally, new theoretical considerations explain why temporally complex movements, such as handwriting, may be fundamentally easier to decode than point-to-point movements. Our results open a new approach for BCIs and demonstrate the feasibility of accurately decoding rapid, dexterous movements years after paralysis.
Introduction
Restoring communication for individuals with severe motor impairments, such as locked-in syndrome or high-level spinal cord injuries, is a primary goal of neural engineering. Current high-performance BCIs typically rely on "point-and-click" paradigms, where users control a 2D cursor to select letters on a virtual keyboard. While effective, these systems are limited by the simplicity of the underlying motor signals, peaking at approximately 40 characters per minute (CPM). The fundamental bottleneck is the neural similarity of point-to-point movements, which makes rapid discrimination difficult.
This study explores a different approach: decoding "dexterous" movements, specifically imagined handwriting. Handwriting involves complex, unique spatiotemporal patterns for each character, which may be more distinct in neural space than simple directional movements. We hypothesized that this increased complexity would facilitate faster and more accurate decoding.
Methods
The study involved a 65-year-old participant (T5) with a C4 spinal cord injury, resulting in paralysis from the neck down. To capture neural signals, two 96-electrode microelectrode arrays were implanted in the "hand knob" area of the precentral gyrus, a region of the motor cortex associated with hand movements.
The experimental task required T5 to "attempt" to write letters as if he were holding a pen on paper. A Recurrent Neural Network (RNN) was trained to map the high-dimensional neural population activity to probabilities for 31 possible characters (26 lowercase letters and 5 symbols). To optimize performance, the RNN utilized a 1-second output delay, allowing it to integrate the full neural trajectory of a character before classification. Due to the limited amount of real-time neural data, the researchers employed data augmentation by stitching together snippets of individual characters to create synthetic training sentences. Finally, a large-vocabulary (50,000-word) bigram language model provided real-time autocorrect.
Results
The handwriting BCI achieved a record-breaking typing speed of 90 characters per minute. This nearly doubles the previous state-of-the-art for BCIs and approaches the average smartphone typing speed of 115 CPM for individuals in the participant's age group. Accuracy was equally impressive: raw character accuracy was 94.1%, which improved to >99% with the use of the language model.
A key insight from this work is that the spatiotemporal complexity of handwriting is a significant advantage for decoding. Unlike point-to-point movements, each letter produces a highly distinct neural signature.
As shown in the t-SNE visualization, neural activity patterns for different characters form well-separated clusters. This high dimensionality in the signal allows the decoder to maintain high accuracy even in the presence of neural noise. Furthermore, the study demonstrated that motor representations for handwriting remain robust and "motoric" even nine years after the initial injury.
Discussion
This research represents a landmark achievement in BCI performance, demonstrating that rapid, dexterous movements can be decoded years after paralysis. The shift from simple cursor control to complex handwriting movements leverages the brain's natural spatiotemporal diversity to overcome previous speed limits.
While this system is a powerful proof-of-concept, several challenges remain for clinical translation. The current implementation lacks capital letters and intuitive editing features (e.g., backspace). Additionally, the transition from wired lab setups to fully implanted, wireless clinical devices is a major engineering hurdle. Future work will focus on expanding the character set and optimizing calibration procedures to make the system more practical for daily use. Combining handwriting decoders with traditional point-and-click interfaces could eventually provide a comprehensive and high-speed communication solution for individuals with severe paralysis.
Summary updated