BCIwiki (bciwiki.com) — A study published in Journal of Neural Engineering on October 23, 2026, reports that researchers at The University of Melbourne evaluated temporal neural features for motor decoding using a Stentrode endovascular brain-computer interface (BCI). One implanted participant attempted to handwrite English characters, which the researchers classified offline and online from recorded brain activity.
The researchers hypothesized that relatively long compound movements generate distinct neural activity patterns captured by temporal features. To isolate temporal information, the model received a one-dimensional trajectory of high-gamma power summed across all channels, rather than retaining separate spatial information from each channel. A small number of classes was reliably decoded across multiple months, indicating stable temporal patterns associated with different handwriting movements.
Across four online sessions, each containing 20-30 trials per character and 80-180 rest trials, the best-performing 3-class model achieved a mean rest recall of 87%, character recall of 65%, and character precision of 59%. Recall describes how many trials belonging to a class were correctly identified; precision describes how many predictions assigned to a class were correct. These measures are not interchangeable with overall accuracy.
Linear temporal warping, which adjusted all trials to the same duration to reduce timing variability, significantly improved decoding compared with classification of raw trials. The authors suggest that temporal information might reduce the spatial footprint required by BCIs and potentially make them less invasive; the study did not demonstrate such a device modification. They call for testing in more participants and other recording modalities, alongside investigation of automatic temporal alignment strategies.