BCIwiki (bciwiki.com) — A study published in Science Advances on October 23, 2026, reports that a collaboration involving West China Hospital of Sichuan University, China, developed Pathology-Aware Consensus Training (PACT) to improve motor decoding in pathological brain networks. The approach embeds biophysical constraints into healthy neural recordings to guide decoding, providing a proof of concept for brain-computer interfaces (BCIs) that must interpret signals from compromised networks.
The researchers tested gesture classification using high-density electrocorticography (ECoG) from patients with gliomas infiltrating the motor cortex. They reported improved classification accuracy with PACT. Correlation between motor-command representations and healthy canonical representations increased from r = 0.66 to r = 0.94. These findings concern classification performance and alignment between neural representations, rather than recovery of patients' motor abilities.
The authors interpret the model's behavior as an emergent "inverse weighting" strategy: it autonomously suppresses high-power pathological noise while redirecting spatial attention toward functional compensatory regions. They propose that healthy neural dynamics can help guide decoding in compromised networks, with BCI deployment in clinical populations remaining a future application.