Neural Noise

Directional hand movement decoding from insular cortex SEEG with RNN and high-gamma

2 min read

  • Directional hand movement can be classified from insular cortex SEEG signals using recurrent neural networks and high-gamma band features.1
  • Insular cortex as a control source expands potential BCI implant targets beyond traditional motor cortex.1
  • The work is published in Nature with strong methods and electrophysiology relevance for iEEG/SEEG motor decoding.1 1

Weekly enrichment (2026-07-20)

  • The study was published in Scientific Reports (Nature Portfolio, article s41598-025-14805-3), received 15 April 2025 and accepted 4 August 2025.2
  • Seven participants with medication-resistant epilepsy underwent stereo-EEG (SEEG) depth-electrode implantation in the insula for clinical seizure monitoring, with recordings made during a cued motor task.34
  • The task used three conditions (left-hand movement, right-hand movement, and no movement), and neural signal processing focused on high-gamma band activity (70–200 Hz).4
  • Feature extraction used demixed Principal Component Analysis (dPCA) for dimension reduction (d = 10), feeding a single-layer bidirectional Long Short-Term Memory (LSTM) network implemented in PyTorch.34
  • The dPCA + LSTM decoder achieved a mean classification accuracy of 72.6% ± 13.0% (mean ± SD), significantly above the 33.3% chance level (p < 0.0001, n = 16 sessions).34
  • Across frequency bands, high-gamma (70–200 Hz) gave the best performance (~73% average across participants), and dPCA transformation significantly improved accuracy over non-transformed input.2
  • For directional decoding the insula outperformed the amygdala (mean 53.8%), cingulate (59.3%), and hippocampus (61.6%), all p < 0.001, and the LSTM outperformed LDA and SVM baselines.24
  • A strong negative correlation was found between the temporal distance of training vs. testing sessions and decoding accuracy (r = −0.868, p < 0.0001), indicating drift in the neural representations over days.4
  • The work was supported by NINDS grant K23 NS114190 under IRB-approved protocol HS-17-00544, and positions the insula as an additional non-motor-cortex source for SEEG-based motor BCIs.23

Footnotes

  1. https://news.google.com/rss/articles/CBMiX0FVX3lxTE5fNjN5VDJDRG12ZVVjYkZMc2VrVlhYR2Q2Z085LWp3ZVZYSEl0UDFFQ3I2NGl5em9qOFZoZWtOQzg0LXBaQjVYR0ZaZmkyWGVpMVMyelB5aVRyMHljSTA4?oc=5 ↩ ↩2 ↩3 ↩4

  2. https://www.nature.com/articles/s41598-025-14805-3 ↩ ↩2 ↩3 ↩4

  3. https://pubmed.ncbi.nlm.nih.gov/40819020/ ↩ ↩2 ↩3 ↩4

  4. https://pmc.ncbi.nlm.nih.gov/articles/PMC12357955/ ↩ ↩2 ↩3 ↩4 ↩5 ↩6

Graph View