Neural Noise

Event-based seizure detection in human iEEG with neuromorphic hardware

2 min read

  • Event-based processing on neuromorphic hardware can run on human intracranial EEG for real-time seizure detection (medRxiv).1
  • The approach offers a path to low-power, implantable or bedside seizure detectors.1 1

Weekly enrichment (2026-07-20)

  • The study (medRxiv preprint, July 2025; not yet peer-reviewed) tested whether a spiking neural network (SNN) on the DYNAP-SE1 neuromorphic chip could detect the monotonic descending “chirp” pattern that marks seizure onset in intracranial EEG.2
  • It analyzed 48 hours of continuous bipolar iEEG from a single patient containing 40 seizures, using only one iEEG channel.2
  • Signals were filtered into six 10 Hz sub-bands spanning 0–40 Hz and encoded into UP/DOWN spike events via software asynchronous delta modulation (ADM) before being fed to the SNN.2
  • The hardware SNN used just 34 adaptive-exponential neurons — 3.3% of the 1024 neurons on one chip — with hierarchical inhibition enforcing the high-to-low band sequence and a disinhibition unit suppressing isolated low-frequency bursts.2
  • Chirps appeared at the onset of all 40 seizures (100%); the hardware SNN detected every seizure (100% sensitivity) with a single false alarm over 48 h (false-alarm rate 0.021/h).2
  • A software-only variant (28 neurons, no disinhibition population) detected 32/40 seizures (80% sensitivity) with 9 false alarms (0.19/h), illustrating the value of the hardware disinhibition circuit.2
  • The pipeline ran in real time with mean processing of 4 h 55 s ± 42 s per 4-hour block (~55 s latency per 4 h of iEEG); event-driven computation only draws energy when spikes occur, suiting always-on battery-powered wearable or implantable monitoring.2
  • Related neuromorphic work reinforces the direction: a two-layer SNN on the DYNAP-SE2 that amplifies seizure-related partial synchronization ran at an estimated 150 µW, supporting ultra-low-power edge seizure detection.3

Footnotes

  1. https://news.google.com/rss/articles/CBMie0FVX3lxTFBJbTBfT2o0eVF6U0N3NDRJYkZmc2VLUHdmeTN2UkJ4RGM3aGRza1FodXZYTFNZN0lpMFNqempEcVlEVFdIQ25fd3NqSkxQTFU3MF9xWVZEV1ZmSlB2bGdEVmF3MXo3Q1VVTEQzNXhsal84dEJOaE5DWkJQWQ?oc=5 ↩ ↩2 ↩3

  2. https://www.medrxiv.org/content/10.1101/2025.07.10.25331024v1 ↩ ↩2 ↩3 ↩4 ↩5 ↩6 ↩7

  3. https://www.nature.com/articles/s41598-025-99272-6 ↩

Graph View