Millisecond-Level Online Spike Sorting under Abrupt Drift

Achieve millisecond-level latency in online spike sorting for streaming extracellular recordings by reducing the latency introduced by the minimum segment duration parameter L_min in the segmentation-based SpikeSift algorithm, while maintaining sorting accuracy under dynamic recording conditions, particularly when abrupt electrode drift occurs.

Background

SpikeSift processes streaming or growing recordings using an adaptive segmentation strategy that imposes a minimum segment duration (L_min). This design introduces a small but nonzero latency, which is acceptable for batch processing but limits true real-time operation at millisecond-level latency.

The authors explicitly note that achieving millisecond-level spike sorting under non-stationary recording conditions, especially in the presence of abrupt electrode drift, remains unresolved. Addressing this challenge would enable closed-loop and real-time applications that demand extremely low latency without sacrificing sorting accuracy.