Real-Time, Low-Power Purkinje Cell Spike Sorting for Headstage Systems in Freely Moving Mice

Abstract

Studying Purkinje cell activity is vital for understanding brain function and movement disorders. However, conventional wired headstage setups in mice restrict natural behavior, while transmitting full neural waveforms wirelessly is prohibitively power-intensive. This work presents a lightweight, real-time spike sorting system implemented on the EFM32PG28 microcontroller, capable of classifying complex spikes and simple spikes from Purkinje cells directly on the headstage. By extracting only relevant spike information, the system enables efficient wireless transmission or local storage. Leveraging knowledge distillation and Matrix Vector Processor (MVP) hardware acceleration, the spike sorter achieves an overall $\mathbf{F 1}$-score of 93.43% and completes detection and classification in 0.765 ms. With over 53 hours of continuous operation on a $50-\mathrm{mAh}$ battery, the proposed solution is well-suited for long-duration, untethered cerebellar experiments in freely moving mice.