Improving an open-source commercial system to reliably perform activity-dependent stimulation. (29th October 2019)

• Record Type: Journal Article
• Authors: Murphy, Maxwell; Buccelli, Stefano; Bornat, Yannick; Bundy, David; Nudo, Randolph; Guggenmos, David; Chiappalone, Michela
• Is Part Of: Journal of neural engineering. Volume 16:Number 6(2019:Dec.)
• Publication Date: 2019-10-29
• Subjects: FPGA -- neuroprosthetics -- spike detection; Neurosciences -- Periodicals; Biomedical engineering -- Periodicals; 612.8
• DOI: 10.1088/1741-2552/ab3319
• Languages: English
• ISSNs: 1741-2560

Abstract

Abstract: Objective . Activity-dependent stimulation (ADS) is designed to strengthen the connections between neuronal circuits and therefore may be a promising tool for promoting neurophysiological reorganization following a brain injury. To successfully perform this technique, two criteria must be met: (1) spikes in the extracellular electrical field potential must be detected accurately at one site of interest, and (2) stimulation pulses generated at fixed (<1 ms jitter), low-latency (<10 ms) intervals relative to each detected spike must be delivered reliably to a second site of interest. Here, we aimed to improve noise rejection in a low-cost commercial system to reliably perform ADS in awake, behaving rats, while maintaining latency requirements. Approach . We implemented a spike detection state machine on a field-programmable gate array (FPGA). Because the accuracy of spike detection can be heavily reduced in awake and behaving animals due to biological artifacts such as movement and chewing, the state machine tracks candidate spike waveforms, checking them against multiple programmable thresholds and rejecting any spikes that fail to meet a programmed threshold criterion. Main Results . A series of offline analyses showed that our implementation was able to appropriately trigger stimulation during epochs of biological artifacts with an overall accuracy between 72% and 97%, fixed computational latency of 167 µ s, and an algorithmic latency of 300 µ s to 800 µ s. Significance . Our improvements have been made open-source and are freely available to all scientists working on closed-loop neuroprosthetic devices. Importantly, the improvements are easily incorporated into existing workflows that utilize the Intan Stimulation and Recording Controller.