Adaptive spike detection and hardware optimization towards autonomous, high-channel-count BMIs
Zheng Zhang1, Timothy G Constandinou2
1Department of Electrical and Electronic Engineering, Imperial College London, South Kensington Campus, London SW7 2AZ, UK.
Journal of Neuroscience Methods
|February 22, 2021
Summary
We developed an adaptive algorithm for real-time spike detection in neural recordings. This low-power, low-resource method enhances signal processing for brain-machine interfaces.
Area of Science:
- Neuroscience
- Biomedical Engineering
- Signal Processing
Background:
- Microtechnology advancements enable simultaneous recording of more neurons, increasing data bandwidth demands.
- Current electrophysiology practices often involve offline processing of raw data for spike event detection.
- Emerging applications necessitate local, real-time processing methods for neural data.
Purpose of the Study:
- To develop an adaptive, low-complexity spike detection algorithm for real-time neural data processing.
- To optimize the algorithm for hardware implementation on low-power embedded systems.
- To enable autonomous and calibration-free spike detection for advanced applications.
Main Methods:
- The algorithm integrates novel components for local field potential removal, signal-to-noise ratio enhancement, and adaptive threshold computation.
- Optimized for hardware implementation, minimizing computations for fixed-point processing.
- Demonstrated on low-power embedded targets.
Main Results:
- Validated on synthetic and real neural recordings, achieving up to 90% detection sensitivity.
- Hardware implementation on an embedded platform required <0.1 KB ROM and 3 KB program flash.
- Consumed an average power of 130 μW.
Conclusions:
- The algorithm enables robust, real-time, autonomous spike detection without calibration.
- Low hardware resource requirements make it suitable for embedded applications.
- Crucial for advancing Brain-Machine Interfaces (BMIs), especially for high-channel-count systems.
