A Hardware-Efficient Novelty-Aware Spike Sorting Approach for Brain-Implantable Microsystems.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Unsupervised spike sorting is crucial for real-time brain-implantable microsystems.
• Nonstationarity in neural signals, including waveform changes and new neuron activation, poses a significant challenge.
• Existing adaptive methods are resource-intensive, while static methods suffer performance degradation.

Purpose of the Study:

• To develop a hardware-efficient, novelty-aware spike sorting approach for implantable applications.
• To balance hardware cost and processing performance in nonstationary neural signal environments.
• To address the limitations of current spike sorting techniques in long-term brain recordings.

Main Methods:

• A novelty detection process is integrated to manage neural signal variations.
• Spike features are tracked, and parameters are updated upon detecting unexpected changes (novelty).
• On-implant computations are minimized, with the computational burden shifted off-implant for agility.

Main Results:

• The approach achieved 94.31% detection of novel spikes and 96.31% classification accuracy on average.
• An FPGA prototype demonstrated higher classification accuracy than the OSORT algorithm with significantly lower hardware resources.
• A CMOS implementation achieved low power consumption (1.78 µW/channel) and small chip area (0.07 mm²/channel).

Conclusions:

• The proposed novelty-aware spike sorting method offers an effective solution for nonstationary neural signals in implantable devices.
• It provides a favorable trade-off between hardware efficiency and processing performance.
• This approach is particularly suitable for brain-computer interfaces requiring long-term, real-time brain interaction under limited hardware constraints.