A framework for on-implant spike sorting based on salient feature selection.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Current on-implant spike sorting methods rely on static feature extraction to reduce hardware costs.
• These static methods may not be optimal for accurately classifying neural spikes in real-time.

Purpose of the Study:

• To develop a novel framework for real-time spike sorting using dynamic feature selection.
• To improve spike discrimination and classification accuracy for neural implants.

Main Methods:

• A dynamic feature selection framework was proposed, optimizing salient features for maximal between-class distances and homogeneity.
• Wave-shape classification utilized a multi-label window discrimination approach.
• An external module configured the on-implant sorter by optimizing a replica of the on-implant operation.

Main Results:

• Hardware implementation for 512 channels achieved an average classification accuracy of approximately 88%.
• The proposed method reduced classification error by a factor of ~2 compared to similar methods.
• Required memory space was reduced by a factor of ~5.

Conclusions:

• The proposed dynamic feature selection framework enables efficient and accurate real-time spike sorting on implantable devices.
• This approach offers significant improvements in classification accuracy and hardware resource utilization for neural signal processing.