An efficient spike-sorting for implantable neural recording microsystem using hybrid neural network.

Area of Science:

• Computational Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Online spike sorting is crucial for real-time neural recording microsystems.
• Existing methods face challenges in efficiency and accuracy.

Purpose of the Study:

• To develop an unsupervised, efficient, and adaptive spike sorting method.
• To improve the classification accuracy of neural signals.

Main Methods:

• Proposed a hybrid neural network combining Principal Component Analysis Network (PCAN) for feature extraction and dimension reduction.
• Utilized a Normal Boundary Response (NBR) Self-Organizing Map Network (SOMN) classifier for spike distribution and clustering in feature space.
• Spike sorting performed by computing NBR to determine neuron classes.

Main Results:

• The hybrid neural network spike sorting algorithm achieved accuracy exceeding 97.91% for signals with five classes.
• PCAN effectively reduced feature dimensions and eliminated correlations.
• SOMN demonstrated clear spike cluster distribution in the feature space.

Conclusions:

• The proposed unsupervised spike sorting method offers high efficiency and accuracy for neural data.
• The novel classification algorithm enhances the adaptability of neural signal processing systems.
• This approach addresses a significant challenge in online neural recording microsystems.