Hierarchical Adaptive Means (HAM) clustering for hardware-efficient, unsupervised and real-time spike sorting

Area of Science:

• Computational Neuroscience
• Machine Learning
• Signal Processing

Background:

• Accurate spike sorting is crucial for analyzing neural data.
• Existing methods often require prior knowledge or extensive training.
• Real-time adaptive clustering for neural spikes remains a challenge.

Purpose of the Study:

• To introduce a novel unsupervised algorithm for real-time adaptive clustering of neural spike data.
• To develop a method that can autonomously determine the number of spike classes.
• To create a computationally efficient algorithm suitable for hardware implementation.

Main Methods:

• The Hierarchical Adaptive Means (HAM) clustering method was developed.
• HAM combines centroid-based clustering with hierarchical cluster connectivity.
• The algorithm adaptively tracks incoming spike data without requiring past history or training.

Main Results:

• HAM achieved near-identical classification accuracy compared to k-means on simulated and recorded datasets.
• The method demonstrated robustness across different feature extraction techniques, exceeding 80% accuracy.
• Low memory and computational requirements were quantified, highlighting suitability for hardware.

Conclusions:

• HAM provides an effective unsupervised, real-time adaptive solution for spike sorting.
• The algorithm's autonomy in determining spike classes and its efficiency are significant advantages.
• The low complexity of HAM makes it highly attractive for future hardware implementations in neuroscience.