A review on cluster estimation methods and their application to neural spike data.

Area of Science:

• Computational Neuroscience
• Signal Processing
• Machine Learning

Background:

• Extracellular action potentials reflect collective neuronal activity, requiring spike sorting to identify individual neuron signals.
• Determining the number of neurons (clusters) is a critical challenge in spike sorting due to noise and signal overlap.
• Manual inspection of neural data is insufficient for processing large datasets.

Purpose of the Study:

• To comprehensively review and implement clustering validity indices for determining the number of neurons in neural datasets.
• To evaluate the performance of these indices on synthetic and empirical neural data.
• To identify reliable indices for accurate neuron cluster determination in spike sorting.

Main Methods:

• Thirty-three clustering validity indices were reviewed and implemented.
• K-means clustering was applied to twenty synthetic and one empirical neural dataset.
• The performance of the indices was compared against ground truth labels.

Main Results:

• The top five clustering validity indices demonstrated consistent performance across varying noise levels.
• These indices proved effective for both synthetic and real neural datasets.
• The selected indices provide robust support for determining neural cluster counts.

Conclusions:

• Specific clustering validity indices can reliably determine the number of neurons in neural spike sorting.
• The identified top-performing indices offer a data-driven solution to a critical challenge in neuroscience.
• This work facilitates more accurate analysis of large-scale neural recordings.