Improved space breakdown method - A robust clustering technique for spike sorting.

Area of Science:

• Computational Neuroscience
• Machine Learning
• Data Science

Background:

• Neuronal spike sorting is crucial for analyzing brain activity.
• Traditional clustering methods struggle with overlapping and imbalanced neuronal data.
• The original Space Breakdown Method (SBM) is effective for low-dimensional data but computationally expensive for high-dimensional datasets.

Purpose of the Study:

• To improve the Space Breakdown Method (SBM) for high-dimensional neuronal data analysis.
• To enhance computational efficiency and performance of the SBM algorithm.
• To introduce a novel clustering validation metric suitable for spike sorting.

Main Methods:

• The original SBM algorithm was modified by replacing the array structure with a graph structure.
• The number of partitions was made feature-dependent, creating the Improved Space Breakdown Method (ISBM).
• A new clustering validation metric was developed that does not penalize overclustering.

Main Results:

• The ISBM demonstrates reduced space and time complexity compared to the original SBM.
• ISBM shows increased performance on high-dimensional neural data.
• Evaluations on synthetic data confirm ISBM's effectiveness against state-of-the-art algorithms.

Conclusions:

• The ISBM significantly improves the applicability of the SBM algorithm to high-dimensional neuronal spike sorting.
• The proposed clustering validation metric provides more suitable evaluations for spike sorting tasks.
• ISBM offers a computationally efficient and high-performing solution for analyzing complex neural data.