An Unsupervised Compressed Sensing Algorithm for Multi-Channel Neural Recording and Spike Sorting.

Area of Science:

• Neuroscience
• Signal Processing
• Machine Learning

Background:

• Neural recordings generate vast amounts of data, necessitating efficient compression and analysis techniques.
• Accurate spike sorting is crucial for understanding neural activity but is challenging without prior information.

Purpose of the Study:

• To develop an unsupervised compressed sensing (CS)-based framework for neural action potential compression, recovery, and clustering.
• To enable integration into high-density multi-electrode neural recording VLSI systems.
• To enhance unsupervised spike sorting using dictionary learning and template matching.

Main Methods:

• Utilized compressed sensing (CS) for data compression and recovery.
• Employed spectral clustering and group structures within dictionary learning for unsupervised spike sorting.
• Incorporated group sparsity for multi-channel recordings (e.g., tetrodes).
• Integrated template matching into sparse coding for joint cluster prediction.

Main Results:

• Achieved high compression ratios (8:1 to 20:1).
• Demonstrated high-quality signal reconstruction performance (>8 dB).
• Obtained high spike sorting accuracy (>90%) without prior labels.

Conclusions:

• The proposed CS-based framework offers an efficient and accurate solution for neural data analysis.
• The unsupervised approach simplifies spike sorting, making it suitable for VLSI integration.
• The method shows significant potential for advancing neural recording system capabilities.