AECuration: automated event curation for spike sorting.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Electrophysiology

Background:

• Spike sorting identifies neural events from extracellular recordings, crucial for neuroscience.
• Manual curation of spike sorting is subjective, error-prone, and not scalable for large datasets.
• Non-neural events like artifacts can compromise spike sorting accuracy.

Purpose of the Study:

• To introduce AECuration, a novel automated method for curating neural spike events.
• To address the limitations of manual curation in spike sorting analysis.
• To improve the specificity and efficiency of electrophysiology data analysis.

Main Methods:

• Developed AECuration, an autoencoder-based method trained on simulated extracellular spike waveform features.
• Utilized reconstruction error to classify neural and non-neural events.
• Applied the method to experimental electrophysiology datasets.

Main Results:

• AECuration achieved 97.46% accuracy in classifying neural events on synthetic datasets.
• Improved spike sorting pipeline sensitivity by up to 20% on ground-truth datasets.
• Increased the ratio of clustered units with low interspike interval violation rates from 55.3% to 85.5%.

Conclusions:

• AECuration offers a generalizable, time-domain evaluation method for automated extracellular recording analysis.
• The trained model can be applied to real datasets without retraining, enhancing its practical utility.
• This method can be integrated into existing spike sorting pipelines to boost accuracy and efficiency.