Spike Sorting Quality Control Fundamentals
Spike sorting assigns extracellular voltage waveforms to putative single neurons. Quality control is essential because sorting errors propagate into all downstream analyses — firing rates, correlations, tuning curves, and population decoding.
ISI Violation Rate quantifies refractory period violations. A true single neuron cannot fire within ~1 ms of its previous spike. Violations of this constraint indicate contamination from other neurons or sorting errors.
Signal-to-Noise Ratio measures waveform detectability. Low SNR clusters have ambiguous spike shapes that are difficult to separate reliably. The threshold of SNR > 5 for “Good” quality reflects the minimum needed for confident waveform classification.
Contamination Estimation (Hill et al., 2011) uses ISI violation counts to estimate the fraction of misassigned spikes, providing a single-number summary of cluster purity.
Common Pitfalls in Spike Sorting QC
Several issues can confound spike sorting quality assessment:
• Drift: Electrode drift causes waveform amplitude to change over time, splitting a single unit into multiple clusters or merging distinct units
• Bursting neurons: Neurons that fire in bursts naturally have short ISIs that can be misclassified as refractory violations if the refractory period threshold is set too long
• Overlapping spikes: When two neurons fire nearly simultaneously, their waveforms superimpose, creating hybrid shapes that confuse sorters
• Noise floor changes: Fluctuating noise RMS (e.g., from movement artifacts) makes SNR unreliable if computed over the full recording
• Low firing rate units: Clusters with very few spikes have unstable ISI violation rates — a single coincidental violation can push the rate above thresholds
• Over-splitting: Aggressive sorting can split one neuron into multiple high-quality clusters, inflating the Good cluster count while reducing each cluster’s spike count
