CosMIC: A Consistent Metric for Spike Inference from Calcium Imaging

Area of Science:

• Neuroscience
• Computational Biology
• Signal Processing

Background:

• Accurate detection of neuronal spiking activity from two-photon calcium imaging is crucial for understanding neural circuits.
• Existing metrics for evaluating spike train detection algorithms, such as spike train correlation and success rate, have known limitations.
• There is a need for more sensitive and accurate metrics to assess the performance of spike detection algorithms.

Purpose of the Study:

• To highlight the limitations of commonly used spike train similarity metrics.
• To propose and validate a novel metric, CosMIC (Correlation of Smoothed மை Calcium Imaging), for assessing spike train accuracy.
• To demonstrate CosMIC's superiority in evaluating precision and recall compared to existing methods.

Main Methods:

• Developed CosMIC, a metric based on the convolution of spike trains with a smoothing pulse derived from imaging data statistics.
• Evaluated CosMIC's performance by comparing it with spike train correlation and success rate using simulated and real two-photon calcium imaging data.
• Analyzed CosMIC's sensitivity to precision, recall, and the number of detected spikes.

Main Results:

• CosMIC effectively discriminates the precision and recall of spike train estimates.
• Unlike spike train correlation, CosMIC is maximized when the correct number of spikes are detected, avoiding rewards for overestimation.
• CosMIC demonstrates higher sensitivity to the temporal precision of spike train estimates compared to the success rate metric.

Conclusions:

• CosMIC offers a more accurate and robust assessment of spike train detection performance from calcium imaging data.
• The proposed metric addresses key limitations of existing methods, providing better insights into algorithm precision and recall.
• CosMIC's foundation in set theory and its sensitivity to temporal precision make it a valuable tool for neuroscientists.