Advances in recording technologies now allow us to record populations of neurons simultaneously, data necessary to understand the network dynamics of the brain. Extracellular probes are fabricated with ever greater numbers of recording sites to capture the activity of increasing numbers of neurons. However, the utility of this extracellular data is limited by an initial analysis step, spike sorting, that extracts the activity patterns of individual neurons from the extracellular traces. Commonly used spike sorting methods require manual processing that limits their scalability, and errors can bias downstream analyses. Leveraging the replication of the activity from a single neuron on nearby recording sites, we designed a spike sorting method consisting of three primary steps: (1) a blind source separation algorithm to estimate the underlying source components, (2) a spike detection algorithm to find the set of spikes from each component best separated from background activity and (3) a classifier to evaluate if a set of spikes came from one individual neuron. To assess the accuracy of our method, we simulated multi-electrode array data that encompass many of the realistic variations and the sources of noise in in vivo neural data. Our method was able to extract individual simulated neurons in an automated fashion without any errors in spike assignment. Further, the number of neurons extracted increased as we increased recording site count and density. To evaluate our method in vivo, we performed both extracellular recording with our close-packed probes and a co-localized patch clamp recording, directly measuring one neuron’s ground truth set of spikes. Using this in vivo data we found that when our spike sorting method extracted the patched neuron, the spike assignment error rates were at the low end of reported error rates, and that our errors were frequently the result of failed spike detection during bursts where spike amplitude decreased into the noise. We used our in vivo data to characterize the extracellular recordings of burst activity and more generally what an extracellular electrode records. With this knowledge, we updated our spike detector to capture more burst spikes and improved our classifier based on our characterizations.
| Identifer | oai:union.ndltd.org:bu.edu/oai:open.bu.edu:2144/19751 |
| Date | 07 December 2016 |
| Creators | Moore-Kochlacs, Caroline Elizabeth |
| Source Sets | Boston University |
| Language | en_US |
| Detected Language | English |
| Type | Thesis/Dissertation |
| Rights | Attribution-NonCommercial 4.0 International, http://creativecommons.org/licenses/by-nc/4.0/ |
Page generated in 0.0059 seconds