Toward Automated Spike Sorting via Ground Truth Neural Recordings

PROJECTSUMMARY
Scaling extracellular electrophysiology to higher channel counts is hindered by the burden of data
handling storageand especially preprocessinge gspike sortingThe burden of spike sorting can in
principle be reduced through a combination of high density multielectrode arrayprobetechnology and
algorithm optimization to yield a spike sorting method that is both highly accurate and fully automatedWith a known good spike sorting method in handthe algorithm can be baked into the data stream
as early as possible to allow for automatic data sorting and a massive reduction in data rate to
downstream storage and processingHoweverit takes an investment of considerable resources to
implement this sort of large scale real time processingand great confidence to throw away raw data and
keeponlyprocesseddataAccuracy and automation of spike sorting increases with the spatial density of recording sitesNeural activity recorded from high density probes can serve as a data corpus for testing the accuracy of
spike sorting algorithmsHoweverto quantify spike sorting performance for comparison between
algorithmsthe ground truth spiking activity of neurons captured in the data corpus must be measuredsuch as by simultaneously recording via patch clamp pipette or some other recording modalityUnfortunatelybecause ground truth recordings are so challenging to performthey remain too rare to
allow for this sort of analysis in a large scalemeaningful wayUntil this need is metspike sorting
development lacks a compassand cutting edge techniques such as supervised machine learning which
require large amounts of labelled data remain out of reachAccordinglywe propose a series of
multimodal neural recordings combining multielectrode array and patch pipette techniques to
generateacorpusofgroundtruthdataforvalidationofspikesortingalgorithms PROJECTNARRATIVE
Electrophysiologicalrecordingsystemsallowdirectobservationofneuralactivityinanimal
subjectsThisfacilitatesthestudyofcrucialneuroscientifictopicssuchasdevelopmentlearningandmemoryandcognitionaswellasbraindiseasessuchasAlzheimer sepilepsyParkinson sanddepressionLeafLabstoolsforcharacterizingandanalyzinghigh channel
countelectrophysiologyrecordingswillallowresearcherstomoreeasilycollectandinterpret
neuraldataatalargescale