guest ::
| Home > Publications database > Evaluation of spike sorting results |
| Home > Publications database > Evaluation of spike sorting results |
| Poster (Other) | FZJ-2016-04996 |
; ; ; ; ;
2016
Abstract: In Parkinson’s disease (PD) the STN plays an important role in the formation of pathological oscillatory activity within the basal ganglia-cortex loop. The primary measure to reveal such oscillations is the local field potential (LFP). While it is assumed that the LFP reflects synaptic input to groups of neurons, the relationship between this population signal and the single neuron activity is still a matter of debate [1, 2]. Our long-term goals are to investigate the spike-LFP relationship in STN recordings obtained during deep brain stimulation surgery, as well as to assess the amount of synchrony between individual neurons in order to elucidate how oscillations on the population level translate to neuronal synchrony.A critical step to achieve this goal is to correctly isolate the spiking activity of single units in extracellular STN recordings from Parkinson patients measured with a Ben Gun five channel micro-marcro-electrode holder. We employed a number of spike sorting algorithms [e.g., 3] and found that different spike sorting methods yield inconsistent results. We quantify these differences by the number of detected single units and the individual assignment of spikes to the detected units. Our long-term goal critically depends on the spike sorting quality [4], as, e.g., spike synchrony evaluation depends on the percentage of correctly identified spikes [5]. Hence, we introduced two additional approaches. Firstly, we developed a set of tools that estimates the isolation quality of single units [6]. These tools calculate the similarity of the spike shapes within one unit compared to other units. Secondly, we generated synthetic ground truth spike data of mixed units with the statistical features of the STN recordings: We selected the two most different spike shapes which we combined linearly to obtain pairs of spikes with a controlled dissimilarity. Assuming Poisson spike rates we generated spike trains by inserting such spike pairs into a noisy background obtained by phase shifting the original noise. These data enable us to calibrate and verify our spike sorting results, i.e., to check if the number of extracted units and the spike-to-unit assignment is correct. By use of these two approaches, we compare and evaluate various spike sorting methods to finally select and apply the most appropriate one for the analysis of our STN recordings.AcknowledgementsSupported by DFG GR 1753/3-1 Klinische Forschergruppe (KFO219, TP12), DFG GR 1753/4-1 Priority Program (SPP 1665), Helmholtz Portfolio Theme Supercomputing and Modeling for the Human Brain (SMHB), EU grant 604102 (Human Brain Project HBP).References[1] Moran et al. 2008 Brain 131[2] Weinberger et al. 2006 J Neurophysiol 96[3] Wild et al. 2012 J Neurosci Methods 203[4] Pazienti & Grün 2006 J Comput Neurosci 21[5] Lourens et al. 2012 Clinical Neurophysiol 124(5)[6] Hill et al. 2011 J Neurosci 31(24)
|
The record appears in these collections: |