Parts | Days | Selection | Search | Updates | Downloads | Help
BP: Fachverband Biologische Physik
BP 14: Posters: Neurobiophysics, Theoretical Neuroscience, Sensory Transduction
BP 14.3: Poster
Monday, March 14, 2011, 17:15–20:00, P3
Analyzing Multi-electrode Array Measurements of Neurons — •Stephan Kramer1, Kai Bröking2, and Annette Witt2 — 1Institut f. Numerische u. Angewandte Mathematik, Uni Göttingen — 2Max-Planck-Institut f. Dynamik u. Selbstorganisation, Göttingen
Measurements from neuronal networks cultured on multi-electrode arrays (MEAs) yield noisy time series of the extracellular potential. As each electrode records signals from multiple neurons a principle component analysis followed by a cluster finding analysis is performed to be able to assign spikes to neurons [1]. Although this procedure can be formulated by means of the basic linear algebraic subroutines (BLAS) library the large amount of raw data requires to investigate non-standard hardware like GPUs to achieve best performance. Due to the inherent hardware-dependence of most BLAS libraries programming effort can only be minimized by abstracting the algorithm employed from BLAS and hence hardware specifics. We show how to resolve this dependency by designing a C++-based domain-specific embedded language [2] so that algorithms can be stated in a hardware-independent, compact vectorized form. We discuss the performance of the algorithm proposed in [1] on different kinds of hardware architectures for a particular example (10000 spikes emitted by several hundred neurons).
[1] S. Shoham et al., 2003. Robust, automatic spike sorting using mixtures of multivariate t-distributions. JNM 127 (2), 111 - 122
[2] D. Abrahams, A. Gurtovoy, 2004. C++ Template Metaprogramming: Concepts, Tools, and Techniques from Boost and Beyond, Addison-Wesley