PhD and Postdoc Seminar
You have a new invitation for the PhD Seminar on 12 December 2017!

December 12, 2017

Featured images

Bernstein Center Freiburg

Tuesday, 12 December 2017


17:15 h


vCal iCal


Lecture Hall (ground floor)
Bernstein Center Freiburg
Hansastraße 9a
79104 Freiburg


Hosted by Benjamin Merkt

Neuron localization and classification from multi-electrode array recordings - A combined biophysical modeling and deep learning approach

Recording neural activity in vivo with dense extracellular polytrodes allow for extracting a multitude of information with high temporal and spa- tial resolution. Focusing on the extracellular action potentials (EAPs) of neurons, the cells can be putatively classified into excitatory or inhibitory units and roughly localized based on simple features like the spike shape and firing patterns. The presented method makes predictions about binary type (excitatory/inhibitory) of neurons and their soma position precisely and robustly based on various features of simulated EAPs. The approach uses Convolutional Neural Networks (CNNs) trained to classify neurons or localize their soma. The datasets of EAP recordings for training the net- works are generated by simulating multi-compartmental cell models from different layer-5 morphologies of the rat's somatosensory cortex combined with the line-source method for calculating the induced electric potential on the multi-electrode array. In case of localization, the CNN predicts the soma position robustly across various cell types with errors of the order 7 μm, where cells are placed randomly in front of the probe and have differ- ent alignments with respect to it. Classification of cells into excitatory and inhibitory units can be done with mean accuracies of 98 %. As the detail and fidelity of neural models increase and high-density recordings become available, this approach could become a viable and robust alternative for classification of neural cell types from in-vivo extracellular recordings.


All BCF members are cordially invited!



Hansastraße 9a
79104 Freiburg im Breisgau