Daniel Berger

Daniel Berger

Postdoctoral Fellow
Daniel  Berger

I am interested in understanding, from a computational standpoint, how brains process information. They do this in a large network of interconnected computational elements (neurons) which communicate with chemical and electrical signals. In mammalian cortex, individual neurons typically collect inputs from, and send outputs to, an area spanning several cubic millimeters. In that region their dendrites and axons meet millions of branches of other cells, some of which they make synaptic contacts with and some of which they ignore.

Much research has been conducted to try to understand this network, but the combination of its large scale and fine detail have made it notoriously difficult to investigate. Even very basic properties of the network remain unknown. What is the number of cell types involved and what is their function? Which aspects of network structure are genetically predetermined and which are adapted during development and learning? To what degree is synaptic connectivity random or follows rules, and what might these rules be?


Currently the only method that allows to follow all wires in a piece of cortex and to see all synapses they make is large-scale serial-section electron microscopy. The combination of large fields of view and very high resolution makes this a powerful new imaging technique. Though promising, this method faces many technical challenges. The key to making it work for very large volumes is robust automation.


To help scaling up the volumes so that entire circuits are contained within, I have worked on computational methods to process and analyze the large amounts of image data produced by serial-section electron microscopy. In particular I have worked on automating image acquisition on the electron microscope, alignment and stitching of large image stacks, computer-assisted manual image segmentation, 3D visualization, and automated analysis of structure and synaptic connectivity in labeled image stacks. This now allows us to deal with data sets in the 100 GB to TB range. Methods from this toolbox are now used in many collaborations to process samples ranging from mouse cortex, LGN, retina, muscle and cerebellum to C. Elegans and zebrafish.

Contact Information


Appointment Type