IMI Interdisciplinary Mathematics InstituteCollege of Arts and Sciences

Information in super-resolution microscopy and automated analysis of large-scale calcium imaging data

  • April 2, 2009
  • 3:30 p.m.
  • Sumwalt 102


Optical microscopy is a classic tool for dissecting neuronal circuits, which enabled the pioneering work of Ramón y Cajal and Golgi a century ago. Currently, a renaissance in optical imaging driven by advances in fluorescent probes, as well as high-resolution techniques for imaging in live animals is again poised to revolutionize our understanding of brain circuit structure and dynamics. Non-trivial computational and statistical approaches to processing microscopy data are central to some of these advanced techniques. In my talk I will discuss two ways in which a statistical analysis of imaging data enables using microscopy to test quantitative hypotheses, rather than simply to visualize biological structures. Using two-photon fluorescence imaging, neuroscientists can study Ca2+-dynamics within large numbers of cells, in anesthetized and awake behaving animals. However, unlike for electrophysiological studies standardized algorithms for assigning optically recorded signals to individual cells have not yet emerged. We developed an automated sorting procedure that combines independent component analysis and image segmentation for extracting cells’ locations and their dynamics with minimal human supervision. In validation studies using simulated data, automated sorting significantly improved estimation of cellular signals compared to conventional analysis based on image regions of interest. We used automated procedures to analyze data recorded by two-photon Ca2+-imaging in awake behaving mice. We focused on the brain circuit in the cerebellum, in an area that is critical for coordinating movement. Our studies reveal correlated activity in networks of neurons and glia, as well as details of the modulation of network activity during spontaneous running behavior. A second recent advance in optical microscopy has broken the classical “diffraction limit” by using single-molecule fluorescence imaging to obtain optical information with ~20 nm spatial resolution in 3 dimensions. Despite the proliferation of super-resolution techniques and applications to studying biological structure and dynamics on the nanometer scale, there is not yet a broadly useful definition of resolution that can compare the information obtained by particular methods. Using Fisher information and the related Cramér-Rao bound on estimation error, we provide a figure of merit for imaging techniques ranging from conventional imaging to true, single-molecule super-resolution. Our framework allows evaluation of the effect on image quality of parameters, such as the instantaneous density of activated emitters or the amount of time spent imaging each emitter. Our quantitative measure of the information content of microscopy techniques allows the design of optimal data processing and image reconstruction algorithms.

© Interdisciplinary Mathematics Institute | The University of South Carolina Board of Trustees | Webmaster