Advances in Electron Microscope Tomography

From NBCRwiki

Jump to: navigation, search


Progress and Prospects for Electron Microscope Tomography

In recent years the biological electron microscopy community has adapted technical improvements in instrumentation, data collection modes and large format digital image detectors. For example images on the order of 4K X 4K pixels are commonplace, and multiple detectors and montaging techniques make series of billion-pixel images achievable without extraordinary effort. As input to tomographic reconstruction this flood of image data plays an important role in determining the three-dimensional structure and function of cells and sub-cellular organelles. Structure may be elucidated across a wide range of spatial scales, ranging from that of neurons in the brain, for example, down to the scale of proteins and protein complexes. The scale and volume of this data, is in itself, a challenge to the transmission, reconstruction and storage of the processed data. In response to the requirements for high-quality three-dimensional reconstructions from EM data this branch of computer tomography is presently under a state of rapid development. Areas of future progress include instrumentation, data collection, reconstruction and other image processing techniques. Because of rapid advances in instrumentation and computer-based reconstruction, the routine imaging of molecular structure in context appears to be likely within the next few years.

Electron microscope (EM) tomography presents a number of special problems. The imagery is low contrast and noisy with limited sampling of projection directions; sample warping and the curvilinearity of electron trajectories make classical techniques of x-ray tomography problematic; and the volume and scale of the data make automated preprocessing and image segmentation necessary. The need for solution of these problems has spurred the introduction of new techniques into EM tomography. For example, the presence of geometric nonlinearities in the basic ray transform requires that the inversion problem be treated in terms of Fourier integral operators rather than Fourier transforms. Recently Bayesian techniques and Markov random field techniques have been applied to other steps in the tomographic reconstruction process such as feature extraction, tracking, alignment, and post processing steps. Progress in all aspects of EM tomography requires deeper mathematical understanding, and the introduction of new algorithms and techniques from computer science.

Up to the present, most conferences on EM tomography have concentrated on the biology. This is also true for more permanent publication such as books and papers, and is in contrast to related areas such as X-ray tomography where the technical aspects have a well-developed presence. Nevertheless, both the instrumentation and computational aspects of EM tomography have shown rapid progress, and this discipline is developing a foundation of its own, in both the mathematics and the engineering disciplines. Therefore, we believe that now is a good time to initiate the exchange ideas on the mathematical foundations, computer science and engineering.

Although we will emphasize applications of EM tomography to the biological sciences, materials science also makes extensive use of this technology. Electron tomography in the materials sciences is a rapidly advancing area of scientific endeavor and affords many opportunities for collaborative activities. (Electron Tomography for Nanoscale Materials Science)

A tutorial on EM tomography, including sessions on instrumentation and the software package TxBR was presented at the ISBI 2009 Conference.

Tomography Day at UCSD

Group Photo for Attendees at the Calit2 Quad

Topic 1. The radon transform and its cousins

The purpose of this session is to introduce the basics of the generalized Radon transform as it arises in EM tomography and it's associated inverse problem. We will cover applications of microlocal analysis and the Fourier integral operator.

Topic 2. The AI problems associated with tomography: tracking and segmentation.

The purpose of this session is to give a quick review of feature extraction, tracking and image segmentation. We will then discuss why these techniques are necessary to the tomographic process, what information we must extract from the data, and the special problems of dealing with EM data. We will also cover some new techniques from graph theory and Markov random fields, as time permits. New results on fiducial free alignment based on extended structures in the EM images will be presented at the CSIE 2009 Conference by Sebastien Phan. A preliminary presentation on alignment on points and one dimensional structures is available. This method is an extension of occluded contour techniques.

Session 3. Algorithms and architectures

The purpose of this session is to review the current state of EM tomography packages, future requirements for speed and computational resources, and new developments in computer hardware such as graphical processor units which may be applied to the reconstruction and data processing problems. Matching architectures to algorithms promises to be an active area of development. GPU boards in workstations can yield substantial performance gains in tomographic applications. Results obtained by researchers at NCMIR will also be presented at the CSIE 2009 Conference. Information relating to new developments in 2011 may be found at the following link: presentation on contour alignment, montaging and applications of tropical semirings.

Topic 4. Tomography, Optics and Instrumentation

The purpose of this session is to provide a quick introduction to the practical side electron microscope tomography followed by a presentation of recent advances in tomography techniques and instrumentation. Emphasis will be on the methods of acquisition of data covering a wide range of spatial scales. A tutorial on instrumentation will be presented at the ISBI 2009 Conference.

Recent Developments

Multiscale Phenomena

EM images span spatial scales ranging from a fraction of a nanometer to about 50 micrometers, and typical 3D reconstructions may cover 1 / 10^15 of the volume of a typical optical microscope reconstruction. The limit of resolution of light microscopy is on the order of a hundred nanometers, and even though super resolution techniques applied in “pointilliste optical tomographymay give much better resolution, but rare events and contextual information are missed.

In order to bridge the gap, the spatial range of the electron microscope has been expanded by various techniques. Large sensor arrays and wide-field camera assemblies have increased the field dimensions by a factor of ten over the past decade, with a further 10X expansion possible and new techniques for serial tomography (z-axis) and montaging (x and y axes) make possible the assembly of tens of thousands of three dimensional reconstructions. Even though we are far from closing the spatial scale gap with a single experimental preparation, the amount of data generated by this enterprise threatens to overwhelm computational resources. A single data set for tomographic processing may have as many as 360 images, each 8K by 8K. Simple backprojection is order N^3 so a data set of this size requires about 10^13 or more coordinate evaluations. Because coordinate evaluations involve nonlinear functions in our case, the number of elementary operations is hundreds of times more, and bit error rates require double precision. The number of reconstruction volumes necessary to image an average cell down to the protein assembly level is of the order 10^4 , so counting exponents, we are into the exascale range. Data sets for this are well into the petascale range.

The present state of the art depends on development of new algorithmic techniques in addition to the scientific instrumentation and computer hardware which permit the experimenter to obtain the data and run the reconstructions. We appeal to a particular example which makes this point. TxBR runs in 4 days instead of 4 years on the particular data set mentioned above because the evaluation of high degree polynomials on a regular grid can be reduced to a simple recursion consisting of additions, and that the number of additions is linear (rather than exponential) in the degree of the polynomial. This, can be performed on a workstation with several GPU cards. Furthermore, the recursion can be made parallel to a very high degree, and the algorithm can be mapped to the simplified processors comprising, for example, a graphics processor unit. Programming this on a GPU board reduces the 4 days to less than 4 hours.

Stochastic Geometry and Multiscale Biology

What is Stochastic Geometry?

Stochastic geometry, or geometrical probability, designates probability problems which incorporate geometrical figures as the 'outcomes' or 'data'. Random positions of a geometrical figure are the center of the classical theory. Applications range from astronomy to cell biology and materials science. Additionally the probabilistic foundations are connected to other branches of pure mathematics. Scientific problems may have a geometrical content in the data, in sampling, in theoretical models, or in all three.

Examples of geometrical data in biology include: maps of neurons, proximity data for interacting macromolecules, particle counts, configurations of linear structures such as filaments, shapes of ultrastructural components, or morphological changes as the result of experimental manipulation or disease, and derived measures such particle counts, length ,surface area, and volumes.

Even the question of summarizing data may be nontrivial. Many experimental sampling techniques can be described as geometrical. Spatial variation of cell types or morphological changes might be sampled along a linear transect, inside an arbitrary grid square or within a fixed radius of an arbitrary reference point.

Solid structures of biological tissues are frequently studied by taking thin sections for microscopic examination. Here the deep problem of relating solid geometry to plane sections is compounded with questions of sampling variability, bias, choice of sampling design and inference. The multidisciplinary science of stereology deals mainly with sectioning problems and is an important area of application for stochastic geometry. Finally, idealized models of natural phenomena often invoke geometrical probability. These tend to be gross simplifications of complex processes, such as the packing of astrocytes in the central nervous system or the attachment of antibodies to a virus or the distribution of fluorescent markers along proteins along filaments, or embedded in biological membranes.

Imaging and Stochastic Geometry

New advanced experiments give access to the molecular level. These inclued including superresolution fluorescence microscopy, cryoelectron microscopy (cryo-EM), and EM tomography on multiple tilt series. These imaging technologies require the development of new computer-based mathematical and statistical methods of analysis. For example cryo-EM and large scale averaging with traditional plastic sections has experienced several successes in recent years but is now facing a number of problems which can only be solved in collaboration with mathematicians, statisticians and computer scientists.

Because of the rapid development of imaging technologies extending the spatial range to molecular scales advanced research world-wide has found a new focus on fundamental issues. This research is inspired by the need to develop new stochastic geometry methods with the ultimate purpose of establishing a mathematical and statistical fundamental to computer based analysis of advanced bioimaging data. These efforts take place at the boundaries between mathematics, statistics, computer science and bioscience.

Multiscale Biology. Biological processes take place on a wide range of spatial and temporal scales, from the molecular to the whole organism. Ongoing work in light and electron microscopy has extend the scale available for imaging to the level of large molecules and molecular assemblies. In addition to the structural information available trom electron microscopy, fluorescence microscopy has added dynamical information as well as information about the location of specific proteins. Modern staining techniques also can locate the site of fluorescence within nanometers. We may exploit the opportunities created by these two imaging technologies in several diverse areas.

Correlated microscopies. The first problem is to correlate the information gained by light and electron microscopy. The immediate task is to overlay images or sections of 3D reconstructions so that identifiable structures are in geometric correspondence. This task is complicated by the range of spatial scales and the nature of the images at the various spatial scales. We discuss the problems of correlation in the sections describing work on 3D reconstruction.

Uncertainty quantification.  Microscopy is Inherently probablistic. Binding and staining are the result of random processes, and florescence itself is subject to quantum uncertainty at the molecular level. These lead to uncertainty in both position and magnitude or relevant physical events. A second source of error is due to uncertainty in the trajectory of the interrogating radiation. Paths of electrons through samples and lenses are subject to local fluctuations in electrical and magnetic fields, and light paths are affected by varying optical densities across material interfaces. A third source of error is in the manual marking of structures for alignment. As microscopy becomes more quantitative, in terms of measurement of structure and dynamics of biological processes the need for identification and quantification of sources of uncertainty becomes more acute.

Systems biology.  Decades of work on the metabolic processes has resulted in the construction of network models relating the chemical interactions and transformations in living cells. There is increasing evidence that this chemical activity is compartmentalized and that geometric proximity is a large factor determining whether specific reactions take place. Accordingly it is necessary to map the metabolic network to the spatial reconstructions available from light and electron microscopy. This would entail the introduction of new mathematical structures into theoretical biology in order to go beyond the database approaches currently in vogue.


In this section we discuss the rationale behind stereology applications in electron microscopy and review our demonstration of stereology software developed at NCMIR. It should be noted that all microscopy methods can be regarded as sampling methods, whether or not the process is guided to capture structures or events of interest. As such, the development of stereology has been shaped by this circumstance and the probabilistic theory of sampling.

Stereological tools are standard for accurate (i.e., unbiased) and precise quanti?cation of any microscopic sample. Research efforts during the past decades have provided a broad spectrum of tools to estimate a variety of parameters such as volumes, surfaces, lengths, and numbers. Current research has focused of extending this to derived geometric measures, such as mean curvature and other related quantities. These techniques require sets of parallel sections that can be produced by either physical or optical sectioning, with optical sectioning being much more e?cient when applicable. At present various EM techniques are available for stereologic applications. These include transmission and the so-called block face EM.

Transmission electron microscopy can not fully exploit image data directly, mainly because of the large depth of ?eld.. Electron tomography yields stacks of slices from electron microscopic sections. With the development of high resolution techniques for plastic fixed objects, parallel reconstructed slices of small thickness (2–5 nm axial resolution) can be produced from EM. These optical slices minimize problems related to overprojection e?ects, and allow for direct stereological analysis, e.g., volume estimation with the Cavalieri principle and number estimation with the optical disector method.

Block face comprises a second class of techniques. While electron microscopy has been around for eight decades, since it’s invention in the 1930’s, the technique of Serial Block face Scanning Elecron Microscopy (SBFSEM) has been made practical (Denk2004) and gained popularity in the electron microscopy only recently (EM) community (refs). In this technique an ultramicrotome is inclueded within a scanning electron microscope (SEM) and is able to remove slices as thin as 40 nm from a block of tissue embedded in resin and typically stained with heavy metals during a process of freeze substitution (ref). By scanning the exposed block face between successive slices, SBFSEM is relatively easy to use and then able to image large volumes of tissue without human intervention.

Using modern machines such as the 2 keV FEI Merlin with Gatan 3View cutting system, this technique can image over a terabyte worth of data per week when left running unsupervised (ref). Such datasets can span as much as 32 x 24k pixel and thousands of slices (Fig 1a), with an optical resolution as low as 2 nm and pixel size as low as 0.5 nm in the XY dimensions. Depending on pixel size, SBFSEM is able to capture on the order of hundreds of microns in XY and Z dimensions, and thus enough to image many cells simultaneously from a specimen of tissue or cultured cells (Fig 1b).

Such SBFSEM 3D images represent a new precedent in the volume and size (multiple terabytes/week) in EM, but also represent an even bigger bottleneck between image acquisition/reconstruction and segmentation/analysis. Despite advances in automated segmentation methods, accurate 3D segmentation of complex tissue, such as the primary cells and neuropil tissue in Fig 1a, requires manual tracing of contour lines on every slice in order to achieve accurate segmentation and correct surface topology of every visible compartment. Manual segmentation in this was is very slow, on the order of seconds per contour, and approximately three orders of magnitude slower than image acquisition. Using modern SBFSEM, a single terabyte of complex tissue can be acquired in approximately three days but would require tens of thousands of man-hours (decades), for complete manual segmentation of all cellular and sub-cellular compartments. With such volumes it has become infeasible for individuals or even groups to fully segment an entire dataset, and thus the only way to accurately quantify the content of these datasets is using rapid randomly sampling methods like the tools offered by stereology.

We have demonstrated the use of stereology on 3D serial block face scanning electron microscopy (SBFSEM) to acquire highly accurate quantification information of large tissue samples all the way down to individual cells. In particular we demonstrate the benefits of applying stereology to such 3D SBFSEM volumes of tissue, most notably:

(1) a small (~5-10 nm for 2 keV) penetration depth, to reduce over-projection (40-60nm) errors experienced in thin serial sections imaged with transmission electron microscopes,

(2) the ability to identify questionable structures by checking adjacent Z-slices, thus decreasing misclassification errors, and

(3) the versatility to project a 3D stereology grids over single cells, thus allowing quantification of individual cells in situations a biologist needs more than just a flat average over tissue.

To test results, we manually segmented an entire neural soma and use point counting stereology of just 1000 points yields a percentage volume of ±0.3% for each compartment measured, and this decreased error predictably when more points were added. To perform stereology we created a “slash stereology plugin” which we have contributed to the IMOD software package, and could use to count and average of about 3000 points per hour over a 3D grid of any scale within our massive 3D SBFSEM volume.

Exascale Computing

Because of the necessity to bridge the gap between electron and light microscopy, we expect that many more orders of magnitude improvement in computational capabilities will be required in the coming decades. Exascale computing, on the other hand, will raise a new set of problems, associated with component energy requirements (cost per operation and costs of data transfer) and heat dissipation issues. As energy per operation is driven down, reliability decreases, which in turn raises difficult problems in validation of computer models (is the algorithmic approach faithful to physical reality), and verification of codes (is the computation reliably correct and replicable)? Leaving aside the hardware issues, many of these problems will require new mathematical and algorithmic approaches, including, potentially, a re-evaluation of the Turing model of computation.

Tomography and Pure Mathematics

Tomography, or more properly from a mathematician's point of view, integral geometry is a core area in mathematics. One may find applications in many areas of mathematics, physics, and engineering. Inverse transport, for example, is an obvious generalization of inversion of the Radon transform, and inverse transport is encountered in many areas of biomedical imaging. Another area of mathematics which shows up, surprisingly, in areas as diverse as electron microscopy (as originally noted by Peter Hawkes in the 1990s) and by brain modelers is topical algebra. In tropical algebra, the mathematical operations of product and sum are replaced by the simpler operations of sum and maximum, respectively. Study of this algebra has recently had a major impact in theoretical mathematics.

Recent Software Developments

Alignment of the individual images of a tilt series is a critical step in obtaining high-quality electron microscope reconstructions. We are continuing research into techniques for producing good alignments, and utilizing the alignment data in subsequent reconstruction steps. Our alignment techniques utilize bundle adjustment. Bundle adjustment is the simultaneous calculation of the position of distinguished markers in the object space and the transforms of these markers to their positions in the observed images, along the bundle of particle trajectories along which the object is projected to each EM image. Bundle adjustment techniques are general enough to encompass the computation of linear, projective or nonlinear transforms for backprojection, and can compensate for curvilinear trajectories through the object, sample warping, and optical aberration. This research has resulted in an advanced code "Transform-Based Tracking, Bundle Adjustment and Reconstruction" (TxBR).

Iterative Methods

Until recently main two approaches to 3D reconstruction have been employed, electron tomography (ET) and single particle analysis. In ET, the same sample volume is observed multiple times under many different viewing angles and the set of images is reconstructed via an implementation of the inverse Radon transform, for example, with the filtered back-projection scheme. Alternatively, in single particle analysis, a collection of numerous snapshots of objects of the same 3D structure having random orientations are extracted from one or more micrographs and transformed via single-particle tomography to provide   3D structure of the generic object at molecular or near molecular scale. While the first approach is prone to reconstruction artifacts and focuses on wider scale phenomena involving many biological entities, the latter offers high-resolution details on the molecular structure of simple small individual objects. The development of TxBR has enabled a hybrid method between these two approaches. We have found that given that proper reconstruction methods are employed a plastic embedded biological specimen can actually be submitted to a rather substantial electron beam dosage, through multiple angles of exposure, without compromising the subsequent reconstruction. Thus the number of electron micrographs of the same area may be significantly increased by more than an order of magnitude compared to common practice in ET. Projections may contribute to a better averaging scheme so higher quality reconstructions can be produced.  The reconstruction method should incorporate corrections to the possible sample distortions and limit any processing artifacts. Implementing this method requires highly automated procedures, both for the data acquisition and the image processing.

Fluorescence and Electron Microscopies

A third theme related to molecular scale resolution comes from fluorescence microscopy.  Although the spatial discrimination along the optical axis is an order of magnitude worse than that afforded by electron tomography, with superresolution techniques spatial discrimination between active molecules in the perpendicular plane is comparable to ET techniques. Because of the chemistry, which is described in other sections of this proposal, fluorescence microscopy makes functional and dynamical information available, particularly in living cells. Unfortunately, because of the specificity, only the active molecular sites are imaged, which at nanometer resolution, gives images of isolated points of light.  Staining techniques, can, however, prepare the sample for electron microscope imaging, and with some effort, sites associated with fluorescence identified, and the surrounding ultrastructure revealed for subsequent analysis.  Nevertheless, some part of the information available at the light microscope level is lost.  Preservation and utilization of the coordinates of active points of light would be useful on several levels during the process of reconstruction and analysis. This is particularly true in cases where large scale imaging and reconstruction is necessary for the elucidation of extended spatial structure and the statistical analysis of large numbers of ultrastructural components.

Coordinated Microscopies

As an initial essay in the arena of coordinated light and electron microscopy (LM and EM) we have identified two key areas where LM information would be useful.  The first area is in the alignment of EM images for 3D reconstruction, and the assembly of many reconstructions for montaging and (possibly) assembly of serial sections. Fluorescence can outline extended structures, and this information can be used to supplement or replace the usual alignment data afforded by deposited particles.  The second area of application is the natural extension of these ideas to the post reconstruction analysis.  Pointwise outlines afforded by light microscope data can be of use in segmentation, and the coordinatization of the active sites has applications in stereology.  One immediate question is whether all of the ultrastructural components of a given species carry the active sites identified by LM. The scale of the data processing problem posed by the analysis of whole or multicellular data at the molecular level is somewhat daunting. The incorporation of information from LM should simplify or streamline the data processing task at subsequent steps in the workflow from LM to EM to reconstruction and finally to the analysis and assembly of statistical data.  Without constraining the volume of the data in some fashion, many of the data processing problems posed by spatial systems biology are in the exascale range.  Clever application of the information from LM may reduce the magnitude of the task by orders of magnitude.

TxBR is available for download from the NCMIR web site. The most recent version is fully automated for use with samples incorporating gold bead markers. This software can be used to assemble large volumes through a combination of serial sectioning amd montaging techniques. Modules for alignment on internal structure, noise reduction and artifact suppression are under development, with plans to add these to the package in 2012. Iterative techniques based on curvilinear backprojection are also under development at NBCR and NCMIR.

Related Links

National Biomedical Computation Resource

National Center for Microscopy Imaging Research

National Center for Research Resources

NBCR Summer Institute 2008



References of Interest

The Landscape of Parallel Computing Research

HPC Wire Interview of two authors of The View from Berkeley

Electron Microscope Tomography

National Center for Microscopy and Imaging Research

Tomography at NCMIR 2009

Problems in Tomography

The Boulder Laboratory for 3-D Electron Microscopy of Cells

Electron Tomography

Transform Based Backprojection

Local Tomography in Electron Microscopy

Inversion of the X-ray transform from limited angle parallel beam region of interest data with applications to electron tomography

Xmipp, "X-Window-based Microscopy Image Processing Package"

Lambda Tomography

A. Faridani, D. V. Finch, E. L. Ritman, and K. T. Smith, Local tomography II, SIAM J. Appl. Math., 57 (1997), pp. 1095-1127.

A. Faridani, E. L. Ritman, and K. T. Smith, Local tomography, SIAM J. Appl. Math., 52 (1992), pp. 459-484.


Markov Random Field Methods

Belief Propagation

Pattern Recognition and Machine Learning Figure downloads, chapter 8 text.

Bundle Adjustment

Euclidean Reconstruction from almost Uncalibrated Cameras

Bundle Adjustment-A Modern Synthesis

Parallel Processing

Graphics Processor Units

The Telescience Project

Mathematical Foundations

G. Beylkin, “The Inversion Problem and Applications of the Generalized Radon Transform”. Comm. Pure Appl. Math., 1984, 37, 579-599

J. Bruning and V. W. Guillemin, Mathematics Past and Present, Fourier Integral Operators, Springer-Verlag, Berlin, 1994.

L Ehrenpreis, "The Universality of the Radon Transform", Clarendon Press, Oxford, 2003.

O. Faugeras and Q.-T. Luong, The Geometry of Multiple Images, MIT Press, Cambridge Mass, 2001.

I. M. Gelfand, S. G. Gindinkin and M. I. Graev, Selected Topics in Integral Geometry, American Mathematical Society, Providence, Rhode Island, 2003.

V. Guillemin, “On Some Results of Gelfand in Integral Geometry”, Proc. Symp. Pure Math., 43, 1985, pp. 149-155.

F. W. Hawkes and E. Kaspar, "Wave Optics", Academic Press, San Diego, 1994.

F. Natterer and F. Wuebbeling, "Mathematical Methods in Image Reconstruction", SIAM, Philadelphia, 2001.

V. P. Palamodov, "Reconstructive Integral Geometry", Birkhäuser, 2004

V. P. Palamodov, "A uniform Reconstruction Formula in Integral Geometry", Arxiv, November 11, 2011.

E. T. Quinto, “An Introduction to X-ray Tomography and Radon Transforms”, Proc. Symp. Pure Math., 67 2006, pp. 1-24.

H. Romer, "Theoretical Optics", Wiley-VCH, Weinheim, 2005.

A. G. Ramm and A. I. Katsevich, "The Radon Transform and Local Tomography", CRC Press, New York, 1996.

V. A. Sharafutdinov, "Integral Geometry of Tensor Fields", VSP, Utrecht, The Netherlands, 1994.

G. Uhlman, editor, "Inside Out", MSRI Publications, Cambridge University Press, 2003.

V. V. Volchkov, "Integral Geometry and Convolution Equations", Kluwer Academic Publishers, Dordrecht, 2003.

Regularization Methods

High noise, low contrast images require special treatment. This is especially true when EM tomography is used to elucidate structure at the level of protein complexes. Regularization methods have proven to be especially effective when dealing with ill-posed problems with additional unknown parameters. Thanks to Ozan Oktem for this material.

Meetings of Interest to Tomographers

Mathematical Sciences Research Institute August-December 2010 Program on Inverse Problems and Applications.

Isaac Newton Institute for Mathematical Sciences July-December 2011 Program on Inverse Problems

7th International Electron Tomography Conference Conference 2014 Ultrastructural Biology

Recent Presentations on EM Tomography

Tutorial on Large Field Electron Microscope Tomography, Part I, presented at IEEE Symposium on Biomedical Imaging, Boston, Massachusetts, June 28, 2009.

Tutorial on Large Field Electron Microscope Tomography, Part II, presented at IEEE Symposium on Biomedical Imaging, Boston, Massachusetts, June 28, 2009.

Large Field and High Resolution Electron Microscope Tomography, presented at Banff workshop 09w5017: Mathematical Methods in Emerging Modalities of Medical Imaging, Oct 25 - Oct 30, 2009.

Recent Developments in Large Scale Electron Microscope Tomography, presented at the NBCR Tech Series: Multiscale and Multi-Physics Tools for Mesoscale Modeling of Heart Diseases using Realistic Geometry, November 18 2010

Sixth International Congress on Electron Tomography 2011 Cryo-Electron Tomography


Abel Lin, NCMIR, University of California, San Diego

Adel Faridani, Oregon State University

Alexandre Cunha, California Institute of Technology

Andreas Rieder, Mathematics, Universität Karlsruhe

Bill Tivol, California Institute of Technology

Farshid Moussavi, Stanford University

Fernando Amat, Stanford University

Hans Rullgard, Mathematics, Stockholm University

James Bouwer, NCMIR, University of California, San Diego

Jane Ding, California Institute of Technology

Jian Shi, California Institute of Technology

Laurent Desbat, TIMC-IMAG, Joseph Fourier University of Grenoble

Mariana Tihova, City of Hope

Mark Ellisman, NCMIR, University of California, San Diego

Mart Malle, Computer Science, University of California, Riverside

Masako Terada, NCMIR, University of California, San Diego

Michael Holst, Mathematics, University of California, San Diego

Neal Young, Computer Science, University of California, Riverside

Peter Arzberger, NBCR, University of California, San Diego

Raj Singh, NCMIR, University of California, San Diego

Rick Giuly, NCMIR, University of California, San Diego

Ryan Hass, Mathematics, Oregon State Univeristy

Sebastien Phan, NCMIR, University of California, San Diego

Stephen Larson, NCMIR, University of California, San Diego

Steve Peltier, NCMIR, University of California, San Diego

Thomas Payne, Computer Science, University of California, Riverside

Todd Quinto, Mathematics, Tufts University

Wilfred Li, NBCR, University of California, San Diego

Zeyun Yu, Mathematics, University of California, San Diego


Albert Lawrence albert (dot) rick (dot) lawrence {at} gmail {dot} com, Sebastien Phan sph {at} ncmir {dot} ucsd {dot} edu

Special Thanks

Peter Arzberger and Mark Ellisman for organizational support.

Personal tools