Research

Transgenic Tg(gata1:GFP) 48 hours post fertilization beating embryonic zebrafish heart. The white dots are red blood cells, which are fluorescent. Compare to the size of a human hair (∅ ≅ 100 μm) on the right. This movie was reconstructed by synchronizing timeseries of confocal microscopy slices. (See Liebling et al., J. Biomed. Opt. (2005). For more movies proceed directly to the supplementary data pages of our J. Biomed Opt. , Science, and Developmental Dynamics papers.)

Integrated Computational Microscopy for Dynamic, Multi-Modal, and In Vivo Cellular Imaging during Morphogenesis
The combination of fluorescence microscopy with a breadth of innovative molecular tools has opened new avenues in biology. High-resolution microscopes, customized fluorescent proteins or the completion of extensive atlases of whole animals that include both anatomical and genetic information iare just some of the recent milestones in this field.

Gathering data of similar extent and resolution but on single individuals, in vivo, at high speed, and over extended periods of time (which are all conditions that currently constitute major challenges) is a prerequisite in order to reach a systematic understanding of how tissue, or entire organs and organisms develop. A key to achieving new breakthroughs appears to be the effective integration of adaptive sample preparation techniques, imaging instrumentation, and novel algorithms for image acquisition, reconstruction, processing, and analysis all within a quantitative framework for biological modeling.

Based on these observations, I have focused my research interests on the development of novel computational imaging and image analysis procedures to specifically address how tissue grows and cells organize during vertebrate heart morphogenesis. There, an integrated approach to imaging and analysis is indispensable since following individual cells is made particularly difficult due to the motion of the heart as it beats, motion that can exceed that of the cells of interest with respect to each other by several orders of magnitude in speed and amplitude.

The thrust of my approach is to develop imaging paradigms that go beyond the traditional, disjoint image-then-analyze pipeline. Instead of optimizing each step by itself, I strive for a more horizontal integration of the imaging and analysis components to take advantage of their individual strengths (e.g. instrumentation over software and vice versa). This translates into the development of novel image acquisition and sampling schemes and computational image reconstruction procedures optimized for quantitative analysis and which take advantage of the wide range of possibilities offered by molecular labeling.

Sampling, Reconstruction, and Analysis Procedures for Dynamic, High-Resolution, and Multi-Modal Microscopy: Tools
An example of such an integrated approach to imaging is the one we developed in order to gain a better understanding of the embryonic heart development. There, we combined a fast confocal slit-scanning microscope, novel strategies for collecting and synchronizing cyclic image sequences to build volumes over the entire depth of the beating heart with high temporal and spatial resolution, and, finally, data analysis and reduction protocols for the systematic extraction of quantitative information to describe phenotype and function. Our imaging pipeline comprises a recently-developed confocal microscope (Zeiss LSM 5 LIVE) which allows for image acquisition at rapid frame-rates (e.g. 120 fps for images of size 512x512 pixels) yet remains too slow for real-time imaging of the heart in 3D. We have developed reconstruction strategies to retrieve 4D volumetric data from nongated 2D image sequences of the beating heart in living zebrafish embryos. Through the methods we developed, a shortcoming of the microscope could be overcome with an alternative acquisition procedure and computational reconstruction algorithms that take advantage of the cyclical movement of the heart.


Four-dimensional imaging: acquisition and reconstruction procedure. (a) Sequential acquisition of two-dimensional slices (confocal microscopy) as a time-series at increasing depths from a cyclically deformed object (cone). (b) Direct reconstruction is not possible from the nongated data. (c) Temporal alignment procedure (d) Reconstruction.

In Vivo Cellular Imaging for the Study of Morphogenesis in Dynamic Conditions: Applications
The development of the vertebrate heart is a subject of particular interest notably because defects during cardiogenesis are the leading form of congenital birth defects. Early in the course of its development, the heart is already functional and is beating. Recently, it has been shown that mechanical forces induced by blood flow in the beating heart influence its development. Some of the key problems are to determine how heart tissue forms while muscle contraction is at work. In particular, is it possible to specifically single out the influence of shear stress induced by hemodynamics along with the genes and molecular pathways that are involved, or determine how drugs and genetic modifications influence development and interfere with mechanical constraints? The advent of powerful microscopes, novel molecular labeling tools, and adapted animal models offer good prospects for answering these questions. One example is the zebrafish as an animal model. It is particularly well suited for these investigations since it is similar in important ways to other vertebrates yet reproduces rapidly and externally, its study has yielded numerous mutant and transgenic lines, and its embryos have excellent optical properties. Nevertheless, the inherent dynamic nature of the problem has made imaging in the heart difficult or even impossible without the availability of novel tools. It takes only one hundredth of a second for a point on a tissue that moves at a speed of one millimeter per second to cover a field of view of 10 μm, the typical size scale that is relevant for studying activity at the cellular level. Using currently available instrumentation, studies at such scales and speed are only possible via an integrated approach to image acquisition, reconstruction, and analysis.
Dynamic three-dimensional reconstruction of the developing Tg (gata1:GFP) zebrafish heart at different stages (28, 33, 60, and 111 hours post fertilization (hpf)) at mid-diastole. These renderings of the three-dimensional architecture at a single time point allow following the morphological changes that are apparent as the heart develops from a tube, then loops to form chambers and to finally become a mature heart. a: atrium, v: ventricle. Arrows indicate flow direction. Scale bar is 100μm

Wavelets
Much of my research is in close connection to wavelets or some of the various flavors of wavelet transforms. We developed wavelets and other basis functions tailored to the specificities of digital holography (Fresnelets) and computerized tomography. I'm particularly interested in the applications that are related to optics such as the modeling of optical systems, phase-retrieval, digital holography, interferometry, microscopy, or data analysis. I organized a special session on that topic, entitled Wavelets in Optics during the 2005 Wavelets XI conference, part of SPIE's Optics & Photonics symposium and I'm again a member of the programm committee of this year's edition, SPIE Optical Engineering + Applications 2007, Wavelets XII.
In 2003, I organized the Catch a Wavelet seminar series at EPFL.

	Dancing Fresnelets. Fresnelets (in red) are wavelet-like functions that are generated from B-splines. B-splines can be constructed using a weighted sum of shifted polynomials (shown in black and blue). Here, Fresnelets are generated from a B-spline of degree 0 (centered indicator function) and their parameter τ=√(λd), where λ is the wavelength and d the propagation distance, is varied from 0 to 4.5… and back.
	In this example, Fresnelets are generated from a B-spline of degree 1, that is, the ‘hat’ function (recognizable when τ=0).
	As the degree of the B-spline is increased—here, the degree is 3—it converges to a Gaussian function. When compared to the Fresnelets generated from B-splines of smaller degree, the spatial spreading is kept to a minimum as the propagation parameter τ increases.

More on Digital Holography, Fresnelets, and Computer Tomography
You can find more about my work on digital holography, Fresnelets and Computerized Tomography on my frozen EPFL research pages.