New Imaging Center Pools the Know-How of Five EPFL Schools

Imaging is an important tool for researchers in nearly all fields, whether to unlock the secrets of atoms or explore the origins of our galaxy. Nearly a quarter of all EPFL labs use imaging systems to study living cells, environmental phenomena, next-generation materials, and even outer space, among others. These systems let scientists observe what cannot be seen with the naked eye. However, a great deal of research also goes into the development of the systems themselves, such as the creation of new methods to capture images, improve the algorithms used to process them, and manage the reams of data generated as a result. Until now, skills in these areas were spread across five different EPFL schools with no single forum for the sharing of information, the formation of cross-disciplinary research groups, and the communication of discoveries—hence the launch of the new EPFL Center for Imaging (ECI). This center, which covers the five schools, avoids the siloing of the teams of experts, provides a single source of funding for cross-disciplinary research projects, and offers support and training services to its users. It also reinforces EPFL’s best-in-class reputation for imaging technology.

Leveraging Synergies Across Nearly 100 Labs

Designed to be an incubator of new ideas and cross-cutting R&D, ECI brings together nearly a hundred labs working in imaging-related fields. For some of them such as life science and microengineering, the need for imaging is obvious; for others like environmental engineering, materials science, and artificial intelligence, the need appears less immediate. Thus, ECI recently issued a call for proposals for cross-disciplinary research projects to foster collaborative ventures: It is also putting together workshops and seminars, including series to showcase discoveries made by scientists around the world. ECI hopes that, by building bridges among the various EPFL units, it will be easier to teach imaging-related topics to students, with greater consistency in their classes and projects. “EPFL is a world-caliber institution when it comes to imaging technology, meaning we’re well positioned to give our students a top-notch education,” says Michael Unser, ECI’s Academic Director. The center is run by a board of representatives from each of the five schools: Suliana Manley from SB; Mackenzie Mathis from SV; Christophe Moser from STI; Devis Tuia from ENAC; and Sabine Süsstrunk from IC. They have all agreed to take the lead in this pioneering cross-school initiative.

Imaging Technology Is Getting Smarter and Smarter

Remarkable advancements have been made in imaging technology of all kinds, regardless of the application. These advancements are being driven by new developments in hardware, data storage, and data processing. At EPFL, researchers have access to state-of-the-art equipment—such as that provided by partner organizations including the BIOP bioimaging platform, the CIME research center, and the CIBM and DCI imaging centers—to capture images, collect data, and measure the effects of real-world phenomena with unprecedented spatial and temporal resolution. For Michaël Unser, “The Nobel Prizes awarded for breakthroughs in high-resolution microscopy in 2014 and in cryo-electron microscopy in 2017 were both the result of several labs working together, which underscores how important cross-disciplinary research is in the field of imaging—and how important that field is to the entire scientific community.”

Fertile Ground for Cross-Disciplinary Research

Modern imaging technologies are being developed at a rapid pace and have all one thing in common: big data. These data cannot be taken advantage of without sophisticated algorithms or the skills of experts. ECI therefore aims to provide such resources to the many labs and centers that do not already have them. “The complex nature of imaging technologies means that it is a demanding task to deploy new methods and discoveries among researchers,” says Laurène Donati, ECI’s Executive Director. The center is in the process of setting up a R&D team to develop advanced methods for image analysis and reconstruction that will be made available through open science.

A cross-disciplinary team of researchers has already developed one such method, available in a program called DeepimageJ that helps imaging-system developers to take advantage of the benefits of deep learning. Deep learning is playing a growing role in imaging technology, where algorithms are trained to sort through thousands of detailed images in the blink of an eye. But the creation of deep-learning tools often requires advanced skills in computer science. DeepimageJ—a user-friendly, open-source plugin—gets around that obstacle by allowing laymen to easily perform the image-processing tasks commonly used in life-science research, such as pixel and object classification, instance segmentation, denoising, and virtual staining. It also has features that allow system developers to easily include new models. DeepimageJ is just one example of the advantages of cross-disciplinary teamwork. “Imaging experts are excited about the opportunities being created by deep-learning algorithms. And the developers of those algorithms need large datasets to train their programs. They’re particularly interested in projects involving real-world scenarios,” says Daniel Sage, an EPFL engineer and one of the DeepimageJ developers.

Student Projects at the Frontier of Imaging Research

Student projects that address concrete problems generated by multiple actors in imaging are another excellent mechanism to promote cross-fertilization across groups. A good example relates to the Master project carried out by Quentin Juppet, an EPFL student in computer science. His project involved coordinating the work being done at BIOP, Unser’s lab at STI, and two SV labs headed by Cathrin Brisken and Martin Weigert. Building on Weigert’s popular open-source StartDist software, Juppet developed a machine-learning algorithm that can distinguish mouse cells from human ones in a tissue sample. Juppet’s findings, published in Journal of Mammary Gland Biology and Neoplasia, can speed the pace of cancer research by automating the cell-classification step. According to Olivier Burri, a BIOP engineer who worked with Juppet: “Thanks to the streamlined communication, we were able to make our microscopic equipment available to the biologists who need it, give them access to the software developed by computational bioengineers, and let them benefit from the experience and oversight of R&D engineers.”