Bioimage Analysis 2014 at EMBL, a recap.
Posted by Julian Sosnik, on 18 June 2014
Last month 39 people from around the world gathered together in the flagship European Molecular Biology Laboratory (EMBL) in Heidelberg, Germany to take part in the Master Course on Bioimage Data Analysis. This was the third edition of the course that had previously been held in Heidelberg and Barcelona, and is aimed at training scientists to meet the growing need to extract measurable and quantitative data from biological images. In this latest incarnation there was a meeting component (invited speaker talks, short talks selected from abstracts and a poster session), a practical component, and a strong “community building” component.
Takeo Kanade from the robotics institute at Carnegie Mellon University started off the course with a fascinating keynote lecture that not only explained some technical aspects of segmentation of biological images, he also gave a great review of the key elements that make a computer capable of “seeing”. In addition, lectures were presented by pioneering researchers such as Fred Hemprecht from the Heidelberg Collaboratory for Image Processing (HCI), Ivo Sbalzarini from the Max Plank Institute of Molecular Cell Biology and Genetics, Nadine Peyriéras from the CNRS, France and Christophe Zimmer from the Institut Pasteur. Diverse aspects of imaging were presented in these lectures, as well as radically different approaches to image analysis. However, all of them had a common thread; how do we teach a computer to “see” the information contained in digital image so that we can quantify this information in a way that is biologically meaningful.
Two underlying themes emerged from these talks. The first focused on how scientists can more easily make computers “understand” what is that they are “seeing” by providing the computers with models of both, the world, and how it appears on the other side of a microscope. Giving computers both of these models as well as the laws that rule them (-i.e. a cell can only divide into two or a nucleus cannot leave a cell-) allows computers to “see” better, requiring less human intervention and data curation. The second theme focused on how to make software designed to quantify images more user friendly. These talks focused on how tool creators (those that design the software) are minimizing the need to change a variety of parameters inside the different tools (known as parameterize) in order to obtain the desired outcome. This is important because unless the user really understands how the tool was implemented, using an application resembles more a leap of faith than a scientific decision. By incorporating computer learning with representative training sets, programmers and tool developers are simplifying the way in which the user applies a given tool, having full control of the quantification process.
The classroom component of the course consisted of structured exercises in which the students learned how to apply the different strategies for segmentation and data analysis using real biological data. Students used the image analysis platform Fiji (ImageJ) and Matlab, and R for part of the analysis. With these tools, students implemented workflows that exemplify some of the most common tasks in image analysis such as the segmentation and visualization of complex 3D structures, tracking particles and cell movements, and analysis of the speed and directionality of biological movements big and small, among others. In addition, students got “insider information” from the experts in the field, regarding how and why they use different strategies (plugins, tools and macros) to accomplish the desired image restoration and segmentation required to quantify a given feature.
Finally, but not less importantly, the attendees got to present their work in short talks and a poster session. This component of the course had a dual goal. On one hand it exemplified how diverse problems in biology are converging in the need to generate quantitative data and how the use of quantitative microscopy will clearly be a first line tool for biologists in years to come. On the other, it generated a sense of community amongst the course participants, imparting the philosophy of the course: to foment and energize the next generation of Image Analysts as members of a growing community that not only applies but also generates tools and research that will propel the specialty well into the future.