Images, whether taken from a satellite, an electron microscope or MRI scanner, contain a wealth of information. Efficiently processing and analyzing all that information with the help of computers is getting harder, in large part because the image data are getting so big.
“A lot of algorithms don’t scale very well to large volumes of data,” says Dr. Kyle Harrington, who will lead the new technology platform “Image Data Analysis” at the Max Delbrück Center for Molecular Medicine in the Helmholtz Association (MDC). Harrington’s group simultaneously serves as the new Helmholtz Imaging Platform (HIP) support unit at the MDC. “You can’t represent the image volumes that are being acquired now because 10 years ago, people didn’t think you could actually have millions or billions of pixels along one axis.”
HIP aims to address this and other challenges around scientific image data, providing tools and support for all researchers in the Helmholtz Association. A key part of Harrington’s mission is to be as forward-thinking as possible while developing algorithms and other universal software that can process any type of image data. “That is one of the key things of the HIP platform,” Harrington says, “we are supposed to be thinking very hard about all our HIP solutions and making sure they will last for a long time.”
Three institutes form the HIP Core, leveraging their strengths to tackle different parts of the image data pipeline. The German Electron Synchrotron (DESY) will innovate new imaging techniques, and the German Cancer Research Center (DKFZ) will focus on data analysis, particularly with machine learning. MDC will focus on the middle section, helping improve ways to process and manage data.
In particular, Harrington and his team will figure out how to better integrate data across multiple scales and data types. For example, combining fluorescence and electron microscopy for biomedical imaging, or different types of environmental data captured from satellites or deep-sea cameras. The goal is to develop tools that any group at the MDC or within the Helmholtz Assoication that is associated to HIP can use, regardless of the type of data or research field. Harrington is confident they can do it.
“We don’t want people to struggle with imaging data in isolation any longer,” explains Professor Thoralf Niendorf, who is one of the three HIP chairs, and heads the Berlin Ultrahigh Field Facility at the MDC. “We want them to get access to tools which are broadly available no matter what the research question is.”
Programmer for life
Harrington is thrilled by the opportunity to work across data types and scales. It’s not often scientists are encouraged to work across such a broad spectrum, so “that’s really the privilege of the position,” he says. It also connects directly to his own research goals. “It’s been a longstanding dream of mine to be able to take data from the molecular to the whole organism scale and synthesize it into complete datasets,” he says.
Growing up in Alexandria, Virginia, just across the river from Washington, D.C., Harrington started coding when he just seven years old. He was immediately hooked and has pursued it ever since. While earning a Ph.D. in computer science from Brandeis University, he assisted Professor Michael Rosbash with image analysis of fruit flies and their circadian rhythms, just a few years before Rosbash shared the Nobel Prize in Physiology or Medicine in 2017 for discovering the molecular mechanisms behind circadian rhythm. From that first exposure to biomedical imaging, Harrington was once again, hooked.
After a postdoc at Harvard Medical School, Harrington became an assistant professor at University of Idaho in 2016. He has spent a lot of time in Germany over the past few years, through collaborations at the Max Planck Institute for Cell Biology and Genetics and Technical University Dresden. He and his family are excited to move to Berlin and be immersed in its vibrant research culture. “I have been describing Germany as the land of research and candy,” he says.
Harrington has a knack for finding creative ways to piece together and process image data. For example, he and his close collaborator, Dr. Ulrik Günther, developed a faster way to “annotate” or label images for supervised machine learning by incorporating eye-tracking into virtual reality headsets. They have also been leading the development of a 3D and virtual reality visualization tool for complex imaging data called Sciview, which they plan to release later this year.
In another recent project, Harrington and collaborators analyzed and visualized the developing zebrafish vascular system, using images of fluorescently-labeled blood cells to infer the location of veins and arteries. They used the data to produce colorful 3D movies of unfurling zebrafish. The images not only help biologists with their research, they also bring science to life for the broader public. Harrington appreciates the dual importance: “If you make good visualizations, then that makes your data and your work much more understandable.”
Text: Laura Petersen