13125 Berlin, Germany
From individual molecules and cells, to whole organisms and ecologies, image data is acquired across all scales. Images are distinctly useful because they capture spatial relationships, between, for example, two particles, or many cells in an organ. That information, especially when combined with other streams of data like gene expression, can tell researchers about what is happening, when and where in the body in much greater detail than before. At the Image Data Analysis platform, we create the digital tools that help researchers access this deeper insight.
The Image Data Analysis platform at the MDC develops computational tools for processing, visualizing and analyzing diverse sources of image data. Image data may be 2D, 3D, or more generically N-dimensional, where additional dimensions may include time, channels, genes, or more. We use modern computing techniques, such as machine learning and virtual reality, to help researchers engage with their data in new ways.
Many teams are taking sequences of 3D images to capture dynamics of biological processes on the microscopic level. Often this can generate volumes of data that quickly exceed the capacity of standard computer memory. We are developing solutions to enable researchers to view and analyze their extremely large datasets.
For example, we are incorporating machine learning to help detect cells in static images, track cells over time, infer properties, and make predictions. This allows us to transform large datasets into compact and efficient representations, such as cell trajectories, which can be easily analyzed.
As part of the Helmholtz Imaging Platform, all of the tools we develop will benefit researchers across the entire Helmholtz Association. The goal is to develop a universal framework that enables image analysis algorithms and tools across a broad range of spatial scales of image data. A key challenge will be to ensure the framework is adaptable and can keep pace with the rapid advancements in imaging and computing power.
Interactive visualizations can help scientists explore relationships hidden in their large image datasets. We have developed an open-source software program called Sciview, which allows researchers to view and handle their data in a virtual reality environment. Researchers can become immersed in their image data, even labeling, or “annotating” their data using eye movements.
Sciview also offers many other advanced features, including the ability to handle very large volumes of data, and using advanced computer graphics that can render complex visual scenes. Sciview can integrate with ImageJ, a popular imaging processing program used by computational and experimental scientists. We are working to make Sciview available for a wide range of applications and anticipate it will help scientists in a range of fields, from molecular biology to environmental engineering.