Will talk about: Electron microscopy circuit reconstruction
Fred Hamprecht is interested in image processing and machine learning, in particular active learning. With his group, he develops automated diagnostic systems with applications both in industrial quality control and the life sciences. Major applications include connectomics, tracking of all cells in a developing embryo and quantitative analysis of high-throughput experiments. The group puts particular emphasis on the the user-friendly training of such systems, and is actively developing open source libraries and program suites such as http://ilastik.org
Fred studied and earned his PhD at the Swiss Federal Institute of Technology (ETH). After a brief period at the Seminar for Statistics, he became a Professor for Multidimensional Image Processing at the University of Heidelberg in 2001. He is a co-founder of the Heidelberg Collaboratory for Image Processing (HCI), and was a fellow of the Marsilius Kolleg in 2010/11.
The tenet of connectomics is that knowledge of the wiring diagram of a brain will facilitate, or even be prerequisite for, an understanding of its function. Serial sectioning electron microscopy now yields brilliant volume images that allow human tracers to accurately follow all neural processes, as well as identify synapses, in tiny parts of a brain. The race is on for microscopic volume imaging of an entire mammalian brain, and the target is likely to be reached in the foreseeable future.
However, human tracing will not scale to such exceedingly large volumes. Besides the actual image acquisition, automated circuit reconstruction is thus becoming the major bottleneck. Current algorithms do not yield flawless segmentations yet, and buying a larger computer does not solve the issue. Instead, better models are required.
I will survey the state of the art in automated reconstruction and explain why algorithm developers are optimistic to still do their bit in good time.