Bioimage analysis with ilastik
Ullrich Koethe
Friday, 7 February 2014, 12:30pm, IMP Lecture Hall (Note the room change)
Tools for bioimage analysis presently face two major challenges: (i) they should work reliably in the hands of end users, and (ii) should be able to handle massive amounts of data. To solve the first problem, our group’s major open-source project “ilastik” (www.ilastik.org) offers generic image analysis workflows (pixel and object classification, interactive segmentation, and tracking) in up to five dimensions (space, time and spectral) which can be adapted to new experiments using modern machine learning methods. ilastik’s intuitive user interface and immediate feedback on all interactions enable biologists to train these methods themselves, without consulting an image analyis expert. The second problem is addressed by means of an execution graph architecture called “lazyflow” which determines the minimum computation required to fulfill any user request and executes each request strictly on demand and in parallel. The talk will give an introduction to the underlying algorithms and software design as well as a short online demonstration of the software’s capabilities.