Unisef
In the context of the Unisef project funded by Helmholtz AI, we implemented the prototype of a webservice to allow training and application of deep learning networks for segmentation running on the HPC infrastructure of DESY. The main idea here is to make DL based segmentation accessible and usable by non-experts, as well as complementing the actual segmentation by active learning to minimise the amount of needed training data, and the computation of confidence scores for the obtained segmentations.
This implementation shall also be understood as a blueprint for envisioned future cloud based imaging services provided in the Helmholtz Imaging context. Part of the Unisef project is also the “Instance segmentation of paper fibers” imaged at the Hereon beamlines at Petra III by KTH scientists.
Other Collaborations
Extracting clinically relevant parameters from real-time MRI images of fontan hearts
Obtaining accurate segmentations of the heart in real-time MRI allows a more realistic view on clinically relevant parameters, such as the stroke volume. Cardiac real-time MRI can assess diastolic filling under breath maneuvers or other cardiac load situations which potentially enhances diagnostics other than CINE breath hold cardiac MRI. Real-time MRI allows rapid acquisitions during […]
Paving the way for future mineral processing and recycling technologies through large-scale analysis of particulate samples
Understanding and quantifying the exact composition of mineral samples paves the way towards advanced methodologies that not only increase the effectiveness of ore processing but also enable future recycling technologies. To this end, samples consisting of ground particles embedded into an epoxy matrix are imaged with computed tomography (CT) at micrometer resolution. Due to the […]