Meeting Banner
Abstract #0973

Deep Learning Based Needle Localization on Real-Time MR Images of Patients Acquired During MR-guided Percutaneous Interventions

Jonathan Weine1,2, Elodie Breton3, Julien Garnon3,4, Afshin Gangi3,4, and Florian Maier1

1Siemens Healthcare, Erlangen, Germany, 2TU Dortmund, Dortmund, Germany, 3ICube UMR7357, University of Strasbourg, CNRS, FMTS, Strasbourg, France, 4Imagerie Interventionnelle, Hôpitaux Universitaires de Strasbourg, Strasbourg, France

Automatic localization of needles in real-time images can facilitate MR-guided percutaneous interventions. It enables automatic slice repositioning and targeting support and, thus, allows for faster workflows. Systematic acquisition of training data for deep learning tasks in the context of interventional MRI can be difficult due to the fact, that treatment quality must not be impaired. Therefore, we investigated whether images of porcine animal experiments can be used to train deep learning algorithms for needle artifact segmentation in human datasets. Results show that transfer is feasible at 1.5T. Additional fine tuning using small amounts of human data further reduces the error.

This abstract and the presentation materials are available to members only; a login is required.

Join Here