MRI lung lobe segmentation of pediatric cystic fibrosis patients using a neural network trained with publicly accessible CT datasets
Orso Pusterla1,2,3, Rahel Heule4,5, Francesco Santini1,3,6, Thomas Weikert6, Corin Willers2, Simon Andermatt3, Robin Sandkühler3, Sylvia Nyilas7, Philipp Latzin2, Oliver Bieri1,3, and Grzegorz Bauman1,3
1Division of Radiological Physics, Department of Radiology, University Hospital Basel, University of Basel, Basel, Switzerland, 2Division of Pediatric Respiratory Medicine and Allergology, Department of Pediatrics, Inselspital, Bern University Hospital, University of Bern, Bern, Switzerland, 3Department of Biomedical Engineering, University of Basel, Basel, Switzerland, 4High Field Magnetic Resonance, Max Planck Institute for Biological Cybernetics, Tübinge, Germany, 5Department of Biomedical Magnetic Resonance, University of Tübingen, Tübingen, Germany, 6Department of Radiology, University Hospital Basel, University of Basel, Basel, Switzerland, 7Department of Diagnostic, Interventional and Pediatric Radiology, Inselspital, Bern University Hospital, University of Bern, Bern, Switzerland
Pulmonary biomarkers quantifications on a lobar level provide improved specificity against whole-lung analyses. However, lobar quantifications of pulmonary MR data are hardly accessible due to the complex work required for the manual segmentations. Supervised neural networks have shown the premise for automatic segmentation, but it is challenging to gather labelled data for the training. To overcome these limitations, in this work, we “translate” publicly accessible chest CT datasets and lobe segmentations to pseudo-MR data, and we then train a network able to segment consistently lung lobes of acquired MRI data. The cross-modality approach has excellent prospects to automatize MRI analyses.
This abstract and the presentation materials are available to members only;
a login is required.