3225
View Abstract / View Presentation
Deep learning improves retrospective free-breathing 4D-ZTE thoracic imaging: Initial experience
Dorottya Papp1, Jose M. Castillo T.1, Piotr A. Wielopolski1, Pierluigi Ciet1, Gyula Kotek1, Jifke F Veenland1, and Juan Antonio Hernandez-Tamames2
1Radiology and Nuclear Medicine, Erasmus Medical Center, Rotterdam, Netherlands, 2Erasmus Medical Center, Rotterdam, Netherlands
FCNNs have been widely used in radiology, they have not been extended for improving free-breathing lung MRI yet. Our aim is to improve image quality of 4D-ZTE in free-breathing using FCNN. When tested on unseen data the predicted images had improved visual image quality and artifacts were reduced
Figure 2. Example images from our validation set. A is the ground truth (the original breath-hold ZTE image), B is the test image (artefact augmented breath-hold ZTE) and C is the prediction of the network.
Figure 3. Example images from our test set for one volunteer. On the top there is the same slice in 4 different respiratory phases from the original 4D-ZTE acquisition and on the bottom there are the same phases predicted by the network.