0710

View Abstract / View Presentation

Comparison of Pure Deep Learning Approaches for Placental Extraction from Dynamic Functional MRI sequences between 19 and 37 Gestational Weeks.

Bryan Quah¹, Anna Dong¹, Neil Rao¹, Patrick Hoang¹, Michael Hirano¹, Manjiri K. Dighe¹, and Colin Studholme¹
¹University of Washington, Seattle, WA, United States

We investigate 4 data schemes: full 3D images, full 2D slices, 3D patches and 2D patches on our dataset of 68 3D R2* images and train a fully automated U-Net model for the task of placenta tissue segmentation to find out the effectiveness of such approaches.

28 week old subject with labeled placenta tissue maps. (Top row: reference label map, Middle row: map generated by model trained on full 2D scheme, Bottom row: map generated by model trained on full 3D scheme.)

19 week old subject with labeled placenta tissue maps. (Top row: reference label map, Middle row: map generated by model trained on full 2D scheme, Bottom row: map generated by model trained on full 3D scheme.)