Christoph Angermann, Markus Haltmeier, Ruth Steiger +2
The paper proposes a projection-based 2.5D U‑net that replaces 3D convolutions with max‑intensity projections and 2D convolutions, enabling faster training and lower memory use whi…