Projective Skip-Connections for Segmentation Along a Subset of Dimensions in Retinal OCT

In medical imaging, there are clinically relevant segmentation tasks where the output mask is a projection to subset of input image dimensions. this work, we propose novel convolutional neural network architecture that can effectively learn produce lower-dimensional than image. The restores encoded representation only in spatial dimensions and keeps unchanged others. newly proposed projective skip-connections allow linking encoder decoder UNet-like structure. We evaluated method on two retinal Optical Coherence Tomography (OCT): geographic atrophy blood vessel segmentation. outperformed current state-of-the-art approaches all OCT datasets used, consisting 3D volumes corresponding 2D en-face masks. fills methodological gap between classification ND