TY - GEN
T1 - Roto-translation covariant convolutional networks for medical image analysis
AU - Bekkers, Erik J.
AU - Lafarge, Maxime W.
AU - Veta, Mitko
AU - Eppenhof, Koen A.J.
AU - Pluim, Josien P.W.
AU - Duits, Remco
PY - 2018/1/1
Y1 - 2018/1/1
N2 - We propose a framework for rotation and translation covariant deep learning using SE(2) group convolutions. The group product of the special Euclidean motion group SE(2) describes how a concatenation of two roto-translations results in a net roto-translation. We encode this geometric structure into convolutional neural networks (CNNs) via SE(2) group convolutional layers, which fit into the standard 2D CNN framework, and which allow to generically deal with rotated input samples without the need for data augmentation. We introduce three layers: a lifting layer which lifts a 2D (vector valued) image to an SE(2)-image, i.e., 3D (vector valued) data whose domain is SE(2); a group convolution layer from and to an SE(2)-image; and a projection layer from an SE(2)-image to a 2D image. The lifting and group convolution layers are SE(2) covariant (the output roto-translates with the input). The final projection layer, a maximum intensity projection over rotations, makes the full CNN rotation invariant. We show with three different problems in histopathology, retinal imaging, and electron microscopy that with the proposed group CNNs, state-of-the-art performance can be achieved, without the need for data augmentation by rotation and with increased performance compared to standard CNNs that do rely on augmentation.
AB - We propose a framework for rotation and translation covariant deep learning using SE(2) group convolutions. The group product of the special Euclidean motion group SE(2) describes how a concatenation of two roto-translations results in a net roto-translation. We encode this geometric structure into convolutional neural networks (CNNs) via SE(2) group convolutional layers, which fit into the standard 2D CNN framework, and which allow to generically deal with rotated input samples without the need for data augmentation. We introduce three layers: a lifting layer which lifts a 2D (vector valued) image to an SE(2)-image, i.e., 3D (vector valued) data whose domain is SE(2); a group convolution layer from and to an SE(2)-image; and a projection layer from an SE(2)-image to a 2D image. The lifting and group convolution layers are SE(2) covariant (the output roto-translates with the input). The final projection layer, a maximum intensity projection over rotations, makes the full CNN rotation invariant. We show with three different problems in histopathology, retinal imaging, and electron microscopy that with the proposed group CNNs, state-of-the-art performance can be achieved, without the need for data augmentation by rotation and with increased performance compared to standard CNNs that do rely on augmentation.
KW - Cell boundary segmentation
KW - Group convolutional network
KW - Mitosis detection
KW - Roto-translation group
KW - Vessel segmentation
UR - http://www.scopus.com/inward/record.url?scp=85054049941&partnerID=8YFLogxK
U2 - 10.1007/978-3-030-00928-1_50
DO - 10.1007/978-3-030-00928-1_50
M3 - Conference contribution
AN - SCOPUS:85054049941
SN - 978-3-030-00927-4
T3 - Lecture Notes in Computer Science
SP - 440
EP - 448
BT - Medical Image Computing and Computer Assisted Intervention – MICCAI 2018 - 21st International Conference, 2018, Proceedings
A2 - Schnabel, Julia A.
A2 - Davatzikos, Christos
A2 - Alberola-López, Carlos
A2 - Fichtinger, Gabor
A2 - Frangi, Alejandro F.
PB - Springer
CY - Cham
T2 - 21st International Conference on Medical Image Computing and Computer Assisted Intervention, MICCAI 2018
Y2 - 16 September 2018 through 20 September 2018
ER -