Local rotation invariance in 3D CNNs.

Details

Serval ID
serval:BIB_3046C1F2B2F4
Type
Article: article from journal or magazin.
Collection
Publications
Institution
Title
Local rotation invariance in 3D CNNs.
Journal
Medical image analysis
Author(s)
Andrearczyk V., Fageot J., Oreiller V., Montet X., Depeursinge A.
ISSN
1361-8423 (Electronic)
ISSN-L
1361-8415
Publication state
Published
Issued date
10/2020
Peer-reviewed
Oui
Volume
65
Pages
101756
Language
english
Notes
Publication types: Journal Article ; Research Support, Non-U.S. Gov't
Publication Status: ppublish
Abstract
Locally Rotation Invariant (LRI) image analysis was shown to be fundamental in many applications and in particular in medical imaging where local structures of tissues occur at arbitrary rotations. LRI constituted the cornerstone of several breakthroughs in texture analysis, including Local Binary Patterns (LBP), Maximum Response 8 (MR8) and steerable filterbanks. Whereas globally rotation invariant Convolutional Neural Networks (CNN) were recently proposed, LRI was very little investigated in the context of deep learning. LRI designs allow learning filters accounting for all orientations, which enables a drastic reduction of trainable parameters and training data when compared to standard 3D CNNs. In this paper, we propose and compare several methods to obtain LRI CNNs with directional sensitivity. Two methods use orientation channels (responses to rotated kernels), either by explicitly rotating the kernels or using steerable filters. These orientation channels constitute a locally rotation equivariant representation of the data. Local pooling across orientations yields LRI image analysis. Steerable filters are used to achieve a fine and efficient sampling of 3D rotations as well as a reduction of trainable parameters and operations, thanks to a parametric representations involving solid Spherical Harmonics (SH),which are products of SH with associated learned radial profiles. Finally, we investigate a third strategy to obtain LRI based on rotational invariants calculated from responses to a learned set of solid SHs. The proposed methods are evaluated and compared to standard CNNs on 3D datasets including synthetic textured volumes composed of rotated patterns, and pulmonary nodule classification in CT. The results show the importance of LRI image analysis while resulting in a drastic reduction of trainable parameters, outperforming standard 3D CNNs trained with rotational data augmentation.
Keywords
Diagnostic Imaging, Humans, Image Processing, Computer-Assisted, Neural Networks, Computer, 3D Texture, Convolutional neural network, Local rotation invariance, Steerable filters
Pubmed
Web of science
Create date
08/07/2020 11:44
Last modification date
16/12/2021 6:32
Usage data