Rotaflip: A New CNN Layer for Regularization and Rotational Invariance in Medical Images
arXiv:2108.02704
Abstract
Regularization in convolutional neural networks (CNNs) is usually addressed with dropout layers. However, dropout is sometimes detrimental in the convolutional part of a CNN as it simply sets to zero a percentage of pixels in the feature maps, adding unrepresentative examples during training. Here, we propose a CNN layer that performs regularization by applying random rotations of reflections to a small percentage of feature maps after every convolutional layer. We prove how this concept is beneficial for images with orientational symmetries, such as in medical images, as it provides a certain degree of rotational invariance. We tested this method in two datasets, a patch-based set of histopathology images (PatchCamelyon) to perform classification using a generic DenseNet, and a set of specular microscopy images of the corneal endothelium to perform segmentation using a tailored U-net, improving the performance in both cases.
arXiv comment13 pages, 6 images. Originally submitted to the European Conference on Computer Vision 2020 (but did not get enough grade to pass the acceptance threshold)
References in corpus (5)
- Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift
- A Survey on Deep Learning in Medical Image Analysis
- Rotation-invariant convolutional neural networks for galaxy morphology prediction
- Group Equivariant Convolutional Networks
- Exploiting Cyclic Symmetry in Convolutional Neural Networks