For proper generalization performance of convolutional neural networks (CNNs) in medical image segmentation,the learnt features should be invariant under particular non-linear shape variations of the input. To induceinvariance in CNNs to such transformations, we propose Probabilistic Augmentation of Data using DiffeomorphicImage Transformation (PADDIT) - a systematic framework for generating realistic transformations that can beused to augment data for training CNNs. The main advantage of PADDIT is the ability to produce transformationsthat capture the morphological variability in the training data. To this end, PADDIT constructs a mean templatewhich represents the main shape tendency of the training data. A Hamiltonian Monte Carlo(HMC) scheme isused to sample transformations which warp the training images to the generated mean template. Augmentedimages are created by warping the training images using the sampled transformations. We show that CNNstrained with PADDIT outperforms CNNs trained without augmentation and with generic augmentation (0.2 and0.15 higher dice accuracy respectively) in segmenting white matter hyperintensities from T1 and FLAIR brainMRI scans.
展开▼
