Probabilistic Atlases to Enforce Topological Constraints

Abstract

Probabilistic atlases (PAs) have long been used in standard segmentation approaches and, more recently, in conjunction with Convolutional Neural Networks (CNNs). However, their use has been restricted to relatively standardized structures such as the brain or heart which have limited or predictable range of deformations. Here we propose an encoding-decoding CNN architecture that can exploit rough atlases that encode only the topology of the target structures that can appear in any pose and have arbitrarily complex shapes to improve the segmentation results. It relies on the output of the encoder to compute both the pose parameters used to deform the atlas and the segmentation mask itself, which makes it effective and end-to-end trainable.

Weakly Supervised Volumetric Image Segmentation with Deformed Templates

“…They are typically built by fusing multiple manually annotated images and incorporated in to CNN architectures as auxiliary inputs to provide localization priors that help the network find structures of interest. The PAs can be either very detailed as in [33,34,41,22,6] to model structures that are known in detail or very rough as in [36] to deal with 3D structures whose shape can vary very significantly.…”

TEDS-Net: Enforcing Diffeomorphisms in Spatial Transformers to Guarantee Topology Preservation in Segmentations

“…Semantic segmentation of medical images is regularly performed using deep convolutional neural networks (CNNs), such as the U-Net [15]. One alternative segmentation method uses a combination of CNNs and spatial transformers to learn the spatial warping required to transform a set of prior shapes into the desired class labels [10,18,17,8,19]. Such methods have outperformed conventional encoder-decoder and state-of-the-art architectures, however, they have no topological guarantees.…”