Probabilistic Atlases to Enforce Topological Constraints

TL;DR

This paper introduces a CNN-based method that leverages probabilistic atlases encoding topology to improve segmentation of complex, variable structures in medical images, enabling more flexible and accurate results.

Contribution

It presents an end-to-end trainable CNN architecture that uses rough topological atlases to enhance segmentation of structures with complex shapes and poses.

Findings

Improved segmentation accuracy on complex structures.

Effective use of topology-encoded atlases in CNNs.

End-to-end trainable architecture.

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.