Self-Supervised Vessel Enhancement Using Flow-Based Consistencies.

Vessel segmentation is an essential task in many clinical applications. Although supervised methods have achieved state-of-art performance, acquiring expert annotation is laborious and mostly limited for two-dimensional datasets with a small sample size. On the contrary, unsupervised methods rely on handcrafted features to detect tube-like structures such as vessels. However, those methods require complex pipelines involving several hyper-parameters and design choices rendering the procedure sensitive, dataset-specific, and not generalizable. We propose a self-supervised method with a limited number of hyper-parameters that is generalizable across modalities. Our method uses tube-like structure properties, such as connectivity, profile consistency, and bifurcation, to introduce inductive bias into a learning algorithm. To model those properties, we generate a vector field that we refer to as a flow. Our experiments on various public datasets in 2D and 3D show that our method performs better than unsupervised methods while learning useful transferable features from unlabeled data. Unlike generic self-supervised methods, the learned features learn vessel-relevant features that are transferable for supervised approaches, which is essential when the number of annotated data is limited.

Estimating uncertainty in MRF-based image segmentation: A perfect-MCMC approach.

Typical methods for image segmentation, or labeling, formulate and solve an optimization problem to produce a single optimal solution. For applications in clinical decision support relying on automated medical image segmentation, it is also desirable for methods to inform about (i) the uncertainty in label assignments or object boundaries or (ii) alternate close-to-optimal solutions. However, typical methods fail to do so. To estimate uncertainty, while some Bayesian methods rely on simplified prior models and approximate variational inference schemes, others rely on sampling segmentations from the associated posterior model using (i) traditional Markov chain Monte Carlo (MCMC) methods based on Gibbs sampling or (ii) approximate perturbation models. However, in such typical approaches, in practice, the resulting inference or generated sample set are approximations that deviate significantly from those indicated by the true posterior. To estimate uncertainty, we propose the modern paradigm of perfect MCMC sampling to sample multi-label segmentations from generic Bayesian Markov random field (MRF) models, in finite time for exact inference. Furthermore, for exact sampling in generic Bayesian MRFs, we extend the theory underlying Fill's algorithm to generic MRF models by proposing a novel bounding-chain algorithm. On several classic problems in medical image analysis, and several modeling and inference schemes, results on simulated data and clinical brain magnetic resonance images show that our uncertainty estimates gain accuracy over several state-of-the-art inference methods.