CRISP - Reliable Uncertainty Estimation for Medical Image Segmentation

TL;DR

CRISP introduces a contrastive learning approach to produce anatomically consistent uncertainty maps in medical image segmentation, outperforming existing pixel-wise methods by leveraging a joint latent space.

Contribution

This paper presents CRISP, a novel contrastive method that encodes valid segmentations and images into a joint latent space for improved uncertainty estimation in medical imaging.

Findings

CRISP outperforms state-of-the-art uncertainty estimation methods.

It provides anatomically consistent uncertainty maps.

Validated on four diverse medical imaging datasets.

Abstract

Accurate uncertainty estimation is a critical need for the medical imaging community. A variety of methods have been proposed, all direct extensions of classification uncertainty estimations techniques. The independent pixel-wise uncertainty estimates, often based on the probabilistic interpretation of neural networks, do not take into account anatomical prior knowledge and consequently provide sub-optimal results to many segmentation tasks. For this reason, we propose CRISP a ContRastive Image Segmentation for uncertainty Prediction method. At its core, CRISP implements a contrastive method to learn a joint latent space which encodes a distribution of valid segmentations and their corresponding images. We use this joint latent space to compare predictions to thousands of latent vectors and provide anatomically consistent uncertainty maps. Comprehensive studies performed on four medical image databases involving different modalities and organs underlines the superiority of our method compared to state-of-the-art approaches.