Causality-inspired Single-source Domain Generalization for Medical Image Segmentation

Abstract

Deep learning models usually suffer from domain shift issues, where models trained on one source domain do not generalize well to other unseen domains. In this work, we investigate the single-source domain generalization problem: training a deep network that is robust to unseen domains, under the condition that training data is only available from one source domain, which is common in medical imaging applications. We tackle this problem in the context of cross-domain medical image segmentation. Under this scenario, domain shifts are mainly caused by different acquisition processes. We propose a simple causality-inspired data augmentation approach to expose a segmentation model to synthesized domain-shifted training examples. Specifically, 1) to make the deep model robust to discrepancies in image intensities and textures, we employ a family of randomly-weighted shallow networks. They augment training images using diverse appearance transformations. 2) Further we show that spurious correlations among objects in an image are detrimental to domain robustness. These correlations might be taken by the network as domain-specific clues for making predictions, and they may break on unseen domains. We remove these spurious correlations via causal intervention. This is achieved by resampling the appearances of potentially correlated objects independently. The proposed approach is validated on three cross-domain segmentation tasks: cross-modality (CT-MRI) abdominal image segmentation, cross-sequence (bSSFP-LGE) cardiac MRI segmentation, and cross-center prostate MRI segmentation. The proposed approach yields consistent performance gains compared with competitive methods when tested on unseen domains.

arXiv commentThis is an early, non-peer-reviewed version. For the final peer-reviewed full version that has been substantially revised, please find: https://ieeexplore.ieee.org/document/9961940. Please find the code at https://github.com/cheng-01037/Causality-Medical-Image-Domain-Generalization

References in corpus (29)

• Adam: A Method for Stochastic Optimization
• Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks
• mixup: Beyond Empirical Risk Minimization
• Attention U-Net: Learning Where to Look for the Pancreas
• Improved Regularization of Convolutional Neural Networks with Cutout
• Deep Domain Confusion: Maximizing for Domain Invariance
• Shortcut Learning in Deep Neural Networks
• Learning deep representations by mutual information estimation and maximization
• Confounding variables can degrade generalization performance of radiological deep learning models
• The Medical Segmentation Decathlon
• Domain Generalization via Invariant Feature Representation
• AugMix: A Simple Data Processing Method to Improve Robustness and Uncertainty
• Manifold Mixup: Better Representations by Interpolating Hidden States
• Domain Generalization via Model-Agnostic Learning of Semantic Features
• Generalizing to Unseen Domains via Adversarial Data Augmentation
• Invariant Risk Minimization
• nnU-Net: Self-adapting Framework for U-Net-Based Medical Image Segmentation
• Deep Neural Networks with Random Gaussian Weights: A Universal Classification Strategy?
• Machine Learning with Multi-Site Imaging Data: An Empirical Study on the Impact of Scanner Effects
• A causal view of compositional zero-shot recognition
• Causal Intervention for Weakly-Supervised Semantic Segmentation
• A Learning Strategy for Contrast-agnostic MRI Segmentation
• AirLab: Autograd Image Registration Laboratory
• Self-Challenging Improves Cross-Domain Generalization
• Representation Learning via Invariant Causal Mechanisms
• Grid Saliency for Context Explanations of Semantic Segmentation
• Selecting Data Augmentation for Simulating Interventions
• Cooperative Training and Latent Space Data Augmentation for Robust Medical Image Segmentation
• Semi-supervised Meta-learning with Disentanglement for Domain-generalised Medical Image Segmentation