Advances in Domain Generalization for Medical Imaging

The field of medical imaging is witnessing significant developments in domain generalization, with a focus on improving model performance across diverse datasets and scanners. Recent research has highlighted the importance of addressing scanner-induced variability, data distribution shifts, and domain-specific variations in imaging conditions. Innovative approaches, such as probabilistic modeling and feature augmentation, are being explored to enhance the generalization of segmentation models. Notably, the introduction of new datasets and frameworks, such as SCORPION and SimCons, is setting a new standard for reliability testing and evaluation of model consistency across diverse scanners. Noteworthy papers include:

  • CLEAR, which proposes a framework for separating task-relevant content features from task-irrelevant style features, leading to better performance when superficial characteristics shift at test time.
  • SCORPION, which introduces a new dataset and framework for evaluating model reliability under scanner variability, and proposes SimCons, a flexible framework for addressing scanner generalization.

Sources

CLEAR: Unlearning Spurious Style-Content Associations with Contrastive LEarning with Anti-contrastive Regularization

Is Exchangeability better than I.I.D to handle Data Distribution Shifts while Pooling Data for Data-scarce Medical image segmentation?

SCORPION: Addressing Scanner-Induced Variability in Histopathology

Exploring Probabilistic Modeling Beyond Domain Generalization for Semantic Segmentation

Learning Semantic Directions for Feature Augmentation in Domain-Generalized Medical Segmentation

Built with on top of