Logo Lanfrica
  • Accueil
  • Atlas
  • Analyses
  • Documentation
  • Sign in

© 2026 Lanfrica. Tous droits réservés. Tous les droits d'auteur des ressources affichées sur le site Web Lanfrica appartiennent aux détenteurs de droits d'auteur d'origine, sauf indication contraire explicite.

Revisiting Invariant Learning for Out-of-Domain Generalization on Multi-Site Mammogram Datasets

Domaine:

healthcare

Type de record:

paperdataset
Créateur:
Vo,ZarLy,Wan
Hôte:avatar
Achieving health equity in Artificial Intelligence (AI) requires diagnostic models that maintain reliability across diverse populations. However, breast cancer screening systems frequently suffer from domain overfitting, degrading significantly when deployed to varying demographics. While Invariant Learning algorithms aim to mitigate this by suppressing site-specific correlations, their efficacy in medical imaging remains underexplored. This study comprehensively evaluates domain generalization techniques for mammography. We constructed a multi-source training environment aggregating datasets from the United States (CBIS-DDSM, EMBED), Portugal (INbreast, BCDR), and Cyprus (BMCD). To assess global generalizability, we evaluated performance on unseen cohorts from Egypt (CDD-CESM) and Sweden (CSAW-CC). We benchmarked Invariant Risk Minimization (IRM) and Variance Risk Extrapolation (VREx) against a rigorously optimized Empirical Risk Minimization (ERM) baseline. Contrary to expectations, standard ERM consistently outperformed specialized invariant mechanisms on out-of-domain testing. While VREx showed potential in stabilizing attention maps, invariant objectives proved unstable and prone to underfitting. We conclude that engineering equitable AI is currently best served by maximizing multi-national data diversity rather than relying on complex algorithmic invariance.

Visit

arxiv.org

Tasks

computer visionimage classification

Tags

Computer Vision and Pattern Recognition

Similaires

Revisiting Invariant am in Early African American Vernacular EnglishWildIng: A Wildlife Image Invariant Representation Model for Geographical Domain ShiftRevisiting the Out of Africa event with a novel Deep Learning approachDeveloping Bilingual English-Setswana Datasets for Space DomainFine-tuning Multilingual Dense Retrievers on Non-English FAQ Pairs for Zero-Shot Low-Resource Domain GeneralizationWhen Unseen Domain Generalization is Unnecessary? Rethinking Data Augmentation

Revisiting Invariant am in Early African American Vernacular English

Scholars of African American Vernacular English (AAVE) have generally assumed that the invariant am

WildIng: A Wildlife Image Invariant Representation Model for Geographical Domain Shift

Wildlife monitoring is crucial for studying biodiversity loss and climate change. Camera trap images

Revisiting the Out of Africa event with a novel Deep Learning approach

Abstract Anatomically modern humans evolved around 300 thousand

Developing Bilingual English-Setswana Datasets for Space Domain

Fine-tuning Multilingual Dense Retrievers on Non-English FAQ Pairs for Zero-Shot Low-Resource Domain Generalization

We present WebFAQ, a large-scale collection of open-domain question answering datasets derived from

When Unseen Domain Generalization is Unnecessary? Rethinking Data Augmentation

Recent advances in deep learning for medical image segmentation demonstrate expert-level accuracy. H