The field of medical imaging is experiencing significant advancements in domain adaptation and test-time adaptation. Researchers are developing innovative methods to address the challenges of domain shift, which occurs when models are trained on one dataset but applied to another. Recent studies have focused on improving the performance of models on unseen data, including the development of test-time adaptation techniques that enable models to adapt to new domains without requiring additional training data. Noteworthy papers in this area include SPADE, which integrates histopathology with spatial transcriptomics to guide image representation learning, and FreeDNA, which proposes a training-free mechanism for endowing diffusion-based dense prediction models with domain adaptation capabilities. Another notable work is DC-TTA, which adapts the Segment Anything Model on a per-sample basis by leveraging user interactions as supervision. These advancements have the potential to improve the accuracy and reliability of medical imaging models in real-world applications.
Advancements in Domain Adaptation and Test-Time Adaptation for Medical Imaging
Sources
SPADE: Spatial Transcriptomics and Pathology Alignment Using a Mixture of Data Experts for an Expressive Latent Space
Weakly-Supervised Domain Adaptation with Proportion-Constrained Pseudo-Labeling
FreeDNA: Endowing Domain Adaptation of Diffusion-Based Dense Prediction with Training-Free Domain Noise Alignment
DC-TTA: Divide-and-Conquer Framework for Test-Time Adaptation of Interactive Segmentation
When Test-Time Adaptation Meets Self-Supervised Models
HASD: Hierarchical Adaption for pathology Slide-level Domain-shift
Single Image Test-Time Adaptation via Multi-View Co-Training
Spatially Gene Expression Prediction using Dual-Scale Contrastive Learning
HistoART: Histopathology Artifact Detection and Reporting Tool
Reducing Variability of Multiple Instance Learning Methods for Digital Pathology
Few-shot Classification as Multi-instance Verification: Effective Backbone-agnostic Transfer across Domains
ADAptation: Reconstruction-based Unsupervised Active Learning for Breast Ultrasound Diagnosis
Is Visual in-Context Learning for Compositional Medical Tasks within Reach?
Learning from Random Subspace Exploration: Generalized Test-Time Augmentation with Self-supervised Distillation
A Multi-Centric Anthropomorphic 3D CT Phantom-Based Benchmark Dataset for Harmonization
Continual Multiple Instance Learning with Enhanced Localization for Histopathological Whole Slide Image Analysis
F^2TTA: Free-Form Test-Time Adaptation on Cross-Domain Medical Image Classification via Image-Level Disentangled Prompt Tuning