Abstract:Disentanglement-based methods effectively facilitate domain-invariant feature learning in Unsupervised Domain Adaptation (UDA) by leveraging complementary domain-specific information relating to statistical distribution differences. However, in scenarios involving cross-domain echocardiography, spectral information is important for determining distribution differences. Existing disentanglement-based methods overlook these spectral differences, resulting in insufficient learning of domain-specific features and unreliable domain- invariant feature disentanglement across different medical centers (domains). In addition, aligning the global marginal distribution in cross-domain disentanglement results in spatially misaligned semantics in domain- invariant features at the pixel level, which hinders robust UDA segmentation. We advance disentanglement learning by designing a novel unified bi-encoder bispace-discriminator disentanglement network (UB2D-Net) 2 D-Net) to disentangle reliable domain-invariant features and adapt the pixel-level aligned semantics preserved therein, thereby achieving robust cross-domain echocardiography segmentation. A space distribution constraint loss is designed to incorporate bispace (spatial and spectral) knowledge into the unified bi-encoder disentanglement path to promote the learning of sufficient domain-specific features in disentanglement. Furthermore, we leverage the confidence-aware weight, which is a novel form of the spatial attention mechanism, to transfer the pixel-level aligned semantics preserved in UDA for the spatial consistency-guided segmentation. We validated our UB2D-Net 2 D-Net for cross-domain left ventricular segmentation on two public echocardiography datasets. Extensive comparison and ablation experiments of UDA segmentation in either CAMUS- EchoNet-Dynamic or EchoNet-Dynamic- CAMUS tasks demonstrated that our method outperformed strong state-of-the-art algorithms.

Disentangle Domain Features for Cross-Modality Cardiac Image Segmentation.

Partial Unbalanced Feature Transport for Cross-Modality Cardiac Image Segmentation

A New Bidirectional Unsupervised Domain Adaptation Segmentation Framework

Disentangled Representation for Cross-Domain Medical Image Segmentation

AttENT: Domain-Adaptive Medical Image Segmentation Via Attention-Aware Translation and Adversarial Entropy Minimization

Multi-Source Domain Adaptation for Medical Image Segmentation.

A 3-D Anatomy-Guided Self-Training Segmentation Framework for Unpaired Cross-Modality Medical Image Segmentation

Unsupervised Domain Adaptation for Cross-Modality Cardiac Image Segmentation Based on Contrastive Image Synthesis

10.1109/ICIPMC62364.2024.10586680 Unsupervised Domain Adaptation for Cross-Modality Cardiac Image Segmentation Based on Contrastive Image Synthesis

Unsupervised Domain Adaptation with Variational Approximation for Cardiac Segmentation

A novel 3D unsupervised domain adaptation framework for cross-modality medical image segmentation

A Structure-aware Framework of Unsupervised Cross-Modality Domain Adaptation via Frequency and Spatial Knowledge Distillation

Unified Bi-Encoder Bispace-Discriminator Disentanglement for Cross-Domain Echocardiography Segmentation

Deep Symmetric Adaptation Network for Cross-Modality Medical Image Segmentation

Toward Accurate Cardiac MRI Segmentation With Variational Autoencoder-Based Unsupervised Domain Adaptation

Unsupervised Cross-modality Cardiac Image Segmentation Via Disentangled Representation Learning and Consistency Regularization.

Reducing Domain Gap in Frequency and Spatial domain for Cross-modality Domain Adaptation on Medical Image Segmentation

Towards Accurate Cardiac MRI Segmentation with Variational Autoencoder-Based Unsupervised Domain Adaptation

Unsupervised Domain Adaptation Method for Medical Image Segmentation Using Fourier Feature Decoupling and Multi-scale Feature Fusion

Dual domain distribution disruption with semantics preservation: Unsupervised domain adaptation for medical image segmentation

TSP-UDANet: Two-Stage Progressive Unsupervised Domain Adaptation Network for Automated Cross-Modality Cardiac Segmentation