SRA-Seg: Synthetic to Real Alignment for Semi-Supervised Medical Image Segmentation

TL;DR

SRA-Seg is a semi-supervised medical image segmentation framework that aligns synthetic and real data in feature space, improving performance by bridging the domain gap with novel alignment and augmentation techniques.

Contribution

The paper introduces SRA-Seg, a novel framework that explicitly aligns synthetic and real medical images in semantic feature space for improved segmentation.

Findings

Achieves 89.34% Dice on ACDC with 10% labeled data

Outperforms existing semi-supervised methods

Matches performance of methods using real unlabeled data

Abstract

Synthetic data, an appealing alternative to extensive expert-annotated data for medical image segmentation, consistently fails to improve segmentation performance despite its visual realism. The reason being that synthetic and real medical images exist in different semantic feature spaces, creating a domain gap that current semi-supervised learning methods cannot bridge. We propose SRA-Seg, a framework explicitly designed to align synthetic and real feature distributions for medical image segmentation. SRA-Seg introduces a similarity-alignment (SA) loss using frozen DINOv2 embeddings to pull synthetic representations toward their nearest real counterparts in semantic space. We employ soft edge blending to create smooth anatomical transitions and continuous labels, eliminating the hard boundaries from traditional copy-paste augmentation. The framework generates pseudo-labels for synthetic images via an EMA teacher model and applies soft-segmentation losses that respect uncertainty in mixed regions. Our experiments demonstrate strong results: using only 10% labeled real data and 90% synthetic unlabeled data, SRA-Seg achieves 89.34% Dice on ACDC and 84.42% on FIVES, significantly outperforming existing semi-supervised methods and matching the performance of methods using real unlabeled data.