Manual annotation for three-dimensional (3D) medical image segmentation is time-consuming and depends on expert knowledge. Semi-supervised learning (SSL) helps reduce labeling work, but most frameworks assume the data come from the same source.

However, images from devices of different hospitals often vary greatly, making SSL, unsupervised domain adaptation (UDA), and semi-supervised domain generalization (Semi-DG) hard to apply. Models may also over-rely on simple frequency features, which can hurt generalization and reinforce pseudo-label errors.

In a study published in Pattern Recognition, a team led by GUI Huaqiao from the Hefei Institutes of Physical Science of the Chinese Academy of Sciences, along with the collaborators, developed a universal semi-supervised artificial intelligence (AI) learning framework designed to reduce the burden of manual annotation in 3D medical image segmentation while improving model generalization across multiple clinical centers.

By combining adversarial training with two new data augmentation modules: the low-frequency adversarial adaptive enhancement (L-AAE) module and the frequency adaptive suppression and enhancement (F-ASE) module, the framework suppressed frequency shortcuts.

L-AAE reduced the model's reliance on dominant low-frequency features and narrowed distribution differences between images from different centers through adversarial adjustment and style optimization. F-ASE adjusted feature weights across frequency bands, helping the model learn richer frequency information and avoid biased features. The original images and optimized adversarial samples were then used together for model training in the SSL framework.

Experiments on several public datasets showed that the framework performed well across SSL, UDA, and Semi-DG, with improved accuracy and better suitability for real-world clinical applications. L-AAE and F-ASE can be easily integrated into many common neural network models, making the method flexible and practical for different systems.

This work offers a solution for improving AI reliability in medical image analysis, especially in real-world clinical settings with data variations across hospitals.