Towards Data-Efficient Medical Imaging: A Generative and Semi-Supervised Framework

2025-10-07

AI Summary

To address the critical bottlenecks of scarce annotated data and severe class imbalance in medical imaging, this paper proposes SSGNet: a unified framework integrating class-specific generative modeling with iterative semi-supervised pseudo-labeling. We innovatively adapt StyleGAN3 for fine-grained, high-fidelity, class-conditional medical image synthesis. Coupled with a confidence-threshold-driven multi-round pseudo-label refinement mechanism, SSGNet dynamically enhances label quality for unlabeled samples. Evaluated across multiple public medical imaging benchmarks—spanning both classification and segmentation tasks—SSGNet achieves significant performance gains: FID improves by 12.6%, and mean Dice score increases by 4.3%. To our knowledge, this is the first work to establish a synergistic, closed-loop optimization between generative modeling and semi-supervised learning in medical image analysis, delivering strong generalizability and markedly reduced annotation dependency.

Technology Category

• Machine Learning: Semi-Supervised Learning
• Computer Vision: Generative Adversarial Networks (GANs) for Vision
• Natural Language Processing: Code Generation / Program Synthesis from Natural Language

Application Category

• Social Networks and Social Media: Generative AI / large language models and their impact on social systems
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Deep learning in medical imaging is often limited by scarce and imbalanced annotated data. We present SSGNet, a unified framework that combines class specific generative modeling with iterative semisupervised pseudo labeling to enhance both classification and segmentation. Rather than functioning as a standalone model, SSGNet augments existing baselines by expanding training data with StyleGAN3 generated images and refining labels through iterative pseudo labeling. Experiments across multiple medical imaging benchmarks demonstrate consistent gains in classification and segmentation performance, while Frechet Inception Distance analysis confirms the high quality of generated samples. These results highlight SSGNet as a practical strategy to mitigate annotation bottlenecks and improve robustness in medical image analysis.

Problem

• Addresses scarce annotated medical imaging data
• Combines generative modeling with semi-supervised learning
• Enhances classification and segmentation performance

Innovation

• Combines generative modeling with pseudo labeling
• Augments training data using StyleGAN3 generated images
• Refines labels through iterative pseudo labeling process