HU-based Foreground Masking for 3D Medical Masked Image Modeling

2025-09-09

AI Summary

Random masking in 3D medical image masked image modeling (MIM) disregards anatomical density variations, limiting representational capacity. To address this, we propose an Hounsfield unit (HU)-guided anatomically aware foreground masking method. By applying HU-based thresholding, our approach precisely isolates organ tissues while excluding diagnostically irrelevant air and fluid background regions, enabling intensity-driven, anatomy-aware mask generation. This is the first work to explicitly incorporate HU priors into the MIM pretraining framework, significantly enhancing discriminative learning of anatomical structures. Evaluated on five major 3D medical segmentation benchmarks—BTCV, FLARE22, MM-WHS, AMOS22, and BraTS—our method achieves Dice scores of 84.64%, 92.43%, 90.67%, 88.64%, and 78.55%, respectively, consistently outperforming random masking baselines. These results advance task-specific, anatomy-informed self-supervised pretraining for medical imaging.

Technology Category

• Computer Vision: Segmentation
• Machine Learning: Hardware-aware ML
• Humans and AI: Human-in-the-loop Machine Learning

Application Category

• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Web Mining and Content Analysis: Large pretrained models with web data
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

While Masked Image Modeling (MIM) has revolutionized fields of computer vision, its adoption in 3D medical image computing has been limited by the use of random masking, which overlooks the density of anatomical objects. To address this limitation, we enhance the pretext task with a simple yet effective masking strategy. Leveraging Hounsfield Unit (HU) measurements, we implement an HU-based Foreground Masking, which focuses on the intensity distribution of visceral organs and excludes non-tissue regions, such as air and fluid, that lack diagnostically meaningful features. Extensive experiments on five public 3D medical imaging datasets demonstrate that our masking consistently improves performance, both in quality of segmentation and Dice score (BTCV:~84.64%, Flare22:~92.43%, MM-WHS:~90.67%, Amos22:~88.64%, BraTS:~78.55%). These results underscore the importance of domain-centric MIM and suggest a promising direction for representation learning in medical image segmentation. Implementation is available at github.com/AISeedHub/SubFore/.

Problem

• Enhancing 3D medical MIM with HU-based masking
• Focusing on diagnostically meaningful anatomical regions
• Improving segmentation quality and Dice scores

Innovation

• HU-based masking focuses on organ intensity distribution
• Excludes non-tissue regions lacking diagnostic features
• Improves segmentation quality and Dice scores consistently