Frequency-Disentangled Domain Generalization for 2D Prostate Segmentation
IEEE Access · 2026 · 14 · 105043-105055
DOI: 10.1109/access.2026.3694433 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Access, 2026. Verfasst von Yukun Wu und 6 weiteren Autor:innen. Die Arbeit nennt 31 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: National Natural Science Foundation of China, Youth Medical Innovation and Practice Research Program of Guangzhou, Guangdong Provincial Clinical Research Center for Urological Diseases.
Abstract (Originalsprache)
Multi-center prostate imaging suffers from strong domain shifts that often degrade segmentation performance. We propose a frequency-aware domain generalization framework for 2D prostate lesion segmentation. The framework uses wavelet decomposition as a task-motivated prior to perturb appearance-sensitive low-frequency components while preserving boundary-sensitive high-frequency responses. It combines input-level low-frequency randomization with learnable high-frequency preservation (MWDis), a spatial-frequency integrated encoder with wavelet self-attention (SFIE), and a multi-scale spatial-frequency prototype alignment objective (MSFFA). The framework is end-to-end and does not require any target-domain adaptation, aiming to improve robustness and boundary fidelity across heterogeneous multi-center test distributions. We instantiate the backbone with Segment Anything Model v3 (SAM3) and low-rank adaptation (LoRA) adapters for parameter-efficient fine-tuning on source domains.
Förderung
- National Natural Science Foundation of China
- Youth Medical Innovation and Practice Research Program of Guangzhou
- Guangdong Provincial Clinical Research Center for Urological Diseases
Zitieren
APA
Wu, Y., Chen, C., Guo, Z., Liu, J., Jiang, S., Chen, J., & Huang, B. (2026). Frequency-Disentangled Domain Generalization for 2D Prostate Segmentation. IEEE Access, 14, 105043-105055. https://doi.org/10.1109/access.2026.3694433
Vancouver
Wu Y, Chen C, Guo Z, Liu J, Jiang S, Chen J, et al. Frequency-Disentangled Domain Generalization for 2D Prostate Segmentation. IEEE Access. 2026;14:105043-105055. doi:10.1109/access.2026.3694433
BibTeX
@article{Wu2026,
author = {Wu, Yukun and Chen, Chuning and Guo, Ziyi and Liu, Jinwen and Jiang, Shuangjian and Chen, Junxing and Huang, Bin},
title = {Frequency-Disentangled Domain Generalization for 2D Prostate Segmentation},
journal = {IEEE Access},
year = {2026},
volume = {14},
pages = {105043--105055},
doi = {10.1109/access.2026.3694433}
}
EUVIMED