FS-DANet: Dual-Domain Signal Enhancement and Dynamic Spatial Calibration for Gastric Ultrasound Artifact Mitigation
Yuyi Bai1,2, Yanmin Luo3,4, Zhikui Chen5
1College of Computer Science and Technology, Huaqiao University, Jimei, Xiamen, 361021, Fujian, China.
Journal of Imaging Informatics in Medicine
|July 20, 2026
Summary
A new Frequency-Spatial Dual-domain Awareness and Dynamic Adaptation Network (FS-DANet) improves gastric cancer segmentation in ultrasound images. This method enhances accuracy in low-quality signals, outperforming existing techniques for better clinical diagnosis.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Biomedical Engineering
Background:
- Accurate segmentation of gastric cancer in ultrasound images is vital for early diagnosis and treatment.
- Challenges include speckle noise, tissue deformation, and high computational costs of fully supervised models.
- Few-shot reference-guided methods are efficient but struggle with feature alignment in low-quality ultrasound.
Purpose of the Study:
- To develop an efficient and accurate automatic segmentation method for gastric cancer in ultrasound images.
- To address limitations of existing few-shot methods in handling low-quality ultrasound signals and cross-frame feature alignment.
- To improve segmentation accuracy for personalized diagnosis in complex clinical environments.
Main Methods:
- Proposed FS-DANet (Frequency-Spatial Dual-domain Awareness and Dynamic Adaptation Network) utilizing a Dual-Frequency Disentangled Block (DFDB) for noise separation.
- Introduced Mask-Guided Feature Calibration (MGFC) to suppress artifacts and align features using a support mask.
- Employed an Iterative Deformable Adaptation (IDA) decoder for progressive reconstruction of lesion boundaries.
Main Results:
- FS-DANet achieved superior performance over state-of-the-art reference-guided methods on both GCUI and BUSI datasets.
- Obtained Dice coefficients of 75.41% on GCUI and 74.19% on BUSI, surpassing fully supervised baselines in a single-reference setting.
- Demonstrated significant improvement in segmentation accuracy within complex ultrasound environments.
Conclusions:
- FS-DANet effectively overcomes limitations in few-shot gastric cancer segmentation from ultrasound images.
- The proposed network shows strong potential for clinical application in personalized diagnosis and treatment planning.
- This approach offers a robust solution for accurate segmentation even with low-quality ultrasound data.
