Related Experiment Videos
DIFNet: A Dual-Branch Interactive Fusion Network for Femoral Nerve Segmentation in Ultrasound Images
Junbo Gao1, Yujie He2, Wei Sun2
1College of Information Engineering, Shanghai Maritime University, Shanghai, 201306, China. jbgao@shmtu.edu.cn.
Journal of Imaging Informatics in Medicine
|July 20, 2026
Summary
A new AI model, DIFNet, accurately segments the femoral nerve (FN) in ultrasound images. This improves nerve blocks by providing more complete and precise nerve visualizations.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Anatomy
Background:
- Accurate femoral nerve (FN) segmentation in ultrasound is vital for nerve blocks.
- Challenges include FN tortuosity, size variations, and low-contrast boundaries.
- Current methods struggle with fragmented results and false positives due to poor dependency modeling.
Purpose of the Study:
- To develop an advanced deep learning model for precise FN segmentation in ultrasound.
- To overcome limitations of existing methods in capturing FN morphology and context.
Main Methods:
- Proposed Dual-branch Interactive Fusion Network (DIFNet) with parallel CNN and Transformer encoders.
- Utilized Cross-Branch Interaction Module (CBIM) for continuity, Multi-Scale Dilated Fusion (MSDF) for variations, and Region-Guided Enhancement (RGEM) for boundary refinement.
Main Results:
- DIFNet achieved superior performance on public and private datasets (e.g., mDice 91.69%, mIoU 84.97% on private data).
- Demonstrated more continuous, complete, and accurate FN segmentations compared to state-of-the-art methods.
Conclusions:
- DIFNet effectively addresses FN segmentation challenges in ultrasound imaging.
- The model's robustness and accuracy enhance its utility for clinical applications like nerve blocks.