Related Experiment Video
Updated: May 10, 2026
Use of MRI-ultrasound Fusion to Achieve Targeted Prostate Biopsy
Published on: April 9, 2019
ProMA-Net: MR-TRUS prostate registration via a dual-stream Swin Transformer-based network with mixed attention
Junxi Kang1, Bowen Zheng2, Yang Guo1
1School of Automation, Guangdong University of Technology, Guangzhou, 510006, China; Guangdong Provincial Key Laboratory of Intelligent Decision and Cooperative Control, Guangzhou, 510006, China; Guangdong-Hong Kong Joint Laboratory for Intelligent Decision and Cooperative Control, Guangzhou, 510006, China.
This study introduces a new dual-stream Swin Transformer network for accurate multimodal registration of magnetic resonance (MR) and transrectal ultrasound (TRUS) prostate images. The method enhances image-guided prostate interventions by improving MR-TRUS data fusion.
Area of Science:
- Medical Imaging
- Computer Vision
- Biomedical Engineering
Background:
- Multimodal registration of magnetic resonance (MR) and transrectal ultrasound (TRUS) prostate images is vital for targeted biopsy and therapy.
- This task is challenging due to complex cross-modality correspondences and feature fusion conflicts.
- Balancing computational efficiency and registration accuracy remains a significant hurdle.
Purpose of the Study:
- To develop a novel deep learning approach for accurate deformable registration between preoperative MR and intraoperative TRUS prostate images.
- To address the challenges of multimodal feature fusion and anatomical alignment in prostate image registration.
Main Methods:
- A dual-stream Swin Transformer-based network processing MR and TRUS volumes separately.
- Window Self-Cross Mixed Attention (WSCMA) for efficient multimodal feature fusion without explicit attention overhead.
- Gating Cross-Feature Fusion (GCFF) module for adaptive feature reweighting and Multi-Scale Patch-Neighborhood Modality-Independent Neighborhood Descriptor (MIND) Contrastive Loss for enhanced alignment.
Main Results:
- The proposed method achieved superior registration accuracy on the μProReg prostate dataset.
- It demonstrated the highest prostate Dice overlap and the lowest target registration error compared to advanced methods.
- The approach effectively fuses MR and ultrasound data for improved anatomical alignment.
Conclusions:
- The novel dual-stream Swin Transformer network significantly advances multimodal prostate image registration.
- The method offers improved accuracy and efficiency, showing potential for enhancing image-guided prostate interventions.
- Publicly available code facilitates further research and application in clinical settings.
