Salient Region Matching for Fully Automated MR-TRUS Registration

TL;DR

This work tackles the challenge of fully automated MR-TRUS registration for prostate cancer guidance by focusing on salient prostate regions. It introduces a three-stage pipeline—prostate ROI segmentation, rigid initialization, and deformable registration—with a dual-stream encoder and cross-modal spatial attention, guided by a Salient Region Matching Loss that combines ROI mutual information and a weighted multi-class Dice term. The approach achieves superior registration performance on the public mu-RegPro dataset, outperforming state-of-the-art label-driven and segmentation-based methods, and demonstrates the importance of rigid initialization, CMSA, and SRML. The framework’s focus on foreground structures and its explicit structure–intensity alignment within the prostate region hold promise for improved targeting in MR-TRUS guided interventions, though validation on larger datasets and potential joint segmentation–registration strategies are discussed as future directions.

Abstract

Prostate cancer is a leading cause of cancer-related mortality in men. The registration of magnetic resonance (MR) and transrectal ultrasound (TRUS) can provide guidance for the targeted biopsy of prostate cancer. In this study, we propose a salient region matching framework for fully automated MR-TRUS registration. The framework consists of prostate segmentation, rigid alignment and deformable registration. Prostate segmentation is performed using two segmentation networks on MR and TRUS respectively, and the predicted salient regions are used for the rigid alignment. The rigidly-aligned MR and TRUS images serve as initialization for the deformable registration. The deformable registration network has a dual-stream encoder with cross-modal spatial attention modules to facilitate multi-modality feature learning, and a salient region matching loss to consider both structure and intensity similarity within the prostate region. Experiments on a public MR-TRUS dataset demonstrate that our method achieves satisfactory registration results, outperforming several cutting-edge methods. The code is publicly available at https://github.com/mock1ngbrd/salient-region-matching.

Figures (4)

• Figure 1: Illustration of our MR-TRUS registration framework.
• Figure 2: The proposed deformable registration network.
• Figure 3: The cross-modal spatial attention (CMSA) module.
• Figure 4: Visualization of the MR-TRUS registration results.