Deformation-Aware MR-TRUS Image Translation for Prostate Cancer Brachytherapy

Transrectal ultrasound (TRUS) is widely used in prostate cancer brachytherapy for real-time imaging and to guide radioactive seed implants. However, its low soft tissue contrast limits precise anatomical localization. In contrast, magnetic resonance imaging (MRI) offers superior soft-tissue contrast but exhibits different structural deformations and nonlinear intensity responses compared to TRUS, preventing its direct use in brachytherapy procedures. To achieve accurate seed localization and soft tissue visualization, it often relies on manual annotations on both images, which are labor-intensive and prone to errors. We propose a region-of-interest (ROI)-guided modality translation framework that synthesizes TRUS images from MRI using structural priors and intensity-aware normalization. By generating synthetic TRUS images that are modality-aligned with real TRUS images, our approach simplifies subsequent deformable registration. The method combines cross-domain image synthesis with anatomical constraints to ensure fidelity in both geometric and intensity representations. Evaluations on multiinstitutional datasets demonstrate significant improvements in synthetic image quality and MR-TRUS alignment. This work advances multimodal medical image translation and supports robust cross-modality signal registration, thereby facilitating image-guided therapies and communication-driven clinical systems.