Organ-Aware Cross-Modality Registration Enables Attenuation Correction Without Repeated CT in Multi-Scan Total-Body PET/CT Imaging.
Authors
Abstract
Multi-Scan Total-Body PET/CT imaging, including dual-time-point and multi-tracer protocols, provides valuable metabolic information for enhanced disease diagnosis. However, the necessity for repeated CT scans due to patient repositioning for attenuation correction (AC) increases radiation exposure, raising safety concerns. To address this, we propose a novel organ-aware cross-modality registration model (OCRM) that enables accurate AC without repeated CT scans. The OCRM predicts a deformation field between the attenuation map of the first scan and the non-attenuation-corrected PET image from the subsequent scan, generating a pseudo-attenuation map that is subsequently used for AC during PET image reconstruction. Furthermore, to enhance registration accuracy and improve generalizability, we introduced a organ-aware mechanism that emphasizes alignment in both key organ regions and the global body contour. In comparison to existing deep learning-based AC methods, the OCRM can be deployed in a zero-shot manner across diverse multi-scan PET/CT imaging scenarios without retraining. We validated the method across four different multi-scan PET/CT scenarios, including dual-time-point and dual-tracer protocols. The proposed method outperformed state-of-the-art CT-free AC techniques, demonstrating superior performance in terms of RMSE, PSNR, and SSIM across both whole-body and specific anatomical regions. Specifically, in all four scenarios, SSIM values for the corrected PET images consistently exceeded 0.995, and the RMSE of standardized uptake values remained below 0.005. These results highlight the practical applicability and generalization capability of the proposed method, enabling its seamless deployment across diverse multi-scan PET/CT imaging scenarios.