DCE image registration via unfolded RPCA and temporal intensity modeling with vision transformers

Accurate registration of Dynamic Contrast-Enhanced (DCE) images is critical for preserving the temporal fidelity of contrast kinetics, which underpins quantitative pharmacokinetic analysis and reliable lesion characterization. However, conventional registration methods often fail to account for the dual challenges of patient-induced motion and dynamic intensity variations caused by contrast agent uptake and washout. To address these limitations, we propose a novel Coarse-to-Fine Vision Transformer framework (NCF-ViT) that explicitly incorporates physiological priors into a ViT-based registration framework. Specifically, the proposed method incorporates a deep unfolded Robust Principal Component Analysis (RPCA) network to disentangle motion-related and enhancement-related components, thereby isolating physiologically meaningful signal dynamics. Furthermore, a time-series regularization constraint based on Time-Intensity Curve (TIC) smoothness is introduced to enforce temporal coherence, while a hierarchical ViT architecture enables coarse-to-fine spatial alignment. Comprehensive evaluations on four datasets, including brain DCE-MRI, brain DCE-CT, liver DCE-MRI, and multi-modal brain imaging, demonstrate that the proposed framework achieves improved performance in both temporal fidelity and spatial registration accuracy compared with state-of-the-art methods. Notably, the improved alignment leads to more accurate downstream lesion delineation and pharmacokinetic parameter estimation. These findings highlight the potential of NCF-ViT as a robust and clinically applicable solution for high-fidelity DCE image registration and quantitative perfusion analysis.