Vascular Morphology Motivated Progressive Structure-Enhanced Restoration for Sparse-view 3D-DSA

Three-dimensional digital subtraction angiography (3D-DSA) is a key technique for the diagnosis and treatment of cerebrovascular diseases. The conventional dense-view acquisition is time-consuming and delivers a non-negligible radiation dose to both patients and doctors, making sparse-view acquisition a highly promising solution. However, sparse sampling induces severe streak artifacts in 3D-DSA images, thereby affecting subsequent treatment procedures. To address these issues, this study presents a Progressive Structure-Enhanced Restoration (PSER) framework for sparse-view 3D-DSA. First, a 3D DNN that can inherently exploit inter-slice and intra-slice correlations is designed for coarse artifact removal and image restoration. The Discrete Wavelet Transform and Weighted Convolution are integrated to encourage the 3D DNN to focus on restoring high frequency vascular structures in both the frequency and spatial domains, respectively. Second, a 2.5D DNN is employed to further restore fine-grained vascular details while avoiding discontinuities in 3D-DSA images. Since brain vessels exhibit snake-like morphological characteristics, Dynamic Snake Convolution is integrated to adaptively model the complex vascular structures through topology-aware deformation, thereby improving vascular continuity. Qualitative and quantitative results on both simulated and real datasets demonstrate the potential of the proposed PSER method in artifact removal, structure restoration, and continuity preservation.