研究目的
To develop a multiple limited-angles (MLA) sampling scheme and a multi-direction total variation minimization (MDTVM) method for accurate CT reconstruction from incomplete projections, reducing X-ray radiation dose or scanning time while suppressing shading artifacts.
研究成果
The MLA sampling scheme effectively balances technical implementation complexity and CT reconstruction difficulty by reducing the need for frequent switching of tube power or collimators and lowering data correlation compared to limited-angle sampling. The MDTVM method, designed specifically for MLA CT, suppresses shading artifacts better than TVM by enhancing sparsity in multiple directions. Experiments on digital phantoms and real data demonstrate that MDTVM achieves higher image quality with clearer edges and reduced artifacts, though further improvements are needed for complete artifact elimination and parameter optimization. Future work should focus on incorporating additional prior knowledge, extending to 3D cases, and developing adaptive parameter selection methods.
研究不足
The MLA sampling requires a wide scanning angular range, which may not be feasible in environments with spatial constraints. The reconstruction parameters (e.g., σ, NMDTV) were chosen manually based on visual inspection, lacking an adaptive method. Shading artifacts are weakened but not completely eliminated due to incomplete projections. The method's performance is limited without additional prior knowledge, and deformations in images may still occur. Computation cost increases with the number of directions (NMLA), though parallel computing could mitigate this.
1：Experimental Design and Method Selection:
The study designed the MLA sampling scheme and the MDTVM reconstruction method. It involved simulations and real experiments using iterative reconstruction methods (SART, TVM, MDTVM) with comparisons based on performance metrics like RMSE, PSNR, and SSIM.
2：Sample Selection and Data Sources:
Digital phantoms (FORBILD head phantom and NCAT phantom) and real CT projections from a gear specimen were used. Projections were acquired with specific angular ranges (e.g., 0°–30°, 120°–150°, 240°–270° for NMLA=3).
3：3).
List of Experimental Equipment and Materials:
3. List of Experimental Equipment and Materials: A personal computer (8.0 GB memory, 2.8 GHz CPU, NVIDIA Quadro K620 card), MATLAB 2014b, Microsoft Visual C++ 2010, and a circular fan-beam CT system with a detector array (553 units), X-ray source, and turntable.
4：0 GB memory, 8 GHz CPU, NVIDIA Quadro K620 card), MATLAB 2014b, Microsoft Visual C++ 2010, and a circular fan-beam CT system with a detector array (553 units), X-ray source, and turntable.
Experimental Procedures and Operational Workflow:
4. Experimental Procedures and Operational Workflow: Projections were collected or simulated with noise (Gaussian noise added in simulations). Reconstruction was performed using SART, TVM, and MDTVM algorithms with iterative steps. Parameters (e.g., σ, NMDTV) were set empirically. Image quality was evaluated visually and quantitatively.
5：Data Analysis Methods:
Performance metrics (RMSE, PSNR, SSIM) were calculated. Image profiles and close-ups were analyzed to assess artifact suppression and edge clarity.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
