| Summary: | Magnetic resonance imaging (MRI) is one of the essential elements in medical areas, particularly in diagnostic procedures. However, due to the under-sampling process, the reconstructed MR image leads to incomplete output, including the edge being blurred and the noise remaining, which are considered fundamental problems of MRI analysis and the diagnosis procedure. The total variational (TV) regularization technique is a standard method for MRI reconstruction. This paper proposes a nonconvex regularization MRI model (LogTV) to construct a logarithm penalty function that could effectively prevent the system due to underestimating characteristics. Moreover, this study will offer an improved alternating direction method of multipliers (ADMM) algorithm and the algorithm's convergence in solving the new nonconvex model. Finally, the numerical proposed model is expected to have a better effect on MRI than similar models. That is, the value of the peak signal-to-noise ratio (PSNR) is more significant, the relative error (RE) is minor, and the structural similarity index measurement (SSIM) is closer to 1 under the proposed model. © 2023, International Association of Engineers. All rights reserved.
|
|---|