Application of Regularization for Improving Reconstruction of Dynamic Contrast-Enhanced MRI of Spinal Cord Injury
C Hui1*, J Herrera2, P Narayana2, (1) UT MD Anderson Cancer Center, Houston, TX, (2) UT Medical School, Houston, TXTU-G-134-7 Tuesday 4:30PM - 6:00PM Room: 134
Purpose: DCE-MRI allows quantitative assessment of vascular permeability and is the preferred method for investigating blood-spinal cord barrier (BSCB) permeability in spinal cord injury (SCI). However, substantial variability in the permeability parameters has been observed in DCE-MRI analysis. We hypothesized that reconstructing DCE-MRI data with regularization will reduce the variability.
Methods: Previous studies have shown that the use of iterative reconstruction with regularization can reduce noise in MRI data. In this study, total variation and wavelet sparsity were used as regularization functions. Reconstruction was done by minimizing the regularization functions using Bregman iteration. Simulations were performed to see if noise reduction resulting from regularization can improve accuracy of permeability parameters. Simulated DCE-MRI data were generated with 5% noise. The corresponding dynamic images were reconstructed using inverse Fourier transform (IFT) and also with regularization. The accuracy of the permeability parameters computed from their respective reconstructions was compared. Finally, in-vivo DCE-MRI experiments were performed in 12 animals to determine the extent of BSCB compromise after experimental SCI. Permeability parameters were calculated from the images reconstructed with the two different methods. The inter-subject variability in permeability parameters was compared between the two reconstruction methods.
Results: From the simulations, the relative root mean square (RMS) error in K^trans was 36% using IFT. Reconstruction with regularization reduced the error of the reconstructed images which resulted in a reduction of RMS error in K^trans to 23%. In our in-vivo SCI study, the inter-subject standard deviation of average K^trans was 39% using IFT. Regularization can produce noticeably smoother images and K^trans maps. As a result, reconstruction with regularization reduced the inter-subject standard deviation of K^trans to 34%.
Conclusion: Iterative reconstruction with regularization can improve quality of the reconstructed images in DCE-MRI. The resulting images yield more accurate permeability parameters and can reduce the inter-subject variability in DCE-MRI.