Improved Random Sampling Reconstruction for In-Vivo Data Using Discrete Cosine Transform
Mir R, Irarrazaval P, Lillo I, Plett J, Guarini M
Pontificia Universidad Catolica de Chile
When a signal or image has a sparse or compressible representation in some domain, it is possible to obtain few random samples and reconstruct the image by minimizing the L1 norm of the coefficients in the sparse domain, maintaining fidelity of the acquired samples. In phantoms it is possible to get exact reconstruction. For in vivo images, our hypothesis is that a windowed 2D-DCT of the image, similar to JPEG, yields an excellent concentration of information, allowing a good combination of localization versus frequency. We propose an MRI reconstruction based on the DCT, using adaptive and non uniform k-space subsampling.