Method For Image Reconstruction Using Low-Dimensional-Structure Self-Learning and Thresholding

A method for reconstructing an image of a subject from undersampled image data that is acquired with an imaging system, such as a magnetic resonance imaging system or computed tomography system, is provided. From the acquired undersampled image data, an image of the subject is reconstructed and used to guide further image reconstruction. For example, a low resolution image is reconstructed from a portion of the undersampled image data, such as from a portion corresponding to the center of k-space when MRI is used. From this image, a number of similarity clusters are produced and processed. The processing may be by hard thresholding, Wiener filtering, principal component pursuit, or other similar techniques. These processed similarity clusters are then used to reconstruct a final, target image of the subject using, for example, a weighted average combination of the similarity clusters.