Abstract #0099

Low-dimensional-Structure Self-Learning & Thresholding (LOST): Regularization Beyond Compressed Sensing for MRI Reconstruction

Mehmet Akcakaya¹, Tamer Basha¹, Beth Goddu¹, Lois Goepfert¹, Kraig V. Kissinger¹, Vahid Tarokh², Warren J. Manning¹, Reza Nezafat¹

¹Medicine, Beth Israel Deaconess Medical Center, Harvard Medical School, Boston, MA, USA; ²School of Engineering & Applied Sciences, Harvard University, Cambridge, MA, USA

We develop an improved image reconstruction technique for undersampled acquisitions that learns and utilizes the structure of images being reconstructed. The results of our retrospective study with coronary MRI imply that the proposed method achieves higher acceleration rates compared to conventional CS reconstructions. The pilot prospective acquisitions confirm this finding, and additionally show that our method provides superior image quality at higher rates compared to traditional parallel imaging reconstruction.