Medical Physics Seminar – Monday, October 9, 2017
Data Consistency Driven CT Image Reconstruction Framework
Yinsheng Li (student of Dr. Guang-Hong Chen)
Research Assistant, Department of Medical Physics, UW-School of Medicine & Public Health, Madison, WI - USA
Current image reconstruction methods used in tomographic imaging modalities such as CT were developed under the assumption that a complete and consistent dataset was acquired during data acquisition. In practice, however, the acquired data are often not consistent. As a result, the application of a well-developed image reconstruction algorithm to an inconsistent dataset generates artifacts in the reconstructed images. Conflicts between classical image reconstruction theory and physics involved in data acquisition procedure motivate us to incorporate the data consistency information into image reconstruction. In the proposed data consistency driven image reconstruction framework, a data inconsistency metric was introduced to classify an acquired dataset into different consistency classes. Conventional single class reconstruction strategy was generalized to reconstruct multiple consistency classes jointly in a matrix completion form using the proposed SMART-RECON algorithm. The proposed framework was applied to improve three-dimensional cone-beam CT (CBCT) image quality, generate time-resolved CBCT angiography from a single short-scan data acquisition, and improve the temporal resolution of CBCT so that high-quality CBCT perfusion imaging can be achieved in interventional suite. These novel imaging techniques will enable physicians to improve their toolbox for better clinical care.