Reconstruction of radial simultaneous multi-slice myocardial perfusion data with self-guided deep image prior

Motivation: Current state-of-the-art (SoTA) radial simultaneous multi-slice (SMS) reconstruction methods heavily depend on temporal constraints and features, potentially creating artifacts and blurring in time frames with large inter-frame motion. Goal(s): Our goal was to examine the synergistic ability of implicit and explicit priors to reconstruct accurate SMS perfusion images without temporal constraints and get comparable image quality to SoTA. Approach: A self-supervised network was tested on two patients with different acquisition protocols. Time curves were examined for temporal consistency. Results: The method was able to generate consistent and comparable results to the SoTA and over the entire dynamic sequence. Impact: The promising results will enable consistent image quality across all time frames, even in the presence of large respiratory and cardiac motion within the dynamic sequence. This will allow for improved visual diagnosis as well as myocardial blood flow quantification.