Characterizing the Signal-to-noise Ratio and Spatio-temporal Resolution of an Imaging Transformer Model for CMR

Deep learning imaging model can improve quality, reduce acquisition time, and increase resolution. A less-studied prerequisite to deploy these models is characterizing the output quality. Previous studies have shown nonlinear imaging methods can achieve high SNR, but at the cost of losing substantial spatial or temporal resolution [1]. In this study, we developed methods to characterize SNR and spatio-temporal resolution and used these methods to evaluate an imaging transformer model for CMR.