Abstract
Dynamic imaging is used in magnetic resonance imaging examinations when several shots of the same anatomical cross section need be acquired at a much high temporal resolution than what normal scan procedures would allow. Present techniques improve the temporal resolution by reducing the number of samples required to reconstruct the dynamic images. This comes at the expense of a reduction in either the spatial resolution or the signal-to-noise ratio. In this work, a new technique is presented for improving both the temporal and spatial resolutions without sacrificing the signal-to-noise ratio. The technique is based on modeling individual lines of the reconstructed images with a series of super-resolution rectangular pulses. The dynamic changes of the object can be determined by estimating the parameters of these pulses at the different shots. In particular, a robust estimation is achieved by using linear associative memories, which have been classically used in pattern recognition applications. The technique starts by acquiring a full resolution image for the imaged slice, from which accurate initial values of the parameters are computed. These values are used to train the linear associative memory system to establish the mapping between a few acquired samples of a certain dynamic image and its model parameters. Results on simulated images demonstrated the value of the technique to accurately reconstruct high spatial resolution dynamic images at a significantly reduced scanning time.