Applications of adaptive analysis in magnetic resonance imaging

The application of a variety of parametric modeling techniques to short complex nuclear magnetic resonance (NMR) data sequences is demonstrated. These techniques have the potential of identifying frequency clusters of signals without being compromised by truncation artifacts. These adaptive algorithms are as rapid as the fast Fourier transform (FFT), and are often a practical alternative to the FFT for generating magnetic resonance images from time-domain data sequences with only 16 complex points. There are two distinct methods of processing nonstationary NMR data, i.e. block and recursive processing. Least-mean-square and modified-least-mean-square algorithms are examples of recursive adaptive procedures, while the Yule-Walker and Burg methods are examples of block processing. Application of the adaptive algorithms yields results where the inherent information content of short time-domain data records are accurately represented. The resolution of these representations is comparable to a DFT analysis with twice the number of samples. This increase in resolution and the accuracy of the signal can be obtained without any increase in acquisition or processing time. Hence, the techniques is well suited for clinical applications on NMR instruments