Analysis of phosphorus magnetic resonance spectra using hidden markov models

A 31P magnetic resonance spectrum is modeled as the sum of three components: a collection of peaks, a baseline, and noise. All three components, as well as prior information, are modeled probabilistically. In this Bayesian context, a maximum a posteriori simultaneous estimate of the baseline and peaks is sought. A nearly global search for this pair is performed by expressing the spectrum as the output of a hidden Markov model and using dynamic programming. Experiments are demonstrated with both synthetic and actual data.