HEURISTIC OPTIMIZATION ALGORITHMS APPLIED TO THE QUANTIFICATION OF SPECTROSCOPIC DATA

The quantification of in vivo MR spectra imposes severe problems becau se of low spectral resolution and poor signal-to-noise ratio. Maximum likelihood methods are often applied. However, with conventional spect rum analysis procedures, the search for a global minimum in a multidim ensional space often terminates in only a local minimum. Heuristic opt imization procedures are able to circumvent this difficulty. Two appro aches, the genetic algorithm and the simulated annealing, have been ad apted to the quantification of MR spectra, For evaluation purposes, th e procedures have been applied to synthetic and in vivo spectra with d ifferent noise levels. They both allowed a reliable spectrum quantific ation. The areas of most peaks were quantified reproducibly, although in some cases, the discrimination between spectroscopically almost ide ntical metabolites (e.g., glutamate and glutamine) was not completely satisfactory. The two algorithms are found to be valuable alternative methods in the quantification of in vivo MR spectra.