Model-based PRFS thermometry using fat as the internal reference and the extended Prony algorithm for model fitting.

A model-based proton resonance frequency shift (PRFS) thermometry method was developed to significantly reduce the temperature quantification errors encountered in the conventional phase mapping method and the spatiotemporal limitations of the spectroscopic thermometry method. Spectral data acquired using multi-echo gradient echo (GRE) is fit into a two-component signal model containing temperature information and fat is used as the internal reference. The noniterative extended Prony algorithm is used for the signal fitting and frequency estimate. Monte Carlo simulations demonstrate the advantages of the method for optimal water-fat separation and temperature estimation accuracy. Phantom experiments demonstrate that the model-based method effectively reduces the interscan motion effects and frequency disturbances due to the main field drift. The thermometry result of ex vivo goose liver experiment with high intensity focused ultrasound (HIFU) heating was also presented in the paper to indicate the feasibility of the model-based method in real tissue.