Accurate quantitation of Magnetic Resonance Spectroscopy (MRS) signals is an essential step before converting the estimated signal parameters, such as frequencies, damping factors, and amplitudes, into biochemical quantities (concentration, pH). Several subspace-based parameter estimators have been developed for this task, which are efficient and accurate time-domain algorithms. However, they suffer from a serious drawback: they allow only a limited inclusion of prior knowledge which is important for accuracy and resolution. In this paper, a new method is presented: KNOB-SVD and its improved variant KNOB-TLS. KNOB-SVD is a recently proposed method, based on the Singular Value Decomposition (SVD), which allows the use of more prior knowledge about the signal parameters than previously published subspace-based methods. We compare its performance in terms of robustness and accuracy with the performance of three commonly used methods for signal parameter estimation: HTLS, a subspace-based method which does not allow any inclusion of prior knowledge, except for the model order; HTLSPK(Dfdeq), a subspace-based method obtained by incorporating in HTLS the prior information that the frequency differences between doublet components are known and the damping factors are equal; and AMARES, an interactive maximum likelihood method that allows the inclusion of a variety of prior knowledge. Extensive simulation and in vivo studies, using 31P as well as proton MRS signals, show that the new method outperforms HTLS and HTLSPK(Dfdeq) in robustness, accuracy, and resolution, and that it provides parameter estimates comparable to the AMARES ones.
Subspace-based MRS data quantitation of multiplets using prior knowledge
Laudadio T;
2004
Abstract
Accurate quantitation of Magnetic Resonance Spectroscopy (MRS) signals is an essential step before converting the estimated signal parameters, such as frequencies, damping factors, and amplitudes, into biochemical quantities (concentration, pH). Several subspace-based parameter estimators have been developed for this task, which are efficient and accurate time-domain algorithms. However, they suffer from a serious drawback: they allow only a limited inclusion of prior knowledge which is important for accuracy and resolution. In this paper, a new method is presented: KNOB-SVD and its improved variant KNOB-TLS. KNOB-SVD is a recently proposed method, based on the Singular Value Decomposition (SVD), which allows the use of more prior knowledge about the signal parameters than previously published subspace-based methods. We compare its performance in terms of robustness and accuracy with the performance of three commonly used methods for signal parameter estimation: HTLS, a subspace-based method which does not allow any inclusion of prior knowledge, except for the model order; HTLSPK(Dfdeq), a subspace-based method obtained by incorporating in HTLS the prior information that the frequency differences between doublet components are known and the damping factors are equal; and AMARES, an interactive maximum likelihood method that allows the inclusion of a variety of prior knowledge. Extensive simulation and in vivo studies, using 31P as well as proton MRS signals, show that the new method outperforms HTLS and HTLSPK(Dfdeq) in robustness, accuracy, and resolution, and that it provides parameter estimates comparable to the AMARES ones.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
