Unbiased measurements of reconstruction fidelity of sparsely sampled magnetic resonance spectra.

Qinglin Wu,Brian E Coggins,Pei Zhou

doi:10.1038/ncomms12281

Qinglin Wu, Brian E Coggins + Show 1 more

Open Access

https://doi.org/10.1038/ncomms12281

Copy DOI

Abstract

The application of sparse-sampling techniques to NMR data acquisition would benefit from reliable quality measurements for reconstructed spectra. We introduce a pair of noise-normalized measurements, and , for differentiating inadequate modelling from overfitting. While and can be used jointly for methods that do not enforce exact agreement between the back-calculated time domain and the original sparse data, the cross-validation measure is applicable to all reconstruction algorithms. We show that the fidelity of reconstruction is sensitive to changes in and that model overfitting results in elevated and reduced spectral quality.

Highlights

The application of sparse-sampling techniques to nuclear magnetic resonance (NMR) data acquisition would benefit from reliable quality measurements for reconstructed spectra
The quality of spectral reconstruction is frequently assessed by direct comparison with artifact-free spectra generated from fully sampled datasets, by examination of algorithm-specific parameters, or by estimating the reduction of aliasing artifacts in the reconstructed spectra
We introduce two algorithm-independent measurements for evaluating the quality of nuclear magnetic resonance (NMR) spectra reconstructed from sparsely sampled datasets, demonstrate their utility in differentiating inadequate modelling from overfitting, and discuss the implication of such quality measurements for the fidelity of NMR spectral reconstruction

Summary

Introduction

The application of sparse-sampling techniques to NMR data acquisition would benefit from reliable quality measurements for reconstructed spectra. The quality of a reconstructed spectrum can be measured by computing the inverse Fourier transform of the spectrum and comparing the resulting time domain data with the raw measurements at the sampled positions.

Results

Conclusion