How the fast Padé transform handles noise for MRS data from the ovary: importance for ovarian cancer diagnostics

Abstract

Advanced signal processing through the fast Pade transform (FPT) can enhance resolution and generates quantitative metabolic information for magnetic resonance spectroscopy (MRS). Herein, we apply both \(\hbox {FPT}^{(\pm )}\) variants to in vitro MRS data as encoded from benign and malignant ovarian cyst fluid and perform detailed analysis with several noise levels. In the presence of higher background noise, all genuine metabolites were unambiguously identified and their concentrations precisely computed, using small fractions of the total signal length by both FPT variants. In the \(\hbox {FPT}^{(-)}\), signal–noise separation was accomplished with the help of the “Stability test”, whereby the non-physical information is binned and denoised spectra are generated. In the \(\hbox {FPT}^{(+)}\) even more stringent signal–noise separation was achieved: the spurious resonances reside exclusively in the negative imaginary frequency domain, whereas the genuine content is all in the positive imaginary frequency region. Pole-zero coincidence of spurious resonance remained complete in the \(\hbox {FPT}^{(+)}\) even at higher noise levels. Via the \(\hbox {FPT}^{(+)}\), a denoised spectrum is generated automatically, without the need for the “Stability test”. The two variants \(\hbox {FPT}^{(\pm )}\) provide self-contained cross-validation of the reconstructed spectral parameters, from which the metabolite concentrations of benign and malignant ovarian cyst fluid are reliably computed. These results are particularly promising for more effective ovarian cancer diagnostics, overcoming the major obstacles that have hindered MRS from becoming the method of choice for non-invasive assessment of ovarian lesions.