The potential for practical improvements in cancer diagnostics by mathematically-optimized magnetic resonance spectroscopy

Abstract

The fast Pade transform (FPT) has been benchmarked as a stable, high-resolution processor. In this paper, the performance of the FPT is examined for in vitro magnetic resonance (MR) spectroscopic data associated with ovarian, breast and prostate cancer as well as benign or normal tissue. We also examine how the FPT handles in vivo MR spectroscopic (MRS) time signals from human brain encoded by high field and clinical (1.5 T) scanners. Salient comparisons are made with the conventional data analysis through the fast Fourier transform (FFT). Separation of noise from genuine signal is carried out with a view to practical applications. Compared to the FFT, the fast Pade transform provided markedly improved resolution of total shape spectra from encoded in vivo time signals from healthy human brain and for in vitro data associated with ovarian cancer. Evidence is presented as to why it is necessary to go beyond MR total shape spectra to calculate metabolite concentrations. It is shown that error spectra, while necessary, are insufficient for accurate assessment of MR data. Two examples from oncology are given to illustrate this point: (1) a marker of breast cancer, phosphocholine, is detected on the component shape spectra, but not on the total shape spectrum, (2) diagnostically important multiplet resonances in prostate cancer spectra can only be detected on the component shape spectra, but not on the total shape spectrum. The FPT provides accurate calculation of metabolite concentrations based on in vitro MR data from three diagnostic problems in clinical oncology: (1) malignant and benign ovarian lesions, (2) breast cancer, fibroadenoma and normal breast tissue and (3) prostate cancer tissue, healthy glandular and stromal prostate tissue. Practical implementation of signal-noise separation is demonstrated for MR time signals encoded in vivo from the human brain on a clinical (1.5 T) scanner. Some 23 stable resonances are thereby identified and quantified. These results provide the basis for the needed next steps: to extensively apply the FPT to in vivo time signals encoded mainly on clinical scanners from e.g. brain tumors, breast, ovary and prostate cancers as well as from benign and normal tissue. The overall goal is that this practical approach through mathematical optimization enables Pade-based MRS to soon be implemented in clinical oncology, including target planning, post-radiotherapeutic follow-up and other aspects of radiation therapy.