Magnetic resonance spectroscopic imaging in breast cancer detection: possibilities beyond the conventional theoretical framework for data analysis

Abstract

Magnetic Resonance Spectroscopic Imaging (MRSI) is a promising method for breast cancer diagnosis, providing, in addition to the anatomic picture, complementary biochemical and physiologic information in the form of spectra. It should be able to identify key biochemical changes before the tumour becomes detectable by other functional imaging methods that rely upon single markers not entirely sensitive or specific for malignant activity. MRSI is potentially well suited for screening and repeated monitoring since it entails no radiation exposure. There are, however, limitations to current applications of Magnetic Resonance Spectroscopy (MRS) and MRSI. Many of these can be directly related to reliance upon the conventional data analytical method, i.e. the Fast Fourier Transform (FFT), which has low resolution, poor signal/noise (S/N) in clinical signals, supplies only shape spectra and requires fitting, which is non-unique, so that the number of metabolites must be guessed in advance. This can lead to spurious peaks (over-fitting) and true metabolites being undetected (under-fitting). These limitations of the FFT can be circumvented by recent mathematical advances in signal processing via e.g. the Fast Padé Transform (FPT). As a high resolution, non-linear, stable parametric method, the FPT substantially improves S/N, and fulfills stringent requirements for tumour diagnostics: no post-processing fitting, provides precise numerical results for all peak parameters, and specifies the exact number of metabolites (including those that overlap) from the encoded data. We illustrate in a realistic synthesized model problem similar to MRS that the FPT can identify overlapping peaks that are entirely missed by the FFT, and we give an example from in vivo MRS of the superior resolving power of the FPT compared to FFT at short acquisition time. We also perform detailed paired and logistic regression analyses of Nuclear Magnetic Resonance (NMR) data on extracted breast specimens from patients with breast cancer. These reveal that a rich store of spectroscopic information for detecting malignant breast lesions is found in closely overlapping resonances, some of which decay rapidly and therefore can only be detected at short acquisition times. Since the FPT offers greater possibilities to extract this information, the next step would be to apply the FPT to in vivo MRS and MRSI signals from patients with breast cancer in comparison to normal breast tissue.