Introduction

Lipid composition in breast cancer, a central marker of disease progression¹, can be non-invasively quantified using 2D MRS method of double quantum filtered correlation spectroscopy (DQF-COSY)². However, the signal retention of only 25%³ and depleted lipids within tumour^4,5, contribute to low signal to noise ratio (SNR). Since 2D MRS demands a long scan time, SNR improvement approaches beyond signal averaging is required for clinically viable applications. We therefore examined current combination algorithms, with a focus on noise decorrelated algorithms, on 2D spectra acquired using DQF-COSY MRS in breast cancer.

Methods

Fifteen healthy female volunteers (mean age, 66 years; age range 58-76 years) participated in the study. Fifteen patients (mean age, 63 years; age range 53-71 years, seven grade II and eight grade III) and a further ten patients (mean age, 52 years; age range 36-63 years, one grade II and nine grade III) with invasive carcinoma were enrolled into the study to acquire lipid composition spectra from peritumoural adipose tissue and tumour respectively. The study was approved by the NHS Research Ethics Committee, and written informed consent was obtained from each participant prior to the study.

Data Acquisition: MRS data were acquired on a 3 T MRI scanner (Achieva TX, Philips Healthcare, Best, Netherlands) using a 16-phased array breast coil. 2D spectra of lipid composition were acquired using single voxel DQF-COSY with TR of 552 ms, initial TE of 25 ms, t₁ increment of 1 ms, 256 increments, 256 sampling points, 2 repeats per increment, spectral bandwidth of 1000 Hz, and encoding gradients of 30/40/100 ms mT/m. Reference spectra without water suppression were acquired using single voxel PRESS sequence with TR/TE of 1250/144 ms, 1024 data points, and spectral bandwidth of 2000 Hz. Data were collected from both breasts in volunteers with a voxel size of 2×2×2 cm³ containing mainly the adipose tissue. In patients, for spectra from tumour, the voxel was positioned snug-fit to the tumour and for spectra from peritumoural adipose tissue, the voxel was positioned at 1 cm from the tumour and was set to 2×2×2 cm³.

Data Processing: All the algorithms were developed in MATLAB (MathWorks, Natick, MA, USA). The raw data were averaged across repeated acquisitions before signal combination using adaptively optimised combination (AOC)⁶, noise decorrelated combination (nd-comb)⁷, whitened singular value decomposition (WSVD)⁸, S/N², S/N, Signal, and Equal Weighting⁹. The averaged signal, organised as a 2D map based on t₁ and t₂ time domain axis for each coil element, was subsequently apodised using squared sine bell along both time domain axes and zero filled to 512 × 512 points. External (denoted by subscript “e”) and internal (denoted by subscript “i”) weighting coefficients were determined from the maximum peak of the reference spectrum and the maximum peak in frequency domain from first increment of DQF-COSY respectively. The weighting coefficient derived for a coil element was applied to the apodised and zero filled signals at the corresponding coil element. The combined 2D time domain signal was the summation across all the coil elements, and the combined 2D spectrum was subsequently derived using 2D Fast Fourier transform. The SNR of diagonal and cross spectral peaks was computed as the peak height divided by the standard deviation of the real part of the noise in a free peak region¹⁰. The SNR of the entire combined 2D spectrum was defined as the SNR of the dominant peak at (1.3,1.3) ppm. The SNR improvement of a spectral peak for an algorithm was computed as the percentage increase with reference to Equal Weighting. The non-uniformity of SNR improvement across the 2D spectrum was subsequently derived as the coefficient of variance (standard deviation divided by the mean) of the SNR improvement across all the spectral peaks. The SNR improvement of the entire spectrum was calculated as the percentage increase with reference to Signal Weighting.

Results

WSVD_i yielded comparable SNR to algorithms with external reference (WSVD_e, AOC_e, nd-comb_e) and higher SNR than AOC_i and nd-comb_i in volunteers, peritumoural adipose tissue and tumours in vivo (Table 1, Figure 1). WSVD_i had low non-uniformity of 5.4% in volunteers, 5.8% in peritumoural tissue, and 20.7% in tumours (Figure 2). WSVD_i improved the SNR by 41.9 ± 31.7% in volunteers, 44.3 ± 38.2 % in peritumoural tissue and 18.4% (9.4-44.5%) in tumours, in comparison to Signal Weighting algorithm (internal reference), enabling the acquisition shorter than 11 min instead of 17 min with the additional acceleration from the removal of reference acquisition.

Discussion

WSVD_i provided maximal SNR in patients with low non-uniformity. Furthermore, WSVD_i utilises the lipid composition spectrum as a reference, eliminating the need of additional acquisition of external reference spectrum, typically 2 min including scanner preparation time and data processing time.

Conclusion

WSVD_i, is the optimal approach for processing lipid composition spectra acquired using 2D MRS on phased array coils from the breast. WSVD_i not only provides maximal SNR improvement without the need of additional reference spectra, but also delivers consistent improvement across lipids with high uniformity. With improved SNR, high sensitivity is maintained at a reduced acquisition time, a step change on the routine clinical assessment of lipid composition.

Acknowledgements

The author would like to thank Dr Matthew Clemence (Philips Healthcare Clinical Science, UK) for clinical scientist support, Ms Bolanle Brikinns, Ms Louisa Pirie, Ms Linda Lett, and Ms Kate Shaw for patient recruitment support, Ms Dawn Younie for logistic support, Mr Roger Bourne and Ms Mairi Fuller for providing access to the patients as well as Mrs Beverly MacLennan, Mrs Nichola Crouch, Mr Mike Hendry, and Ms Laura Reid for radiographer support.