Characterising the variance and reproducibility of low rank denoising methods for spectroscopic data

Low rank denoising methods can reduce the noise present in spectroscopic data, a typically signal-to-noise limited technique. Low rank denoising is a data driven technique, typically exploiting the linear predictability or spatio-temporal separability of the data. Despite high levels of apparent denoising it is not clear whether denoising decreases the final uncertainty in the measurement, as the denoised data can be biased. In this work we assess the uncertainty reduction using Monte Carlo simulation and by measuring reproducibility of noisy and denoised 1H MRSI. We find low rank denoising reduces uncertainty but by much less than is apparent.