Unsupervised denoising of Raman spectra with cycle-consistent generative adversarial networks

Ciaran Bench,Mads S. Bergholt,Mohamed Ali al-Badri

2023-08-01

Abstract:Raman spectroscopy can provide insight into the molecular composition of cells and tissue. Consequently, it can be used as a powerful diagnostic tool, e.g. to help identify changes in molecular contents with the onset of disease. But robust information about sample composition may only be recovered with long acquisition times that produce spectra with a high signal to noise ratio. This acts as a bottleneck on experimental workflows, driving a desire for effective spectral denoising techniques. Denoising algorithms based on deep neural networks have been shown superior to `classical' approaches, but require the use of bespoke paired datasets (i.e. spectra acquired from the same set of samples acquired with both long and short acquisition times) that require significant effort to generate. Here, we propose an unsupervised denoising approach that does not require paired data. We cast the problem of spectral denoising as a style transfer task and show how cycle-consistent generative adversarial networks can provide significant performance benefits over classical denoising techniques.

Medical Physics