Accelerating the Raman spectroscopic discrimination of normal and cancerous tissues with low-rank constraint

Abstract

Raman spectroscopic detection has been extensively validated as a potentially powerful analytical tool for high-efficiency cancer diagnosis. However, Raman scattering is inherently weak, and consequently Raman signal differences between normal and cancerous tissues are easily submerged in the random detecting noise. In order to improve the signal-to-noise ratio, Raman spectral acquisition is generally performed with a relatively long integration time, which strongly hinders the rapid discrimination of normal and cancerous tissues and probably photodamages the tissues under the long-time laser irradiation. In this work, spectral denoising based on low-rank constraint has been investigated and compared with the classical Savitzky–Golay smoothing and wavelet-transform denoising, demonstrating better performance in terms of spectral quality and tissue discrimination as well as simplicity of parameter selection. Specifically, Raman spectra of normal and osteosarcoma tissues were captured with the micro-Raman spectral system, and then classified using the unsupervised K-means clustering. As a result, the spectral classification following the low-rank denoising can remain accurate, even when shortening the integration time by more than 2 orders of magnitude, i.e. from 120 s to 1 s. With the help of low-rank SNR enhancement, Raman spectroscopic detection can be significantly accelerated, presumably promoting the application of Raman imaging to characterize the tissue heterogeneity for more accurate cancer diagnosis.