Fast discrimination of tumor and blood cells by label-free surface-enhanced Raman scattering spectra and deep learning

Abstract

Rapidly and accurately identifying tumor cells and blood cells is an important part of circulating tumor cell detection. Raman spectroscopy is a molecular vibrational spectroscopy technique that can provide fingerprint information about molecular vibrational and rotational energy levels. Deep learning is an advanced machine learning method that can be used to classify various data accurately. In this paper, the surface-enhanced Raman scattering spectra of blood cells and various tumor cells are measured with the silver film substrate. It is found that there are significant differences in nucleic acid-related characteristic peaks between most tumor cells and blood cells. These spectra are classified by the feature peak ratio method, principal component analysis combined with K-nearest neighbor, and residual network, which is a kind of deep learning algorithm. The results show that the ratio method and principal component analysis combined with the K-nearest neighbor method could only distinguish some tumor cells from blood cells. The residual network can quickly identify various tumor cells and blood cells with an accuracy of 100%, and there is no complex preprocessing for the surface-enhanced Raman scattering spectra. This study shows that the silver film surface-enhanced Raman scattering technology combined with deep learning algorithms can quickly and accurately identify blood cells and tumor cells, indicating an important reference value for the label-free detecting circulating tumor cells.

Surface-enhanced Raman spectroscopy for bioanalysis and diagnosis

“…236 Most recently, blood cells and circulating tumor cells (CTCs) could be rapidly discriminated by employing label-free SERS and deep learning algorithms. 237 The SERS spectra obtained from eight kinds of tumor cell line were classified by using the characteristic peak ratio method, and the PCA-K-nearest neighbour (KNN) algorithm was applied for tumor cell and blood cell discrimination. Furthermore, the SERS spectra were used for training and testing a ResNet model constructed by the authors.…”

Machine Learning of Raman Spectroscopy Data for Classifying Cancers: A Review of the Recent Literature

“…Wu et al also performed wavenumber shifting, up to 4 cm , as well as adding linear combinations of 2–5 random spectra from the same class to create a new spectrum, thus increasing the sample from 233 to 2420 spectra. Fang et al linearly combined several spectra to create a new spectrum and also performed ‘wavenumber shifting’ and added ‘random noise’, creating 6600 spectra from 510 spectra [ 51 ]. Xia et al augmented the training set up to an unspecified number, shifting the wavenumber axis and adding noise to the magnitude at each wavenumber, a process which more closely resembles the Poisson noise typical of Raman spectra, compared to adding Gaussian noise [ 36 ].…”

“…The test data are a representation of the general population of interest and adding noise could be considered dangerous as it may then become less representative. Four of the reviewed studies which performed data augmentation did so on the entire dataset before splitting into training and test sets [ 26 , 27 , 35 , 51 ]. There is a technique called test-time augmentation which is becoming more common, particularly with small datasets.…”

A Framework for Biosensors Assisted by Multiphoton Effects and Machine Learning

“…A molecule scatters irradiant light from a source laser in the Raman method, which is a light scattering technique [ 232 ]. Most of the scattered light is of the same wavelength as the laser source and hence useless, but a tiny quantity of light is dispersed at various wavelengths and so is beneficial [ 233 ].…”