Raman data preprocessing tutorial

The preprocessing performed on Raman spectral data depends on the particular application, and should always be validated to ensure there are no significant artifacts induced. However, for Raman spectroscopy there are some common data preprocessing steps that can be applied in the large majority of cases. These includes:
  1. Subtraction of system/sample background spectrum (laser on with no sample present)
  2. Intensity calibration of Raman spectrum using a NIST standard (e.g., SRM2241 or SRM2242)
  3. Cropping of Raman spectrum to desired spectral ranges e.g., 400-1800 cm-1 and/or 1800-2700 cm-1 or 2700-3600 cm-1
  4. Baseline (auto-fluorescence) removal .
  5. Smoothing of the Raman spectrum
  6. Normalization of Raman spectrum (Application specific).
References:
  1. Byrne, H. J., Knief, P., Keating, M. E., Bonnier, F., Chemical Society Reviews, 45, 7, 2016.
  2. Savitzky, A., Golay, M. J. E., Analytical Chemistry, 36, 8, 1964

1. Subtraction of system/background spectrum

Subtraction of a background spectrum is important for most applications. Background subtraction can be used to account for system contributions (e.g. background signal generated by the optical components in use) and/or to account for Raman spectroscopic signal arising from the substrate in use. To perform background subtraction, a background spectrum (either of air or the substrate being used) is acquired using the same parameters as are to be used for imaging of the sample under investigation. The background spectrum is simply subtracted from the spectra of the sample. Notice, for some samples (e.g., scattering or absorbing samples) the background spectrum might not contribute linearly and hence more advanced background subtraction is required.

Glass/water background spectrum

Overlay

Cell + glass/water background spectrum

Cell  - glass/water (subtracted) spectrum

2. Intensity calibration of Raman spectrum

Different microscopes, spectrometers and CCD cameras exhibit different transmission and sensitivity to light which is highly wavelength dependent. For this reason, to enable comparison of relative intensities across laboratories, intensity calibration should generally be performed using a standard. To perform this, the spectrum of a known (fluorescent) standard can be measured (e.g., SRM2241 for 785 nm or SRM2242 for 532 nm). The transfer function of a Raman microscope is then calculated by dividing the measured standard with the true spectrum of the standard (provided by NIST). Multiplying this transfer function with new measured Raman spectra enables the experimenter to largely correct for instrument variability.

Before intensity calibration

After intensity calibration

3. Crop Raman spectrum

Spectral cropping is an important step in the analysis of Raman spectra as unwanted spectra features such as the laser contribution, peaks arising from the substrate background, silent region or other undesired spectral regions can confound the analyses. In addition, when working with large Raman spectral images or datasets, spectral cropping can help to reduce processing times and thus aide analysis. In many biomedical cases most biological information can be found in the 600-1800 cm-1 and 2700-3600 cm ranges.

Original

Cropping: Substrate Background

Cropping: removing low frequecy laser Contribution

Cropping: Spectral Regions

4. Subtraction of autofluorescence

Autofluorescence subtraction in Raman spectroscopy is used when samples emit unwanted autofluroescence. There exists many variants of autofluoresecence subtraction and the specific choice is application dependent. NOTICE autofluoresence subtractions can lead to unwanted spectral artifacts overwhelming the Raman signals through overfit or a poor fit. Hence, extra care should be taken in chosing a robust baseline subtraction.

Raw data

Over-Subtraction

Good Subtraction

Under-Subtraction

4. Smoothing a Raman spectrum

Spectral smoothing can be employed since Raman signals can be weak and noisy.  The most commonly applied method is the Savitzky-Golay filter [2], a technique that uses a linear least squares method to fit low-degree polynomials over data windows. However, it should be noted that parameters such as the window size and the order of Savitzky-Golay filters can have a significant impact on Raman spectra and as such Savitzky-Golay and other methods of spectral smoothing should be used with caution as important molecular information can be removed.

Un-smoothed Raman spectrum

Over-Smoothed

Good smoothing

Under-Smoothed

5. Normalization of Raman spectrum

To account for variations in laser intensity, detector sensitivity, tissue contour and enable quantitative cross-comparison of Raman spectra acquired at different times, it is common to normalise Raman spectra. Typically this is done by normalising to a constant peak, the area under the curve. This is highly application dependent and

Original

Normalised