All researchProject overview

Spectroscopic sensing · 2026

Cross-device Raman spectroscopy

Aligning spectra from portable Raman instruments to a laboratory reference for bacterial analyte classification.

Qatar University · Qatar National Research Fund-supported bacterial diagnostics research

Graphical abstract showing spectra from portable and reference Raman instruments, device transformation, and bacterial classification.
Graphical abstract by Mahmud et al., Spectrochimica Acta Part A (2026). Reproduced under CC BY 4.0.View full-size figureSource articleCC BY 4.0

Approach

Can spectra recorded on different instruments support more consistent recognition of the same analyte classes?

SERS-D2DNet transforms spectra from four portable devices toward a Renishaw reference. The study then uses SuperRaman to classify the original and transformed spectra across 20 analyte classes, using 50 spectra per class to train the classifiers.

  1. Record spectra on five instruments
  2. Transform portable spectra with SERS-D2DNet
  3. Classify with SuperRaman

Reported results

Classification accuracy increased after transformation for all four portable devices, with the largest gains for Enwave and Rapid ID. The transformed spectra also matched the reference more closely than the outputs of five benchmark sequence-to-sequence models, with R² between 98.80% and 99.57%.

Enwave accuracy
97.58% → 99.98%
Rapid ID accuracy
96.54% → 98.87%
Bar chart of R² between transformed and reference spectra for four portable devices. For every device, SERS-D2DNet has the highest bar, between 98.80% and 99.57%, above FPN, U-Net, U-Net++, UNet3+ and MultiResUNet. The untransformed spectra score between 63.97% and 87.99%.
Figure 6 from Mahmud et al. (2026). R² between transformed and reference spectra for SERS-D2DNet and five benchmark sequence-to-sequence models. The black bars compare the untransformed portable spectra with the reference.View full-size figureSource articleCC BY 4.0
Four panels compare the original and transformed mean spectra with the Renishaw reference. Shaded bands show one standard deviation.
Figure 4a from Mahmud et al. (2026), cropped to the four device-level comparisons. After transformation, the mean spectra follow the Renishaw reference more closely in these comparisons. Green shows the original spectra, red the transformed spectra, and black the reference. Shading shows one standard deviation. Raman shift in cm⁻¹. CC BY 4.0.View full-size figurePublished figure and license

My contribution

I led the modelling and evaluation of the spectral transformation and recognition workflow, including implementation, generalization analysis, visualization, and writing.

Study scope

The study used prepared analytes across five instruments, including the laboratory reference. The reported comparisons apply to those samples and acquisition conditions. The Zenodo record explains how to request the restricted files.