STIS: A novel approach for standardization of spectrometric instruments in fungal Raman spectroscopy

More than 1.5 million annual deaths from invasive fungal infections underscore the demand for rapid and accurate diagnostics. Although single-cell Raman spectroscopy is a sensitive and label-free technique, its clinical translation is hindered by spectral inconsistencies between instruments. This work introduces a spectral transformer for instrument standardization (STIS), which replaces the localized linear approximations of traditional DS and PDS with a global self-attention mechanism. By encoding spectral position via embeddings and modeling full-range dependencies, STIS captures complex instrumental deviations without overfitting. Evaluated on Raman spectra from 47 clinical fungal strains, STIS markedly surpasses existing methods, raising the average correlation to 0.9897 and reducing the transfer error to 5.13%. Notably, consistent cross-device alignment is attained using only 10 standardization samples. With STIS calibration, slave instrument spectra achieve recognition accuracies of 95.74% (SVM) and 97.87% (CNN) on a held-out test set-An improvement exceeding 30% over uncalibrated data and clearly outperforming DS and PDS.