Fig 1.
Mechanism of fibre evanescent wave spectroscopy (FEWS).
A) Schematic propagation of the light wave in the fibre. B) General set up for FEWS with the light direction from Mid Infra Red (MIR) light source towards the detector. (With the agreement of Cui et al [16]).
Fig 2.
Representative spectra of a patient that was alive (blue) and a patient that was deceased (red) at 6 months.
The frequency band assignments and absorption domains are displayed: blue, protein; orange, lipids; green, nucleic acids; and dark brown, carbohydrates. ν and δ symbols describe the type of vibration: ν depicts stretching and δ for bending, as for asymmetric and s for symmetric vibrational modes. A) Mid-infrared absorbance spectrum. Analysis domain (3200–2800; 1800-950cm-1) is displayed as red lines. B) Second derivatives spectra within the analysis domain.
Fig 3.
Illustrative histogram of spectral variables selection by FADA algorithm and the process of variable selection optimization for Mid-infrared predictive model construction.
The red line shows the minimum for variable selection.
Table 1.
Clinical and biological characteristics of patients included.
Fig 4.
Representative spectra of one patient deceased (red) and one alive (blue) at 6 months.
The seven spectral variables selected to build the predictive model are represented by stars. A) Second derivative spectra in the 3200–2800; 1800-950cm-1 domain. B) An expanded view in the 1530-1380cm-1 domain. C) An expanded view in the 1090-950cm-1 domain.
Table 2.
Selected variable by FADA algorithm to discriminate patients alive or dead at 6 months.
Fig 5.
Boxplots of AUROC, cut-off, sensibility and specificity values across the 100 runs of 90%/10% Monte Carlo cross-validation for the MIR FEWS model.
Boxplots of sensibility and specificity were obtained with a 0.30 cut-off value.
Fig 6.
Comparison of receiver-operating characteristic (ROC) curves for the prediction of 6 months mortality.
MIR model (red) (AUROC:0.94, CI95: 0.87–0.99), MELD score (blue) (AUROC:0.77, CI95: 0.66–0.89) and Child-Pugh score (AUROC: 0.76, CI95: 0.66–0.88) (black).
Fig 7.
Schematic illustrations of spectral and biomolecular networks.
A) Shows the network of spectral biomarkers. Rather than individual values at these 7 wavenumbers, the diagnostic prediction relies on the whole network as it is shown from the correlations between markers. Positive correlations are featured with blue edges, negative with red ones. Edge’s thickness reflects the degree of correlation (r value). Every correlation is significant (p<0.05 at least). B) Complete network of disease markers including spectral (blue), clinical and biological (red) markers and scores (green). Correlations are coded as in (A).