Fig. 5.

(a) Examples of HSI images of two patients, shown with respective patient ID, from the HELICoiD dataset.³⁹ Each pixel in the shown 2D image possesses a spectral signature with 826 bands. From this signature, we predict the molecular concentration change for (b) hemodynamic $\delta c_{\mathrm{HbT}}=\delta c_{{\mathrm{HbO}}_2}+\delta c_{\mathrm{HHb}}$ and (c) metabolic $\delta c_{\mathrm{diffCCO}}=\delta c_{\mathrm{oxCCO}}-\delta c_{\mathrm{redCCO}}$ characterization. Here, we use the Beer–Lambert model with scattering, as it provides a closer fit to real spectra than the model without scattering. We observe that performing the spectral unmixing on the HSI measurement of brain tissue allows us to better contrast the vessel tree (b) and tumor area (c) than on the RGB image.