Specific features of chromophore concentration estimation from multispectral data
D. Musaeva¹, T. Torokhov¹,³, B. Sheludko², S. German¹, R. Korobeinikov¹, M. Kuzyuk¹, V. Shibaeva¹, S. Korchagin², E. Ershov², D. Gorin¹, S. Perkov¹,²
¹Skolkovo Institute of Science and Technology (Skoltech), Moscow, Russia; ²Kharkevich Institute for Information Transmission Problems, Russian Academy of Sciences, Moscow, Russia ³Prokhorov General Physics Institute, Russian Academy of Sciences, Moscow, Russia
Abstract
Multispectral imaging occupies an intermediate position between RGB photography and hyperspectral imaging: it provides a few to a few tens of spectral samples rather than hundreds, yet acquires a full image in a single short exposure and allows the illumination bands to be chosen freely. The talk examines the advantages and the drawbacks of multispectral data with respect to both RGB imaging and hyperspectral imaging — in terms of spectral information content, acquisition time and robustness of the subsequent processing — together with the techniques that compensate for its weaknesses. The setup consists of eight narrow-band LEDs (450–939 nm, with individually measured emission spectra) and an RGB camera with a Bayer sensor, recording raw frames; a Spectralon standard is imaged as the flat-field reference, and the retrieval is validated against a ColorChecker Digital SG chart of certified reflectance. The samples are liquid phantoms (bilirubin 8–263 µM, hemoglobin, Lipofundin as the scatterer) and gelatin phantoms.
The following are addressed: approaches to per-band reflectance retrieval (raw frames processing, ways of combining the color channels; reaches an RMSE 0.021 and correlation 0.98–0.996 against known reflectance); attempts to reconstruct a continuous reflectance spectrum — direct inversion proves unstable (effective spectral dimensionality ≈ 3.5), whereas projection onto a statistical basis yields R² = 0.955 but cannot represent narrow absorption bands; and concentration estimation — hemoglobin is recovered with a coefficient of variation of 1.3%, bilirubin by a narrow-band differential estimator with 1–7% error down to 33 µM and a detection limit of ~13 µM, which, for the present instrument configuration and single-frame exposure, matches the photon-noise limit of the measurement. Further developments are discussed: frame averaging, correction for the additive stray light that shortens the effective optical pathlength, and a procedure for selecting the minimum set of illumination bands needed for the joint determination of n chromophores.
Speaker
Musaeva Dariia
Skolkovo Institute of Science and Technology (Skoltech), Moscow, Russia
Russian Federation
Discussion
Ask question
