Unsupervised hyperspectral data mining and bioimaging by information entropy and self-modeling curve resolution
share
Unsupervised estimation of the dimensionality of hyperspectral microspectroscopy datasets containing pure and mixed spectral features, and extraction of their representative endmember spectra, remains a challenge in biochemical data mining. We report a new versatile algorithm building on semi-nonnegativity constrained self-modeling curve resolution and information entropy, to estimate the quantity of separable biochemical species from hyperspectral microspectroscopy, and extraction of their representative spectra. The algorithm is benchmarked with established methods from satellite remote sensing, spectral unmixing, and clustering. To demonstrate the widespread applicability of the developed algorithm, we collected hyperspectral datasets using spontaneous Raman, Coherent Anti-stokes Raman Scattering and Fourier Transform IR, of seven reference compounds, an oil-in-water emulsion, and tissue-engineered extracellular matrices on poly-L-lactic acid and porcine jejunum-derived small intestine submucosa scaffolds seeded with bovine chondrocytes. We show the potential of the developed algorithm by consolidating hyperspectral molecular information with sample microstructure, pertinent to fields ranging from gastrophysics to regenerative medicine.
READ FULL TEXT