We report a novel variant of band target entropy minimization (BTEM), adaptive band target entropy minimization (A-BTEM), which offers an improved ability to accurately decompose mixed spectra obtained from complex multicomponent systems. Several key challenges have existed in the application of the basic BTEM approach to decompose spatially offset Raman spectra of bone underneath soft tissues and other multilayer systems demonstrating high collinearity between the spectra of individual components; these have included instabilities, signal mixing, and spectral artefacts that have precluded the reliable use of BTEM in such situations. By using automatic factor selection, an adaptive penalty function in addition to metaheuristic optimization strategies, we demonstrate that high-quality spectral reconstructions of underlying pure component spectra can be obtained with typical correlation coefficients >0.996. Furthermore, we ascertain the behaviour of A-BTEM with different input datasets, both synthetic and real, displaying varying signal-to-noise ratio, signal composition, and numbers of spectra. We thereby identify the multifarious parameter space in the application of A-BTEM and the quality of data required for the most accurate spectral estimates of pure component spectra.
KEYWORDSband target entropy minimization, bone disease, multivariate curve resolution, self-modelling curve resolution, spatially offset Raman spectroscopy