Raman spectroscopy tool (RS-tool)

Implementation of a full cycle of biomedical machine learning classification problem. The program is designed for a realtime full cycle of processing Raman spectra: preprocessing, decomposition of spectra, creation of classifier models for further work with new measured spectra. New original ExModPoly (Extended ModPoly algorithm for baseline correction) and decomposition methods available.

Installation

1. Download the installer in the Releases section and install.
2. Install Graphviz https://graphviz.org/download/. (only needed to display XGBoost trees).

Program functionality

• Import of files in .txt, .asc format as a two-dimensional array with spectra.
• Saving all data in a .zip project file. All information is converted to binary format and compressed in a zip archive with the maximum compression ratio.
• Interpolation to bring spectra with different wavelength ranges to one nm range.
• Despike (removal of high-frequency random noise (electrical interference)).
• Conversion of spectra from the nm wavelength range to wavenumbers cm-1.
• Normalization of spectra by 6 methods: EMSC, SNV, Area, Trapezoidal rule area, Max intensity, Min-max intensity.
• Smoothing of spectra by 14 methods: MLESG, CEEMDAN, EEMD, EMD, Savitsky-Golay filter, Whittaker smoother, Flat window, Hanning window, Hamming window, Bartlett window, Blackman window, Kaiser window, Median filter, Wiener filter.
• Baseline correction using 42 methods:: Poly, ModPoly, iModPoly, ExModPoly, Penalized poly, LOESS, Quantile regression, Goldindec, AsLS, iAsLS, arPLS, airPLS, drPLS, iarPLS, asPLS, psaLSA, DerPSALSA, MPLS, iMor, MorMol, AMorMol, MPSpline, JBCD, Mixture Model, IRSQR, Corner-Cutting, RIA, Dietrich, Golotvin, Std Distribution, FastChrom, FABC.
• Raman spectrum decomposition algorithm.
• Line profiles used for decomposition: Gaussian, Split Gaussian, Skewed Gaussian, Lorentzian, Split Lorentzian, Voigt, Split Voigt, Skewed Voigt, Pseudo Voigt, Split Pseudo Voigt, Pearson4, Split Pearson4, Pearson7, Split Pearson7.
• Optimization methods used in line decomposition: "Levenberg-Marquardt", "Least-Squares, 'Nelder-Mead', 'L-BFGS-B', 'Powell', 'Conjugate-Gradient', 'Cobyla', 'BFGS', 'Truncated Newton', 'trust-region for constrained optimization', 'Sequential Linear Squares Programming', 'Basin-hopping', 'Adaptive Memory Programming for Global Optimization', 'Dual Annealing optimization', 'Simplicial Homology Global Optimization'
• Training classifiers and classifying dimensions using models: 'LDA', 'Logistic regression', 'SVC', 'Decision Tree', 'Random Forest', XGBoost.
• Reducing the dimensionality of data using the method PCA.
• Possibility of using the program for classification of new measured spectra according to trained models.
• Automatic generation of all necessary graphs for articles and the ability to save them in various formats.
• Intuitive graphical interface.

Citation

Decomposition Method for Raman Spectra of Dentine. DOI: 10.18287/JBPE24.10.030303. https://jbpe.ssau.ru/index.php/JBPE/article/view/9074