Competitive evaluation of data mining algorithms for use in classification of leukocyte subtypes with Raman microspectroscopy

• Published in: Analyst (London) 2015-04, Vol.14 (7), p.2473-2481
• Main Authors: Maguire, A, Vega-Carrascal, I, Bryant, J, White, L, Howe, O, Lyng, F. M, Meade, A. D
• Format: Article
• Language: English
• Subjects: Algorithms; Cellular; Classification; Data mining; Data Mining - methods; Devices; Discriminant Analysis; Humans; Leukocytes; Leukocytes - classification; Lymphocytes; Principal Component Analysis; Reduction; Spectrometers; Spectrum Analysis, Raman; Support Vector Machine
• ISSN: 0003-2654; 1364-5528
• DOI: 10.1039/c4an01887g

Raman microspectroscopy has been investigated for some time for use in label-free cell sorting devices. These approaches require coupling of the Raman spectrometer to complex data mining algorithms for identification of cellular subtypes such as the leukocyte subpopulations of lymphocytes and monocytes. In this study, three distinct multivariate classification approaches, (PCA-LDA, SVMs and Random Forests) are developed and tested on their ability to classify the cellular subtype in extracted peripheral blood mononuclear cells (T-cell lymphocytes from myeloid cells), and are evaluated in terms of their respective classification performance. A strategy for optimisation of each of the classification algorithm is presented with emphasis on reduction of model complexity in each of the algorithms. The relative classification performance and performance characteristics are highlighted, overall suggesting the radial basis function SVM as a robust option for classification of leukocytes with Raman microspectroscopy. In this study Raman spectral data from peripheral blood mononuclear cells (PBMCs) is used for the competitive evaluation of three data-mining models in discriminating a highly pure population of T-cell lymphocytes from other myeloid cells within the PBMCs fraction.