Assessment of robustness and transferability of classification models built for cancer diagnostics using Raman spectroscopy

摘要: Over recent years, Raman spectroscopy has been demonstrated as a prospective tool for application in cancer diagnostics. The use of this purpose relies on pattern recognition methods that have developed to perform well data achieved under laboratory conditions. However, the routine clinical is likely result imperfect due instrument-to-instrument variation. Such corruption pure tissue spectral expected negatively impact classification performance diagnostic model. In paper, we present thorough assessment robustness approach. This was by perturbing set spectra different ways, including various linear shifts, nonlinear shifts and random noise using previously optimised models predict class membership each spectrum testing set. loss predictive power with increased used calculate score, which allows an easy comparison model robustness. For approach, three types models, discriminant analysis (LDA), partial least square (PLS-DA) support vector machine (SVM), built lymph node diagnostics were subject testing. results showed perturbation had highest all models. Among methods, gradient y-shift resulted loss. Thus, factor most affect outcome when systems y-shift. Copyright © 2010 John Wiley & Sons, Ltd.