Optimizing Excitation Light for Accurate Rapid Bacterial Species Identification with Autofluorescence.

Rapid identification of bacterial species in patient samples is essential for the treatment of infectious diseases and the economics of health care. In this study, we investigated an algorithm to improve the accuracy of bacterial species identification with fluorescence spectroscopy based on autofluorescence from bacteria, and excitation wavelengths suitable for identification. The diagnostic accuracy of each algorithm for ten bacterial species was verified in a machine learning classifier algorithm. The three machine learning algorithms with the highest diagnostic accuracy, extra tree (ET), logistic regression (LR), and multilayer perceptron (MLP), were used to determine the number and wavelength of excitation wavelengths suitable for the diagnosis of bacterial species. The key excitation wavelengths for the diagnosis of bacterial species were 280 nm, 300 nm, 380 nm, and 480 nm, with 280 nm being the most important. The median diagnostic accuracy was equivalent to that of 200 excitation wavelengths when two excitation wavelengths were used for ET and LR, and three excitation wavelengths for MLP. These results demonstrate that there is an optimum wavelength range of excitation wavelengths required for spectroscopic measurement of bacterial autofluorescence for bacterial species identification, and that measurement of only a few wavelengths in this range is sufficient to achieve sufficient accuracy for diagnosis of bacterial species.