Accurate identification of inflammation in blood based on laser-induced breakdown spectroscopy using chemometric methods.

Diagnosis of inflammation in the blood requires multiple medical examinations that rely heavily on doctors' experiences, which offer a large room for error. Thus, diagnosis becomes complex and time consuming. Herein, we proposed the use of laser-induced breakdown spectroscopy (LIBS) and chemometric methods to rapidly and accurately diagnose inflammation in blood. Serum samples were collected from 10 healthy people and 10 patients suffering from inflammation in blood. The training and test sets were divided according to each individual in a ratio of 7:3. After excluding considerably different divisions of training and test sets (DTTs) using the Wilcoxon test, 10 DTTs were randomly selected. With these 10 DTTs, backpropagation neural network (BPNN) models were built for substrate optimization, and glass slides exhibited the best performance with an average accuracy of 80.87%. Using glass slides and a multiplicative scatter correction–mean impact value–backpropagation neural network (MSC–MIV–BPNN) model, the accuracy of the best DTT was improved to 93.00%. For the identification effect validation of the improved model, the same features were selected from other DTTs to build new MSC–MIV–BPNN models. The accuracies of all the DTTs were increased by 4.67%–16.33% compared to the original BPNN models. Furthermore, the validation time of a single DTT was ∼0.7 s. These results demonstrated that LIBS combined with the MSC–MIV–BPNN model could accurately and rapidly identify samples with inflammation in blood from healthy blood samples for emergency applications. [Display omitted] • The training and test sets were divided according to each individual. • Significant differences between training and test sets were tested by Wilcoxon test. • The average accuracy was improved from 59.13% to 80.87% by substrate optimization. • With multiple scattering correction and mean impact value methods, the accuracies were improved by 4.67% to 16.33%. [ABSTRACT FROM AUTHOR]