The surface-enhanced Raman scattering method for point-of-care atrial fibrillation diagnostics

We suggest a new method for the detection of paroxysmal atrial fibrillation by analyzing surface-enhanced Raman scattering (SERS) spectra of blood serum of patients in question in comparison with SERS spectra of the serum of healthy donors. Spectral measurements were carried out on compact SERS substrates in dried blood serum droplets with immediate subsequent processing. To process the spectra, machine learning methods were used, in particular, the logistic regression method and the principal component method. Furthermore, thanks to the possibility of the physical-chemical interpretation of the coefficients of the method, the vibrational bands responsible for the signs of atrial fibrillation were identified and their correlation was carried out. Evaluation metrics were presented for the classification, among which the accuracy value was 0.82, that is a high indicator when analyzing samples directly from the blood serum of patients with the disease under study. It was shown that a small number of measured spectra for each sample (near 35 measurements) was sufficient to carry out the study. A comparative analysis of the logistic regression method and other commonly used machine learning methods was also carried out: support vector machines and random forest. Each method was evaluated and the advantages of logistic regression in solving the problem presented in this study were shown. The receiver operating characteristic curve (ROC) analysis was also used for graphical representation and comparison of methods. The presented study shows the prospects for using the described method for the analysis of diseases associated with cardiac risks.