Entropy-based analysis and classification of acute tonic pain from microwave transcranial signals obtained via the microwave-scattering approach

Abstract

The use of microwave technology to detect pain-related neural activities has been demonstrated in the context of a cold pressure test (CPT). However, the selection of appropriate algorithms for extracting and selecting CPT and no pain (NP) features and the quantitative assessment of specific pain types in microwave signals remain problematic. For this purpose, multiscale fluctuation-based dispersion entropy (MFDE) and power spectral Shannon entropy (PSSE) are proposed in this study on the basis of time- and frequency-domain changes as features to ameliorate the problem of concern, respectively. First, the time series is decomposed into several components by using two algorithms, namely, empirical mode decomposition and variational mode decomposition (VMD), and the components in the specified frequency domain are selected in accordance with the spectral diagram. Second, the selected components are used to extract MFDE and PSSE entropy features, and the minimal-redundancy-maximal-relevance (mRMR) criterion and principal component analysis (PCA) algorithms are used to select the features. The performance of different feature selection models is evaluated and compared on the basis of support vector machine (SVM), K-nearest neighbors, linear discriminant analysis, and naive Bayes. Results showed that the highest classification performance is obtained using SVM. The entropy-based features in the VMD-mRMR domain obtain high classification values in accuracy (93.25 %), sensitivity (94.44 %), specificity (90.91 %), positive predictive value (89.47 %), and area under curve (0.8238) in the SVM classifier. This classifier exhibits a broad application prospect for the detection of brain activities and the recognition of microwave neural signals via the microwave-scattering method.