Classification of electromyographic signals by their entropic characteristics for differential diagnostics of low back pain using the random forest method

Abstract

The article presents the results of an investigation into the possibilities of using entropic characteristics of electromyographic (EMG) signals for differential diagnostics of pain syndromes in the lumbar spine. Two independent sets of EMG signals recorded in three diagnostic groups were used as initial data: healthy individuals without complaints of back pain, conditionally healthy individuals with complaints of pain (dysfunctional pain), and patients with degenerative diseases of the spine (functional pain). Four entropy indicators were selected to describe the signals – median and mean entropy, as well as median and mean spectral entropy. The random forest algorithm was used as a classification method. Model training was carried out on a data set with a significant class imbalance, and testing was performed using another, independent set. During the study, the number of trees in the ensemble was varied (100, 200, 300 and 500), and the feasibility of taking into account weighting coefficients inversely proportional to the representation of classes in the data was also tested. The quality of the models was assessed based on the classification accuracy, F1-score, area under the ROC curve (AUC), ROC curve and confusion matrix. The results obtained showed that increasing the number of trees above 100 does not provide an increase in the quality of classification, and weighting the training data in most cases does not improve the results compared to unweighted models. The best indicators were achieved when distinguishing the group of patients with dysfunctional pain from the group with functional pain: the F1-score was 0.99, the AUC was 1.00, and the accuracy was 99.09%. Thus, the results confirm the feasibility of using the entropic characteristics of EMG signals in combination with the random forest method to create reliable clinical decision support tools. The greatest diagnostic value is the ability to distinguish the type of pain syndrome (dysfunctional or functional), which can contribute to a more informed choice of treatment tactics in patients with low back pain.