Entropy-Based Surface Electromyogram Feature Extraction for Knee Osteoarthritis Classification

Knee osteoarthritis (KOA) is one of the major causes of lower limb disability. This study aims to develop a computer-based approach to discriminate KOA individuals from controls by using entropy-based features, and therefore to provide an auxiliary, quantitative tool for KOA diagnosis. The surface EMG (sEMG) data were collected from the vastus lateralis, vastus medialis, biceps femoris, and semitendinosus when KOA participants and controls were walking barefoot on ground at a self-paced speed. We employed and compared three different entropy measures, including 1) approximate entropy, 2) sample entropy, 3) fuzzy entropy, for extracting KOA-related features from the sEMG signals for classification. The differences between the KOA group and healthy controls are primarily shown in the fuzzy entropy features extracted from the vastus medialis and biceps femoris muscle pair. Among all tested measures, the fuzzy entropy yielded the best performance in distinguishing KOA patients from controls, with 92% of accuracy, 91.43% of sensitivity and 93.33% of specificity. The results indicate that the fuzzy entropy method is applicable for extracting KOA-related features from sEMG, which can be developed as a sensitive metric for computer-assist diagnosis of knee osteoarthritis.