On the establishment of reference values of clouds of electromyography interference pattern by linear regression method and percentile method and comparison of sensitivity and specificity of both methods

OBJECTIVE: Turn-amplitude clouds were widely used in automatic electromyography (EMG) interference pattern analysis. Earlier works employed the intercept ± 2SD (standard deviation) of the linear regression equation as the upper and lower boundaries of the clouds. The goal of this study was to employ the linear regression method and percentile method to calculate the reference value of turn-amplitude clouds, identify the determining criteria in accordance with the receiver operator characteristic curve (ROC), and analyze the sensitivity and specificity of the linear regression cloud, percentile cloud, and quantitative assessment of the motor unit potential (QMUP). METHODS: First, we explore what factors affect the number of turns per second and the mean amplitude. Then, their logarithms were taken for the normal test. All muscle data were used to calculate the reference values of percentile clouds. However, the reference values of the linear regression clouds were obtained for the muscles with a bivariate normal distribution, homogeneous variances and a linear correlation. We calculated the prediction interval with the standard errors of the intercept and slope of the linear regression equation, which can determine the upper and lower boundaries of the linear regression clouds. Furthermore, we obtained ROCs of these clouds, which were used as the determining criteria to determine the optimum cut-off values. Finally, our study analyzed the sensitivity and specificity of the linear regression cloud, percentile cloud, and QMUP. RESULTS: We here presented the reference values and ROCs of the linear regression clouds and percentile clouds. We suggest the determining criteria be based on ROCs. The areas under the curve (AUC) of both clouds are larger than 0.8, revealing that they have significant diagnostic value. Our results display that the specificities of the linear regression clouds, percentile clouds, and QMUP are almost identical to each other, whereas the sensitivity of percentile cloud is higher than those of QMUP and linear regression clouds. CONCLUSION: According to ROCs, the researchers determine the determining criteria of the linear regression clouds and percentile clouds. Our findings suggest that the percentile clouds possess a wide application range and significant diagnostic value, therefore it may be the optimum for automatic EMG interference pattern analysis.