Improving Generalized Regression Neural Networks With Black Widow Optimization Algorithm for Predicting Waist Muscle Strength

As one of the indispensable and important clinical indicators, muscle strength provides an important judgment basis for the diagnosis and rehabilitation of patients with muscle injuries. At present, muscle strength prediction has been applied to rehabilitation equipment for the upper and lower extremities, but there are few studies on muscle strength prediction of lumbar muscle groups. For the muscle strength of the lumbar region, a GRNN model based on the input parameters (lumbar surface electromyography (sEMG) signal and angle information) was developed to predict it in this study. First, the sEMG signals and angle information of the lumbar muscle groups under different movements of the experimental subjects are extracted, and the standard muscle strength is obtained through the lumbar model of OpenSim software, and then established the dataset; then, the black widow optimization (BWO)-GRNN algorithm model is formed by optimizing the parameters of the existing GRNN network, and compared with similar optimization algorithms using the dataset; finally, the effect of the prediction model constructed in this article on the prediction of lumbar muscle strength was verified by basic experiments and clinical trials. According to the experimental results, compared with other prediction models, the GRNN model optimized by the BWO algorithm has a better prediction effect on the muscle strength. Among them,
${R}^{{2}}$
can reach 0.864 in six movements. Compared with GRNN, the prediction accuracy of each action of the improved GRNN model is improved, and the mse is reduced by 92.4% on average. At the same time, in clinical trials, the Bland-Altman plots and ICC plot show that the prediction model is suitable for variety samples.