Predicting hand grip force based on muscle electromyographic activity using artificial intelligence and neural networks

Abstract

The research aims to find predictive values for hand grip strength based on electromyographic activity, in addition to identifying differences between measured grip strength and the predicted grip strength. The research sample included 12 advanced handball players, with their medical records verified. Researchers measured grip strength using a device designed to read Newton force, recording data in real-time with a sampling window of 0.1 seconds. This measurement was synchronized with the recording of muscle electromyographic activity (sEMG) using the Noraxon myoMOTION technique, with a frequency and number of channels set at 400Hz and 8 channels, respectively. The recommended methodology and conditions were strictly adhered to, with the process repeated for each player with complete rest intervals. The following research variables were adopted: peak electromyographic activity, root mean square, time to peak, time ratio between peak and minimum values, average peaks, area under the curve, peak sustain time, peak changes, and voluntary maximum contraction. Grip strength measurements using the designed device were conducted at three stages (50%, 75%, 100%), maintaining the specified intensity for 3 seconds. After data collection, preliminary processing involved isolation and purification to identify the most influential factors. IBM Statistical was the chosen technique for implementing neural networks and using artificial intelligence techniques to process data with a database synchronized using Python. The results generally supported some of the proposed ideas, with interesting findings revealing statistically insignificant and slight differences between recorded and expected grip strength