Prediction of Handgrip Strength Based on Convolutional Neural Network Using Electromyography Signals

Rehabilitation is essential in restoring the function of damaged limbs caused by post-stroke conditions, especially in the fingers. As time goes by, traditional techniques in rehabilitation cause boredom and loss of motivation in post-stroke patients. Technology is needed to support the rehabilitation process, which can increase motivation and recovery. In this era, rehabilitation technology has developed rapidly, such as with the help of therapeutic robots based on Artificial Intelligence to predict hand grip strength based on electromyography (EMG) signals. Many AI algorithms, especially deep learning, are used to predict hand grip strength based on EMG signals. This research focuses on evaluating a variation of the Convolutional Neural Network (CNN) deep learning algorithm to predict hand grip strength based on EMG signals in the forearm muscles. Variations of CNN used include 3-layers CNN, which were created independently and VGG16 CNN. Evaluation is calculated using the R2 score and Root Mean Square Error (RMSE) parameters. This research involved five subjects in healthy conditions. The results of the test show that the 3-layers CNN has superior performance compared to the VGG16 CNN. In testing the model in each subject, the R2 value reached 0.915, and RMSE reached 0.871.