A neural network approach to electromyographic signal processing for a motor control task

The authors propose a novel signal processing technique employing both neural networks and classical signal processing methods to effectively map the surface electrical signal concomitant with muscle contraction (EMG) to human muscle activation. With a computational musculoskeletal model it is shown that these predicted muscle activations, accurately estimate joint torque for various ballistic flexion exercises. Through the systems ability to generalize across exercise trials and predict a classic ballistic triphasic activation pattern, a hybrid musculoskeletal system may be able to accurately and reliably model extremely complex physiological systems with clinical implications.