A neural network approach to motor-sensory relations during postural disturbance.

This study explored whether artificial neural networks (ANN) can be used to quantify the motor-sensory relationship during postural disturbance. An ANN model was constructed with seven mechanical stimuli to the visual, vestibular and somatosensory systems (i.e., head angular and linear accelerations, eye-target distance, ankle joint rotation and velocity, as well as normal and shear ground contact forces under the feet) as inputs, and electromyographic activities of tibialis anterior and gastrocnemius muscles as outputs. These inputs and outputs were directly measured during a sudden toes-up-down rotation of the supporting base in two groups of elderly subjects: people with peripheral neuropathy (NP) who have severe loss of mechanoreception in the sole of their feet and people without NP. The products of ANN weights were used in a summary statistic called the Q-value to estimate the contribution of each mechanical stimulus to sensory systems in determining each leg muscle activity. It was found that: (1) the stimuli to the vestibular system and/or ankle proprioceptors have greater contributions to leg muscle activities, especially the TA muscle, in people with NP than people without NP; (2) the stimuli to somatosensory receptors have the greatest contribution, and the stimuli to the vestibular system have the least contribution to both muscle activities in both groups. These findings are supported by previous studies and have demonstrated the potential of the Q-value concept in the ANN model in studying the motor-sensory relationship in human postural control.