Anatomically and experimentally based neural networks modeling force coordination in static multi-finger tasks

Abstract

The goal of the study was to construct and test a neural network model that incorporates available knowledge about the finger functioning. The fingers are served by the multi-digit muscles that provide the tendons to the four fingers and the uni-digit muscles that serve individual fingers. Such an anatomical arrangement results in the interdependence of the fingers. The network accounts for (1) existence of multi-digit extrinsic and single-digit intrinsic muscles; (2) the force sharing pattern; (3) the ceiling effect (force deficit) and (4) the force enslaving. Human subjects were instructed to voluntarily exert static force with various fingers while relaxing the remaining fingers. Three-layer feedforward neural networks were constructed and tested. The networks were composed of divergent/convergent connections between the layers modeling the multi-digit muscles and one-to-one shortcut connections from the neurons in the input layer to the neurons in the output layer that model the uni-digit muscles. To validate the networks, 19 task subsets were employed for network training and testing. The predicted finger forces closely matched the experimental data lending support to the assumptions on which the network was based. The current paper demonstrates how available anatomical and physiological facts can be utilized to help build networks so as to enhance our understanding of control mechanisms of the neuromotor system.