Assistive Human-Machine Interfaces via Artificial Neural Networks

This paper reports the study and results of modeling and online compensating of movement disorder stemming from multiple sclerosis (MS) via artificial neural networks. We trained and tested a cascade-correlation neural network with Kalman filtering on data collected from 11 subjects with MS. The test subjects use headcontrolled mouse emulators to move a cursor to a series of random targets on screen. Simulated real-time testing of the trained neural networks shows that the networks successfully make the cursor trajectories of all the 11 subjects less chaotic, and hence more controllable. The neural networks also reduce the time needed to reach the targets by an average of 31.8%. The neural network approach can be easily applied to other human-machine interfaces such as computer mice and joysticks, or powered wheelchairs. This technique is also applicable to movement disorders resulting from certain geriatric diseases such as Parkinson’s disease and essential tremor.