This paper studies improvement of the clinical practicality of the control command input method for motor disabled patients using the recognition of specific motions via an artificial neural network (ANN). Several good ANNs that had different combinations of input signals and/or different number of neurons in the hidden layer were selected for each subject. The final recognition was determined by the majority decision rule by three good ANNs. The AND operation of the outputs of selected ANNs was also used to reduce the number of misrecognitions. The results with neurologically intact subjects and a hemiplegic patient showed that the proposed method would be effective clinically compared to the method using a single ANN with a fixed combination of input signals for all patients. It was also shown that trained ANNs with the proposed method could be used on other days with good performance.
