Search

Abstract: A brain machine interface (BMI) provides the possibility of controlling such external devices as prosthetic arms for patients with severe motor dysfunction using their own brain signals. However, there have been few studies investigating the decoding accuracy for multiclasses of useful unilateral upper limb movements using non-invasive measurements. We investigated the decoding accuracy for classifying three types of unilateral upper limb movements using single-trial magnetoencephalography (MEG) signals. Neuromagnetic activities were recorded in 9 healthy subjects performing 3 types of right upper limb movements: hand grasping, pinching, and elbow flexion. A support vector machine was used to classify the single-trial MEG signals. The movement types were predicted with an average accuracy of 66±10% (chance level: 33.3%) using neuromagnetic activity during a 400-ms interval (−200ms to 200ms from movement onsets). To explore the time-dependency of the decoding accuracy, we also examined the time course of decoding accuracy in 50-ms sliding windows from −500ms to 500ms. Decoding accuracies significantly increased and peaked once before (50.1±4.9%) and twice after (58.5±7.5% and 64.4±7.6%) movement onsets in all subjects. Significant variability in the decoding features in the first peak was evident in the channels over the parietal area and in the second and third peaks in the channels over the sensorimotor area. Our results indicate that the three types of unilateral upper limb movement can be inferred with high accuracy by detecting differences in movement-related brain activity in the parietal and sensorimotor areas. [Copyright &y& Elsevier]