Detection of human movement intention based on a multilayer feed-forward neural network with dictionary learning

Detection of human movement intention can be used for effective neuro-feedback in rehabilitation and self-paced brain computer interface (BCI) applications. Movement-related cortical potentials (MRCPs), representing movement intention, preparation, and execution, is a slow negative deflection observed in EEG signals up to 1.5s prior to the onset of movement. However, detecting movement intention from MRCPs is still a challenging task. In this paper, we proposed a new movement intention detection model using a multilayer feed-forward neural network (MFNN) with dictionary learning during a voluntary reaching task. The detection features were extracted automatically using the dictionary learning algorithm. The detection performances of MFNN were compared with the common classifiers, i.e. random forest (RF) and support vector machine (SVM), in the EEG analysis field. Our results showed that the detection performances of MFNN outperformed those of RF and SVM. Moreover, the detection accuracies were further improved with the features extracted by the dictionary learning method. And the optimal number of dictionary learning components was proved to be 5. In conclusion, the proposed MFNN with dictionary learning based movement intention detection method is superior to previous approaches and is a promising technique in EEG signal analysis and BCI applications where it is difficult to obtain large datasets from individual subject.