Convolutional Neural Networks with Reusable Full-Dimension-Long Layers for Feature Selection and Classification of Motor
In the present article the author addresses the task of classification of motor imagery in EEG signals by proposing innovative architecture of neural network. Despite all the successes of deep learning, neural networks of significant depth could not ensur
- PDF / 1,126,454 Bytes
- 13 Pages / 439.37 x 666.142 pts Page_size
- 36 Downloads / 197 Views
Abstract. In the present article the author addresses the task of classification of motor imagery in EEG signals by proposing innovative architecture of neural network. Despite all the successes of deep learning, neural networks of significant depth could not ensure better performance compared to shallow architectures. The approach presented in the article employs this idea, making use of yet shallower, but productive architecture. The main idea of the proposed architecture is based on three points: full-dimension-long ‘valid’ convolutions, dense connections - combination of layer’s input and output and layer reuse. Another aspect addressed in the paper is related to interpretable machine learning. Interpretability is extremely important in medicine, where decisions must be taken on the basis of solid arguments and clear reasons. Being shallow, the architecture could be used for feature selection by interpreting the layers’ weights, which allows understanding of the knowledge about the data cumulated in the network’s layers. The approach, based on a fuzzy measure, allows using Choquet integral to aggregate the knowledge generated in the layer weights and understanding which features (EEG electrodes) provide the most essential information. The approach allows lowering feature number from 64 to 14 with an insignificant drop of accuracy (less than a percentage point). Keywords: Motor imagery · Feature selection · Convolutional neural network · Reusable convolutions
1 Introduction While analyzing EEG signals, researchers discovered an effect appearing when an examined person was imaging hand movements without even actually performing them [7]. The effect was called motor imagery and attracted serious attention of researchers. Further research revealed, that one of the fields, where motor imagery could be used in therapy purposes, was post-stroke rehabilitation, helping patients to recover faster [2, 4].
© Springer Nature Switzerland AG 2020 I. Farkaš et al. (Eds.): ICANN 2020, LNCS 12396, pp. 79–91, 2020. https://doi.org/10.1007/978-3-030-61609-0_7
80
M. Tokovarov
Another field, where motor imagery presents interest, is brain computer interface (BCI): it could potentially become a source of control signals for brain computer interfaces, able to help severely disabled people in communication and rehabilitation [14]. The author of the present research proposes a novel model which can solve the task of subject-independent classification with a high level of accuracy. The structure of the model also allows to interpret the contribution of each EEG electrode, thus performing feature selection and ensuring interpretability of the model. The proposed method of feature selection allows to significantly decrease the number of electrodes: down to 14 from 64 electrodes by the cost of slightly lower accuracy (less than a percentage point). The method of network interpretation based on fuzzy integral calculation is also proposed, so it does not simply compute the weighed sum of all filters, but takes into account their intera
Data Loading...
