Adaptation of Convolutional Neural Networks for Multi-Channel Artifact Detection in Chronically Recorded Local Field Potentials

Neural recording, known as local field potentials, offer valuable knowledge on how neural processes work and contribute to neural circuits. The recording can be contaminated by different internal and external sources of noise, because of the involvement of complex electronic apparatuses and the natural electrical activity throughout the body. In order to successfully utilize these signal, artifacts must be identified and removed. Thus, in this paper, an artifact detection method using a one-dimensional convolutional network referred to as 1D-CNN is proposed. The presented method achieved an improved accuracy and reduced computational time over the existing methods which use a multi-layered feed-forward neural network and a long-short term memory network. It also provided insight of the criteria behind the classification with gradient attribution maps.