A Flexible Software-Hardware Framework for Brain EEG Multiple Artifact Identification

Abstract

This chapter presents an energy efficient and flexible multichannel electroencephalogram (EEG) artifact identification software-hardware framework using depthwise and separable convolutional neural networks (DS-CNN). EEG signals are recordings of the brain activities. The EEG recordings that are not originated from cerebral activities are termed as artifacts. Our proposed model does not need expert knowledge for feature extraction or preprocessing of EEG data and has a very efficient architecture implementable on mobile devices. The network presented in this chapter can be reconfigured for any number of EEG channel and artifact classes. Experiments were done with the proposed model with the goal of maximizing the identification accuracy while minimizing the weight parameters and required number of operations. The network presented in this chapter achieves 93.14% classification accuracy using EEG dataset collected by a 64 channel BioSemi ActiveTwo headsets, averaged across 17 patients and 10 artifact classes. The hardware architecture designed in this chapter is fully parameterized with number of input channels, filters, depth, and data bit-width. The number of processing engines (PE) in the proposed hardware can vary between 1 and 16 providing different latency, throughput, power, and energy efficiency measurements. A custom hardware architecture is implemented on Xilinx FPGA (Artix-7) which on average consumes 1.4–4.7 mJ dynamic energy with different PE configurations. Energy consumption is further reduced by 16.7× implementing on application-specified integrated circuit (ASIC) at the post-layout level in 65-nm CMOS technology. The FPGA implementation is 1.7× to 5.15× higher energy efficient than some previous works. Moreover, the ASIC implementation is also 8.47× to 25.79× higher energy efficient compared to previous works. In this chapter, it is also demonstrated that the network is reconfigurable to detect artifacts from another EEG dataset collected in a lab by a 14 channel Emotiv EPOC+ headset and achieved 93.5% accuracy for eye blink artifact detection.