Application of Convolutional Neural Networks to Four-Class Motor Imagery Classification Problem

The increased popularity of brain-computer interfaces (BCIs) has created a new demand for miniaturized and low-cost electroencephalogram (EEG) acquisition devices for entertainment, rehabilitation, and scientific needs. The lack of scientific analysis for such system design, modularity, and unified validation tends to suppress progress in this field and limit supply for new low-cost device availability. To eliminate this problem, this paper presents the design and evaluation of a compact, modular, battery powered, conventional EEG signal acquisition board based on an ADS1298 analog front-end chip. The introduction of this novel, vertically stackable board allows the EEG scaling problem to be solved by effectively reconfiguring hardware for small or more demanding applications. The ability to capture 16 to 64 EEG channels at sample rates from 250 Hz to 1000 Hz and to transfer raw EEG signal over a Bluetooth or Wi-Fi interface was implemented. Furthermore, simple but effective assessment techniques were used for system evaluation. While conducted tests confirm the validity of the system against official datasheet specifications and for real-world applications, the proposed quality verification methods can be further employed for analyzing other similar EEG devices in the future. With 6.59 microvolts peak-to-peak input referred noise and a −97 dB common mode rejection ratio in 0–70 Hz band, the proposed design can be qualified as a low-cost precision cEEG research device.

Privacy and data security have become the new hot topic for regulators in recent years. As a result, Federated Learning (FL) (also called collaborative learning) has emerged as a new training paradigm that allows multiple, geographically distributed nodes to learn a Deep Learning (DL) model together without sharing their data. Blockchain is becoming a new trend as data protection and privacy are concerns in many sectors. Technology is leading the world and transforming into a global village where everything is accessible and transparent. We have presented a blockchain enabled security model using FL that can generate an enhanced DL model without sharing data and improve privacy through higher security and access rights to data. However, existing FL approaches also have unique security vulnerabilities that malicious actors can exploit and compromise the trained model. The FL method is compared to the other known approaches. Users are more likely to choose the latter option, i.e., providing local but private data to the server and using ML apps, performing ML operations on the devices without benefiting from other users’ data, and preventing direct access to raw data and local training of ML models. FL protects data privacy and reduces data transfer overhead by storing raw data on devices and combining locally computed model updates. We have investigated the feasibility of data and model poisoning attacks under a blockchain-enabled FL system built alongside the Ethereum network and the traditional FL system (without blockchain). This work fills a knowledge gap by proposing a transparent incentive mechanism that can encourage good behavior among participating decentralized nodes and avoid common problems and provides knowledge for the FL security literature by investigating current FL systems.

Electroencephalography (EEG) is widely used in clinical diagnosis, monitoring and Brain - Computer Interface systems. Usually EEG signals are recorded with several electrodes and transmitted through a communication channel for further processing. In order to decrease communication bandwidth and transmission time in portable or low cost devices, data compression is required. In this paper we consider the use of fast Discrete Cosine Transform (DCT) algorithms for lossy EEG data compression. Using this approach, the signal is partitioned into a set of 8 samples and each set is DCT-transformed. The least-significant transform coefficients are removed before transmission and are filled with zeros before an inverse transform. We conclude that this method can be used in real-time embedded systems, where low computational complexity and high speed is required.

Brain-Computer interface (BCI) systems require intensive signal processing in order to form control signals for electronic devices. The majority of BCI systems work by reading and interpreting cortically evoked electro-potentials across the scalp via an electro-encephalogram (EEG). An important factor affecting the efficiency of BCI is the number of EEG features. To reduce the number of features is an important way to improve the speed. In this paper, we consider application of discrete cosine transform (DCT) on EEG signals. DCT takes correlated input data and concentrates its energy in just first few transform coefficients. This method is used as a feature extraction step and allows data size reduction without losing important information. For classification we are using artificial neural networks with different number of hidden neurons and training functions. We conclude that the method can be successfully used for the feature extraction and dataset reduction.