EEG functional connectivity and deep learning for automatic diagnosis of brain disorders: Alzheimer’s disease and schizophrenia

Abstract

Mental disorders are among the leading causes of disability worldwide. The first step in treating these conditions is to obtain an accurate diagnosis. Machine learning algorithms can provide a possible solution to this problem, as we describe in this work. We present a method for the automatic diagnosis of mental disorders based on the matrix of connections obtained from EEG time series and deep learning. We show that our approach can classify patients with Alzheimer’s disease and schizophrenia with a high level of accuracy. The comparison with the traditional cases, that use raw EEG time series, shows that our method provides the highest precision. Therefore, the application of deep neural networks on data from brain connections is a very promising method for the diagnosis of neurological disorders.

Diagnosis of autism spectrum disorder based on functional brain networks and machine learning

“…EEG is a relatively inexpensive method readily available in most contexts and has a good temporal resolution. Data from EEG has been used to study brain organization [13][14][15]. On the other hand, fMRI has a low temporal resolution but high spatial one, thus being well suited for analyses of spatial brain dynamics [16,17].…”

Analysis of functional connectivity using machine learning and deep learning in multimodal data from patients with schizophrenia

“…After the best brain connectivity metric had been determined, the following ML classifiers were used: Random Forest (RF) [56], Naive Bayes (NB) [57], Multilayer Perceptron (MLP) [58], tuned Convolution Neural Network (called here CN N tuned and CN N untuned ) implemented in [59], and Long Short-Term Memory neural networks (LSTM) [60]. In addition to the CNN deep learning used in prior work [37], the LSTM network is a form of recurrent neural network commonly used to identify patterns in time series. Subsequently, the SHAP value method was used for the biological interpretation, as it explains the predictive power of each attribute.…”

“…The EEG dataset used for diagnosis of SCZ, also used in [37], contains a 16-channel EEG time series recorded at a sampling frequency of 128 Hz over one minute, including From these time series are extracted EEG measurements which are widely used in the literature, as there are spectral entropy [111,112], Hjorth mobility and complexity [113][114][115] and Lempel-Ziv complexity [116,117].…”

“…In contrast to articles in the literature that use only one of these levels of abstraction, this study uses all two levels in both data sets in a multimodal fashion for the first time in literature. In addition, we define which of these abstraction levels is the [21,36,37] in finding the best brain areas, connections between brain regions, and measurements of complex networks that may be utilized to evaluate the effects of the SCZ on the brain. As a final result of this research, for the first time in literature, we combine EEG measures extracted from time series and complex network measures as input for the ML method to evaluate which measure is more critical to distinguish SCZ from control patients.…”

Spatial–temporal graph convolutional network for Alzheimer classification based on brain functional connectivity imaging of electroencephalogram

“…Based on the fast Fourier transform for extracting spectral features of EEG for AD diagnosis, Bi and Wang (2019) (Bi & Wang, 2019;Deepthi et al, 2020;Huggins et al, 2021;Ieracitano et al, 2019;Li, Wang, et al, 2021) focus more on learning the locally and continuously changed multiscaled features on the Euclidean space from the EEG signals, neglecting the functional connectivity features. Although Alves et al (2021) used the connection of EEG channels as the CNN input, it neglected the temporal EEG channels features and the input connections topology feature cannot be modelled effectively due to the arranged order of EEG channels.…”