Clinical methods for dementia detection are expensive and prone to human errors. Despite various computer-aided methods using electroencephalography (EEG) signals and artificial intelligence, a consistent separation of Alzheimer’s disease (AD) and normal-control (NC) subjects remains elusive. This paper proposes a low-complexity EEG-based AD detection CNN called LEADNet to generate disease-specific features. LEADNet employs spatiotemporal EEG signals as input, two convolution layers for feature generation, a max-pooling layer for asymmetric spatiotemporal redundancy reduction, two fully-connected layers for nonlinear feature transformation and selection, and a softmax layer for disease probability prediction. Different quantitative measures are calculated using an open-source AD dataset to compare LEADNet and four pre-trained CNN models. The results show that the lightweight architecture of LEADNet has at least a 150-fold reduction in network parameters and the highest testing accuracy of 98.75% compared to pre-trained models. The investigation of individual layers of LEADNet showed successive improvements in feature transformation and selection for separating AD and NC subjects. A comparison with the state-of-the-art AD detection models showed that the highest accuracy, sensitivity, and specificity were achieved by the LEADNet model.

Alcoholism, a progressive loss of control over alcohol consumption, deteriorates mental and physical health over time. Automatic alcoholism detection can aid in early interventions and timely corrective actions. For this purpose, electroencephalogram (EEG) signals are investigated using explainable artificial intelligence (XAI) techniques to obtain biomarkers. EEG signals are decomposed into five frequency bands to generate vectors of band powers of all channels. These vectors are input to different machine learning models and their ensembles to detect alcoholism. The reliability and generalization of these models are investigated using three oversampling techniques: Synthetic minority over-sampling technique, Adaptive Synthetic Sampling data augmentation, and Gaussian. Comparative analysis using an open-source EEG dataset showed superior performance for artificial neural network (ANN), with an accuracy of 97.36% and an F1-score of 97.88%. The oversampling techniques further improved performance across various ML models, and SMOTE-based ANN outperformed with an accuracy of 97.93% and an F1-score of 97.99%. The best ANN model is investigated using three XAI techniques, Local Interpretable Model-agnostic Explanations (LIME), Submodular Pick LIME, and Morris sensitivity analysis, for explaining the sensitivity of EEG channels and bands in alcoholism detection. The frequency band explanations showed relatively higher performance using beta and gamma bands with 95.45% and 97.36% accuracy, respectively. The EEG channel explanations showed higher performance using biomarkers from parietal and central regions, providing 95.41% and 91.50% accuracy, respectively. A combined explanation of all three XAI techniques indicated that three from the parietal and three from the central regions are the most important for improving detection performance. This study validates the effectiveness of the ANN in detecting alcoholism using EEG signals and emphasizes the significance of frequency bands and EEG channels. These explanations could be applied for early detection and monitoring.