LEADNet: Detection of Alzheimer’s Disease Using Spatiotemporal EEG Analysis and Low-Complexity CNN

IEEE Access, Год журнала: 2024, Номер 12, С. 113888 - 113897

Опубликована: Янв. 1, 2024

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

Язык: Английский

---

Intra- and Inter-Regional Complexity in Multi-Channel Awake EEG Through Multivariate Multiscale Dispersion Entropy for Assessing Sleep Quality and Aging

Biosensors, Год журнала: 2025, Номер 15(4), С. 240 - 240

Опубликована: Апрель 9, 2025

Aging and poor sleep quality are associated with altered brain dynamics, yet current electroencephalography (EEG) analyses often overlook regional complexity. This study addresses this gap by introducing a novel integration of intra- inter-regional complexity analysis using multivariate multiscale dispersion entropy (mvMDE) from awake resting-state EEG for the first time. Moreover, assessing both provides comprehensive perspective on dynamic interplay between localized neural activity its coordination across regions, which is essential understanding substrates aging quality. Data 58 participants—24 young adults (mean age = 24.7 ± 3.4) 34 older 72.9 4.2)—were analyzed, each group further divided based Pittsburgh Sleep Quality Index (PSQI) scores. To capture complexity, mvMDE was applied to most informative sensors, one sensor selected region four methods: highest average correlation, entropy, mutual information, principal component loading. targeted approach reduced computational cost enhanced effect sizes (ESs), particularly at large scale factors (e.g., 25) linked delta-band activity, PCA-based method achieving ESs (1.043 in adults). Overall, we expect that inter- intra-regional will play pivotal role elucidating mechanisms as captured various physiological data modalities—such EEG, magnetoencephalography, magnetic resonance imaging—thereby offering promising insights range biomedical applications.

Язык: Английский

---

EEG features associated with Alzheimer’s disease and Frontotemporal dementia are not reflected by processed indices used in anesthesia monitoring

Journal of Clinical Monitoring and Computing, Год журнала: 2025, Номер unknown

Опубликована: Апрель 21, 2025

Abstract Patients with dementia face increased risks after general anesthesia. Improved perioperative electroencephalogram (EEG) monitoring techniques could aid in identifying vulnerable patients. However, current technology relies on processed indices to measure “depth-of-anesthesia”. Analyzing OpenNeuro Dataset ds004504, we compared resting-state, eyes-closed EEG recordings of healthy controls ( n = 27) patients diagnosed Alzheimer’s disease (AD, 35) and Frontotemporal (FTD, 23). We focused prefrontal recordings. Analysis included spectral analysis, the “fitting-oscillations&-one-over-f”-algorithm for aperiodic periodic signal features, as well calculations openibis, permutation entropy (PeEn), (SpEn), edge frequency (SEF). Spectral differences were pronounced, including a higher alpha/theta-ratio (2.62 [95%CI: 1.54–3.62]) both AD (0.55 0.26–1.92], P < 0.001, AUC: 0.765 [0.642–0.888]) FTD (0.83 0.33–1.65], 0.007, 0.779 [0.652–0.907]). Oscillatory peak detection within alpha band was more robust control (versus AD: 0.003, Cramér’s V 0.374; versus FTD: 0.414). Processed index parameters did not show clear trend. associated openibis (95.53 93.43–97.39]) than (91.98 89.46–96.27], 0.033, 0.717 [0.572–0.862]) an elevated SEF (23.68 14.10–25.57] Hz) (16.60 14.22–22.22] Hz, 0.041, 0.676 [0.532–0.821]). are baseline abnormalities, standard montage, used intraoperatively, present promising technical screening approach cognitive vulnerability. these features obscured by currently neuroanesthesia monitoring. ds004504 “A dataset from: disease, Healthy subjects” (doi: https://doi.org/10.18112/openneuro.ds004504.v1.0.7 ).

Язык: Английский

---

Implications of Aperiodic and Periodic EEG Components in Classification of Major Depressive Disorder from Source and Electrode Perspectives

Sensors, Год журнала: 2024, Номер 24(18), С. 6103 - 6103

Опубликована: Сен. 21, 2024

Electroencephalography (EEG) is useful for studying brain activity in major depressive disorder (MDD), particularly focusing on theta and alpha frequency bands via power spectral density (PSD). However, PSD-based analysis has often produced inconsistent results due to difficulties distinguishing between periodic aperiodic components of EEG signals. We analyzed data from 114 young adults, including 74 healthy controls (HCs) 40 MDD patients, assessing alongside conventional PSD at both source electrode levels. Machine learning algorithms classified versus HC based these features. Sensor-level showed stronger Hedge’s g effect sizes parietal frontal than source-level analysis. individuals exhibited reduced relative HC. Logistic regression-based classifications that slightly outperformed PSD, with the best achieved by combining features (AUC = 0.82). Strong negative correlations were found activities higher scores Beck Depression Inventory, anhedonia subscale. This study emphasizes superiority sensor-level over detecting MDD-related changes highlights value incorporating a more refined understanding disorders.

Язык: Английский

---

Spatial Cognitive EEG Feature Extraction and Classification Based on MSSECNN and PCMI

Bioengineering, Год журнала: 2024, Номер 12(1), С. 25 - 25

Опубликована: Дек. 31, 2024

With the aging population rising, decline in spatial cognitive ability has become a critical issue affecting quality of life among elderly. Electroencephalogram (EEG) signal analysis presents substantial potential assessments. However, conventional methods struggle to effectively classify states, particularly tasks requiring multi-class discrimination pre- and post-training states. This study proposes novel approach for EEG classification, utilizing Permutation Conditional Mutual Information (PCMI) feature extraction Multi-Scale Squeezed Excitation Convolutional Neural Network (MSSECNN) model classification. Specifically, MSSECNN classifies states into two classes-before after training-based on features. First, PCMI extracts nonlinear features, generating matrices across different channels. SENet then adaptively weights these highlighting key Finally, MSCNN captures local global features using convolution kernels varying sizes, enhancing classification accuracy robustness. systematically validates training data from brain-controlled car manually operated UAV tasks, with state assessments performed through cognition games combined signals. The experimental findings demonstrate that proposed significantly outperforms traditional methods, offering superior accuracy, robustness, capabilities. model's advantages provide valuable technical support early identification intervention decline.

Язык: Английский

---

Research on the Application of Improved Particle Swarm Optimization Algorithm Assisted Multi-feature EEG Data Processing in the Diagnosis of Frontotemporal Dementia

Опубликована: Дек. 27, 2024

Язык: Английский

---