Investigating Brain Lobe Biomarkers to Enhance Dementia Detection Using EEG Data
Background Introduction
Dementia is a global health issue that significantly impacts patients’ quality of life and places a substantial burden on healthcare systems. Alzheimer’s Disease (AD) and Frontotemporal Dementia (FTD) are two common types of dementia, and their overlapping symptoms make accurate diagnosis and targeted treatment development particularly challenging. Early detection and accurate diagnosis are crucial for effective dementia management. Traditional diagnostic methods, such as clinical assessments and neuroimaging techniques (MRI, PET scans), while effective, are costly, time-consuming, and not widely accessible. Therefore, researchers have begun exploring non-invasive, cost-effective alternatives, such as Electroencephalography (EEG).
EEG captures the brain’s electrical activity through electrodes placed on the scalp, offering high temporal resolution, affordability, and ease of use. Changes in brain function associated with dementia can be reflected in EEG signals, particularly in different brain lobes responsible for cognition, memory, and sensory processing. However, existing EEG-based methods often fail to pinpoint specific biomarkers, especially changes in brain lobes. Thus, studying the role of brain lobes in dementia detection is essential for improving diagnostic accuracy.
Research Source
This paper was co-authored by Siuly Siuly, Md. Nurul Ahad Tawhid, Yan Li, Rajendra Acharya, Muhammad Tariq Sadiq, and Hua Wang. The authors are affiliated with institutions such as Victoria University, University of Southern Queensland, University of Dhaka, and University of Essex. The study was published in 2025 in the journal Cognitive Computation, titled “Investigating Brain Lobe Biomarkers to Enhance Dementia Detection Using EEG Data.”
Research Process
1. Data Preprocessing
The study used the publicly available EEG dataset OpenNeuro ds004504, which includes EEG recordings from 88 participants, divided into AD (36), FTD (23), and Healthy Control (HC, 29) groups. EEG signals were first filtered using a Butterworth band-pass filter (0.5-45 Hz) to remove noise, followed by Automatic Artifact Rejection (ASR) and Independent Component Analysis (ICA) to further eliminate eye and jaw artifacts. The signals were then resampled to 256 Hz and segmented into 3-second time frames to improve computational efficiency while retaining critical information.
2. Brain Lobe Grouping
EEG channels were grouped based on their corresponding brain lobes, including Frontal, Central, Temporal, Parietal, and Occipital lobes. EEG signals from each lobe were analyzed separately to investigate the role of different brain lobes in dementia detection.
3. Spectrogram Generation
Short-Time Fourier Transform (STFT) was used to convert EEG signals into spectrograms. Spectrograms provide a time-frequency representation of brain activity, capturing changes in brain wave patterns associated with cognitive decline.
4. Convolutional Neural Network (CNN) Classification
The generated spectrograms were input into a deep learning model based on a Convolutional Neural Network (CNN) architecture, which includes four convolutional layers, three Dropout layers, one fully connected layer, and a classification layer. The CNN model automatically extracts and learns features from the spectrograms, enabling the classification of dementia.
5. Model Evaluation
The study employed 10-fold Cross-Validation to evaluate the model’s performance, calculating metrics such as Sensitivity, Specificity, Precision, Accuracy, F1 Score, and False Positive Rate. Additionally, the Grad-CAM method was used to enhance the interpretability of the results, providing meaningful visual insights.
Key Findings
1. Brain Lobe Analysis
The study found that the Parietal lobe exhibited the most significant changes in AD and FTD detection. In the AD vs. HC classification task, the Parietal lobe achieved an accuracy of 92.25%, while in the FTD vs. HC classification task, it reached an accuracy of 95.72%. The Temporal and Frontal lobes also showed high classification performance, while the Central lobe performed poorly.
2. Full Brain Region Analysis
When using EEG signals from the full brain region for classification, the accuracy for AD vs. HC was 95.59%, and for FTD vs. HC, it was 93.14%. This indicates that while the Parietal lobe plays a crucial role in dementia detection, combining EEG signals from the full brain region can further improve classification performance.
3. Grad-CAM Visualization
The Grad-CAM method revealed the most influential regions in the spectrograms for classification decisions. Spectrograms from the Parietal lobe and full brain region showed significant heatmap areas, indicating that these regions provide critical information for dementia detection. In contrast, the Central lobe’s spectrograms had fewer heatmap areas, consistent with its lower classification performance.
Conclusion and Significance
This study developed a novel framework combining STFT and CNN to identify key brain lobe biomarkers for dementia detection. The results demonstrate that the Parietal lobe plays a significant role in AD and FTD detection, and combining EEG signals from the full brain region can further enhance diagnostic accuracy. This research provides a non-invasive, cost-effective tool for early dementia detection, with significant clinical application value.
Research Highlights
- Innovative Framework: First to combine STFT with CNN for EEG signal analysis in dementia detection.
- Brain Lobe-Specific Analysis: First systematic study of the role of different brain lobes in dementia detection, identifying the Parietal lobe as the most critical biomarker.
- High-Performance Classification: Achieved accuracies of 95.59% and 95.72% in AD and FTD classification tasks, significantly outperforming existing methods.
- Enhanced Interpretability: Used the Grad-CAM method to provide visual explanations of model decisions, increasing the credibility of the results.
Additional Valuable Information
The study also compared the performance of existing methods on the OpenNeuro ds004504 dataset, further validating the superiority of the proposed framework. Future research can continue to explore the application of EEG in dementia detection and monitoring, promoting the adoption of this technology in clinical practice.
Through this study, we have deepened our understanding of the pathological mechanisms of dementia and provided new insights for developing more effective diagnostic tools. We hope this achievement will lead to earlier and more accurate diagnoses for dementia patients worldwide, ultimately improving their quality of life.