Abstract
High-throughput analysis of EEG data has significantly contributed to understanding neural dynamics in Alzheimer's disease diagnosis. However, the complexity and high dimensionality of EEG signals pose challenges for traditional classification methods, which often fail to capture intricate patterns. To address this, we propose a hybrid approach integrating Topological Deep Learning (TDL) with machine learning models-including Support Vector Machines (SVM), Random Forest (RF), Neural Networks (NN), and Logistic Regression (LR)-for Alzheimer's disease classification. By leveraging TDL, our method extracts topological and neural features from EEG data, enhancing the identification of disease-specific patterns that conventional models may overlook. The dataset consists of EEG recordings from 88 individuals, categorized into AD patients, FTD patients, and CN, providing a robust foundation for model evaluation. Our findings demonstrate that NN augmented by TDL achieve the highest classification accuracy, reaching up to 90% in distinguishing AD, FTD, and CN cases. These results highlight the potential of TDL-enhanced deep learning models in clinical applications, offering a more accurate and detailed tool for Alzheimer's disease diagnosis and differentiation from other neurodegenerative conditions. This work is presented as a proof-of-concept demonstrating that persistence-based topological descriptors can enhance EEG classification; multicenter validation on larger, diverse cohorts will be required to confirm generalizability.