Graph Signal Processing-based EEG Analysis with Quantum Machine Learning for Early Detection of Alzheimer's Disease

The timely intervention in Alzheimer disease (AD) requires early detection, which is difficult to achieve in conventional machine and deep learning models because the brain networks have a complex functional connectivity and nonlinear dynamics. To mitigate this shortcoming, this paper advances a topology-conscious framework of spectral graph signal processing (GSP) and quantum machine learning (QML) to EEG-based early AD detection. Phase-locking value (PLV) was used to model multichannel EEG data of 162 (AD, mild cognitive impairment (MCI), and healthy controls) as functional networks of connectivity. Features of spectral graphs that are obtained through graph fourier transform can maintain the topology of a network and are encoded into a variational quantum classifier to improve nonlinear separability. The proposed method has accuracy of 94.6, sensitivity of 93.2, specificity of 95.8 and an AUC of 0.968, which is better than the conventional models. The analysis of ablation proves the complementary roles of the GSP and QML. The framework shows that it has a high potential to be used as a scalable and efficient solution to next-generation EEG-based neurodiagnostic.