EEG-Based Dementia Classification Using CS-EMD Synchrony Features and Quantum Machine Learning

Dementia is a condition that affects a very large population, and an effective clinical diagnosis can lead to early intervention to delay progressive cognitive decline. A relatively low-cost, noninvasive, and highly accessible biomarker measurement, electroencephalography (EEG), may be used for this purpose. Many EEG processing and classification methods, which are based on classical techniques, have been explored by researchers over the years. On the other hand, quantum machine learning (QML), which combines the concepts of quantum computing and machine learning, has recently gained much attention with the rising anticipation that this emerging research field may provide significant advantages over classical methods. This paper presents a healthy-dementia classification framework using synchrony features of CS-EMD decomposed EEGs and QML. In this study, three QML models, a variable quantum classifier (VQC), a quantum support vector machine (QSVM), and a quantum neural network (QNN) were explored. Their performance was evaluated through classification experiments using real EEG data. The performances of these QML methods, along with two popular classical classifiers (support vector machine and neural network), were compared. In our experiments, the quantum neural network outperformed all the other tested classifiers, reaching 96.58% accuracy, 97.11% specificity, 96.83% precision, 96.34% sensitivity, 96.28% F1 score, 97.69% AUC, an MCC of 0.9320, and a Kappa of 0.9268. Compared to the runner-up, a classical neural network, the QNN classifier shows an improvement by 8.92% in accuracy, 12.03% in specificity, 13.49% in precision, 5.57% in sensitivity, 10.29% in F1, 3.75% in AUC, 17.83% in MCC, and 18.90% in Kappa. The proposed framework, utilizing a quantum neural network, demonstrates effective classification of healthy individuals and dementia patients with high accuracy, showing good potential as a screening tool for dementia. Similar techniques may also be adapted for other biosignal analyses.