EEG analysis and classification based on cardinal spline empirical mode decomposition and synchrony features

Dementia is a major cause of disability and dependency among older adults. Diagnosis is most effective at an early stage of the disease, as patients can start early treatment to delay progressive cognitive decline. While other diagnostic methods for dementia are available, electroencephalography (EEG) is noninvasive, more accessible, and less complicated than other biomarker measurements. Moreover, it may be orders of magnitude less expensive, thereby offering the possibility of low-cost mass screening. This paper presents a novel digital signal processing method called cardinal spline empirical mode decomposition (CS-EMD) for EEG processing. This new method uses a different signal envelope interpolation algorithm to separate EEG signals into constituent components, called intrinsic mode functions (IMFs), with better signal decomposition properties than classical empirical mode decomposition (EMD). The IMFs obtained from the new method are then used to compute longitudinal and transversal synchrony measures, which are explored as features for healthy and dementia classification using a support vector machine (SVM). The performance of the proposed method is studied on a publicly available EEG dataset. The results show that using multiple synchrony measures of both longitudinal and transversal EEG channels on five IMFs produces the best classification result of 90% accuracy, 96.67% specificity, 83.33% sensitivity, and 96.15% precision, outperforming the classical EMD method and various other approaches. This new, data-driven CS-EMD method shows good potential as a dementia screening tool. CS-EMD may also be applied in processing other nonlinear and nonstationary biosignals.