A Big-Data-Analytics Framework for Supporting Classification of ADHD and Healthy Children via Principal Component Analysis of EEG Sleep Spindles Power Spectra

摘要

Attention Deficit Hyperactivity Disorder (ADHD) diagnosis is essentially clinical and research of biomarkers represents a current great challenge. The interest in sleep spindle has been increased after the description of their role in cognitive functions and of their involvement in neurodevelopmental disorders. We aimed to investigate this peculiar aspect of sleep through EEG spectral analysis of three different spindle epochs (ante, spindle, post), in order to provide more and detailed information on sleep brain functioning in ADHD. These features can be analyzed via well-known big data analytics methods. In our case, they were evaluated by using classification methods to support ADHD diagnosis. We combined ADHD’s related PSD features (i.e. theta, beta and sigma bands) with principal component analysis (PCA) for data dimensional reduction, and Linear Supported Vector Machine (Linear-SVM) as classification algorithm. In all bands and epochs, power values in Control group were higher than in ADHD children, although not statistically significant in all cases. Significant differences between ADHD and Control group were not detected for spindle epoch, while for ante and post epochs spectral power differed significantly in theta, beta and sigma bands. Results highlighted the possibility of using our new approach as a possible hallmark for ADHD. Indeed the analysis of PSD parameters combined with PCA and Linear-SVM classification resulted in a highly (94.1%) accurate discrimination between the two groups. The novelty of the approach is PSD analysis of different sleep spindles epochs combined with principal component analysis and Linear Supported Vector Machine classification. This study demonstrated the importance of analyzing sleep microstructures in ADHD. Encouraging results supports the potentiality of using EEG measures with specific methodologies we applied and should be confirmed in a large clinical study.