Assisted diagnosis of Attention-Deficit Hyperactivity Disorder through EEG bandpower clustering with self-organizing maps

The electroencephalogram is an attractive clinical tool given its non-invasive nature, its ability to reflect real-time changes in local cortical activity, and the load of objective bioelectrical measurements that can be derived from it. For decades, the electroencephalogram has been successfully used for diagnosing epilepsy and schizophrenia, among other brain disorders. This paper focuses in the design and implementation of a computer-aided diagnostic tool for establishing the likelihood of presence of Attention-Deficit Hyperactivity Disorder in children, out of routine electroencephalographic recordings obtained during a specific visual stimulation protocol. Classical bandpower features from multiple differential recordings are computed and used as features in a classifier built from a cooperative ensemble of labeled self-organizing maps. Classification accuracy of the proposed system is 0,7 ± 0,11, as estimated from unseen data, a result that points to the idea that such a quantitative diagnostic aid could adequately support the diagnostic task of a clinical expert.