Classification of ADHD Using fNIRS Signals Based on Functional Connectivity and Interval Features

Attention deficit hyperactivity disorder (ADHD) is a common neurodevelopmental disorder among children. Traditional ADHD diagnosis usually lack physiological measurements. To address this issue, we introduce functional near-infrared spectroscopy (fNIRS), which can capture oxygenation and hemodynamic changes in human cerebral cortex. fNIRS signals can be view as time series. Most of the time series classification algorithms focus on time domain features. In this paper, we propose a half fold windowing (HFW) method on fNIRS signals to get sub-series and extract interval features from them. We also adopt a novel way of extracting inter-channel features through functional connectivity (FC) by correlation, coherence and wavelet coherence. It utilizes inter-channel and frequency domain information. Furthermore, we perform a classification on the ADHD fNIRS dataset we collected based on our feature extraction methods with a feature union. The experiment shows that using HFW-based interval features combined with FC features for fNIRS signal classification achieves a better accuracy compared with other methods.