Use transfer learning to promote identification ADHD children with EEG recordings

Electroencephalography (EEG) is widely used in Attention-deficit/hyperactivity disorder (ADHD) and it's a neurobehavioral disorder in children. Convolutional neural network (CNN) is a mainstreamed deep learning algorithm, which could be used to identify ADHD children. In this study, we validated the feasibility of applying CNN on ADHD identification problem by using transfer learning. A total of 50 ADHD children and 58 normal children were recruited, from whom a 32-channel resting-state EEG was obtained. Phase lag index was utilized to construct brain networks, which were elaborately designed as input to CNN models. We pre-trained an image-classification model and transferred specific layers to another CNN model for identifying ADHD children. Then we Compared with the model trained from scratch, the model applying transfer learning achieved an enhanced performance and robustness with an accuracy of 94.39%, sensitivity of 97.83% and specificity of 91.80%. We demonstrated the effectiveness of transfer learning and the deep learning techniques for identification problem of ADHD children.