Computer aided diagnosis system using deep convolutional neural networks for ADHD subtypes

Background: Attention deficit hyperactivity disorder (ADHD) is a ubiquitous neurodevelopmental disorder affecting many children. Therefore, automated diagnosis of ADHD can be of tremendous value. Unfortunately, unlike many other applications, the use of deep learning algorithms for automatic detection of ADHD is still limited. Method: In this paper, we proposed a novel computer aided diagnosis system based on deep learning approach to classify the EEG signal of Healthy children (Control) from ADHD children with two subtypes of Combined ADHD (ADHD-C) and Inattentive ADHD (ADHD-I). Inspired by the classical approaches, we proposed a deep convolutional neural network that is capable of extracting both spatial and frequency band features from the raw electroencephalograph (EEG) signal and then performing the classification. Result: We achieved the highest classification accuracy with the combination of beta(1), beta(2), and gamma bands. Accuracy Recall, Precision, and Kappa values were %99.46, %99.45, %99.48, and 0.99, respectively. After investigating the spatial channels, we observed that electrodes in the Posterior side had the most contribution. Conclusions: To the best of our knowledge, all previous multiclass studies were based on fMRI and MRI imaging. Therefore, the presented research is novel in terms of using a deep neural network architecture and EEG signal for multiclass classification of ADHD and healthy children with high accuracy.

Automated summary

This study presented a novel computer aided diagnosis system based on deep learning approach for accurate classification of ADHD children and healthy children. By developing a deep convolutional neural network, the system achieved automatic diagnosis of EEG signals, with specific frequency band combinations resulting in the highest classification accuracy.