Hybrid Deep Learning Classification Model for Attention-Deficit-Hyperactivity Disorder using functional Magnetic Resonance Imaging

A child with a neurodevelopmental disorder, such as attention deficit hyperactivity disorder (ADHD), faces many challenges in school, like not paying attention when the teacher is taking class, poor performance in academics, and difficulty following classroom behaviour. The functional connectivity of the brain is an essential feature to be investigated to identify such disorders. Developing a computer-aided diagnosis method using a deep learning algorithm to identify ADHD children is the aim of the research. In this study, a hybrid 3D convolutional neural network and bidirectional long short-term memory (3D CNN-BiLSTM) classification model is proposed to classify ADHD and typically developing conditions using resting-state functional magnetic resonance imaging of the brain. The result shows that the accuracy of the proposed hybrid deep learning model gives a maximum accuracy of 75.4% on validation. This indicates that the proposed 3D CNN-BiLSTM classification model for ADHD outperforms the recent existing algorithm by approximately 3% in terms of accuracy using only the functional magnetic resonance imaging dataset obtained from the ADHD-200 repository. Medical professionals may be able to diagnose ADHD in children efficiently by combining the proposed classification model with the present diagnosis based on the child's behaviour.