An improved model for the diagnosis of attention deficit/hyperactivity disorder (ADHD) using resting-state fMRI data

While attention-deficit/hyperactivity disorder (ADHD) stands as a prevalent psychiatric condition, creating a precise diagnostic approach has proven challenging. In recent years, advancements in machine and deep learning techniques have facilitated studies aiming to classify ADHD. This research presents a robust deep learning model for the accurate diagnosis of Attention Deficit/Hyperactivity Disorder (ADHD) leveraging Resting-state functional Magnetic Resonance Imaging (rs-fMRI) data. The study utilizes the ADHD-200 dataset to capture intrinsic brain activity patterns. To identify relevant regions of interest (ROIs) from the fMRI data, we employ Nilearn's CanICA, a powerful group-level Independent Component Analysis method. The proposed model employs a Gated Recurrent Unit (GRU) architecture, a recurrent neural network, to capture temporal dependencies within the resting-state fMRI time series. The GRU model demonstrates exceptional performance, achieving an average accuracy of 87.9% in ADHD diagnosis. The integration of advanced deep learning techniques, coupled with the informative ROIs identified by CanICA, not only enhances the interpretability of the model but also contributes to its high diagnostic accuracy.