Finding Essential Parts of the Brain in rs-fMRI Can Improve ADHD Diagnosis Using Deep Learning

Abstract

Although attention-deficit/hyperactivity disorder (ADHD) is a common psychiatric disorder, it is difficult to develop an accurate diagnostic method. Recently, studies to classify ADHD using resting-state functional magnetic resonance (rs-fMRI) imaging data have been conducted with the development of computing resources and machine learning techniques. Most of them use the entire brain’s regions when training the models. As opposed to the common approach, we conducted a study to classify ADHD by selecting essential areas for using a deep learning model. The experiment used rs-fMRI data from the ADHD-200 global competition. To obtain an integrated result from the multiple sites, each region of the brain is evaluated using ’leave- one-site-out’ cross-validation. As a result, when we only used 15 important regions of interest (ROIs) for training, 70.6% accuracy was obtained, significantly exceeding the existing results of 68.6% from all ROIs. Additionally, to explore the new structure based on SCCNN-RNN, we performed the same experiment with three modified models: (1) separate channel CNN - RNN with attention (ASCRNN), (2) separate channel dilate CNN - RNN with attention (ASDRNN), (3) separate channel CNN - slicing RNN with attention (ASSRNN). The ASSRNN model provides a high accuracy of 70.46% when trained with only 13 important ROIs. These results show that using deep learning to identify the crucial parts of the brain in diagnosing ADHD yields better results than using every area.

Keywords

• ADHD
• deep learning
• rs-fMRI
• AAL116
• ROI