Application of artificial intelligence in attention-deficit hyperactivity disorder deteaction and response to treatment: A systematic review

Abstract

Attention-deficit hyperactivity disorder (ADHD) is a widespread neurodevelopmental condition that significantly impacts many children. This review systematically explores the application of artificial intelligence (AI), particularly conventional machine learning (ML) and deep learning (DL), in diagnosing ADHD and predicting treatment responses from clinical data (demographics, questionnaires, cognitive tests, and biological variables) and neuroimaging modalities including electroencephalogram (EEG), functional magnetic resonance imaging (fMRI), structural MRI (sMRI), magnetoencephalogram (MEG), diffusion tensor imaging (DTI), single photon emission computed tomography (SPECT), functional near-infrared spectroscopy (fNIRS), and positron emission tomography (PET). We searched papers published until October 2024 via Scopus, Web of Science (WOS), and PubMed and reviewed 147 studies on ADHD diagnosis and treatment response prediction. This study's primary contribution is the broad integration of its studies on both diagnosis and response prediction of pharmacological and non-pharmacological treatments in ADHD using AI methods, relying on feature extraction/selection methods and biomarkers rather than focusing on just one part. Furthermore, it is the first review to specifically evaluate the application of AI in predicting response to ADHD treatment. To have a better view, we investigated the data in clinical/demographic categories, neuroimaging techniques, and using neuroimaging/clinical biomarkers for ADHD diagnosis and predicting response to treatment. Our findings emphasize the widespread application of AI in ADHD detection and show promising results with EEG signals (accuracy up to 99.95 %) and MRI modalities (accuracy up to 92.8 % with a combination of sMRI and fMRI) data while highlighting the limited application of AI in predicting treatment responses. Support vector machines (SVMs) and convolutional neural networks (CNNs) methods have been used more among AI methods. Also, extracting and selecting features from EEG signals is more prevalent than other neuroimaging techniques, and functional connectivity biomarkers in fMRI showed superior performance. Future research should aim to develop integrated AI models that can accurately diagnose and predict personalized treatment responses. © 2025 Elsevier B.V., All rights reserved.