Comparison of Machine Learning Methods for ADHD Medical Imaging Diagnosis

ADHD is a chronic disorder that significantly affects patients’ occupational development, interpersonal relationships, and quality of life. In this study, based on the ADHD dataset and combining feature extraction with visualization analysis, we evaluated the classification performance of seven models: RF, XGB, SVM, LR, GNB, 3D_CNN and 3D_ResNet. Results of 10-fold cross-validation showed that 3D_ResNet and SVM achieved the highest accuracy, at 70% and 63% respectively. 3D_ResNet incurred more expensive computational costs, whereas SVM demonstrated particular advantage in reducing missed diagnoses to meet clinical requirements, recall is 87.67%. The RF, XGB, and GNB models exhibited balanced overall performance, while LR and 3D_CNN showed less stability. Through comparative analysis of these models, this study provides a reference for optimizing auxiliary diagnostic tools for ADHD.