Deep Learning-Based ADHD Classification Using MRI Images: A Comparative Study of VGG19 and EfficientNetB0 with Otsu's Thresholding

Attention Deficit Hyperactivity Disorder (ADHD) is a common neurodevelopmental disease that affects attention, increases impulsivity, and causes hyperactivity in children and adolescents. Traditional diagnosis is mostly based on subjective psychological assessments, which may lack objectivity and consistency. This study looks at how well deep learning methods can sort ADHD cases using brain scan (MRI) data. The VGG19 and EfficientNetB0 models, which are types of neural networks, were used for this task. Otsu's thresholding technique was used as a preprocessing step to improve picture segmentation and classification performance. The VGG19 model gave accuracy of 88.07 % and showed AUC of 0.95, whereas EfficientNetB0 obtained a similar AUC of 0.95 and a little better accuracy of $\mathbf{8 8. 1 5 \%}$. These results show that both models worked well in telling apart children with normal growth and those with ADHD-Hyperactive type. The steady AUC values prove that the models gave stable and reliable results for both groups.