Classification of ADHD and its Sub-Types using Machine learning Models

Abstract

ADHD, a complex neurodevelopment disorder, exhibits diverse manifestations across individuals, making its diagnosis challenging. Convolution Neural Networks (CNNs) offer a crucial edge in ADHD detection due to their proficiency in processing and analyzing intricate patterns within neuroimaging data, such as fMRI scans. Given the multi-dimensional nature of ADHD, CNNs excel in capturing subtle neurological variations that might escape conventional analysis, thereby providing a more nuanced and accurate approach to discerning the complex patterns associated with this condition. Their ability to automatically extract hierarchical features from imaging data makes CNNs indispensable in unraveling the intricate neurobiological markers crucial for ADHD identification and classification. Convolution Neural Networks (CNNs) have revolutionized medical imaging, offering unprecedented potential in the diagnosis and categorization of complex conditions like Attention-Deficit/Hyperactivity Disorder (ADHD) and its subtypes. In the context of ADHD, CNN and VGG-16 models are trained on functional MRI (fMRI) scans to identify unique neural patterns associated with the disorder. This paper introduces machine learning (ML) models designed for ADHD classification, assessed using various statistical metrics such as accuracy, F1-score, precision, and recall through 5-fold cross-validation. The results from the study showcase the capability of the Identification of Lung Cancer (IOLC) model in identifying lung cancer, gauged through accuracy, precision, recall, F-Measure, and error rate metrics. The model demonstrates 91.68% accuracy, 89.8% precision, 89.3% recall and 89.2% F-Measure. The ROC curve confirms the effectiveness of the proposed model as a classifier for ADHD types, and comparative results against VGG-16 demonstrate the proposed model's superior performance, albeit moderately.