Attention Deficit Hyperactivity Disorder identification: FMRI data analyzed with CNN and seed-based approach

Abstract

Attention Deficit Hyperactivity Disorder (ADHD) is a prevalent mental disorder affecting both adults and children, frequently leading to academic difficulties. This study aims to improve the diagnosis of ADHD in children by using resting-state Functional Magnetic Resonance Imaging (fMRI) data. The method use seed coherence to identify functional connections between specific seed areas and all brain voxels, focusing on Default Mode Network (DMN) regions pertinent to the diagnosis of ADHD. Convolutional Neural Networks (CNNs) are utilized in classification tasks because of their capacity to learn intricate spatial hierarchies. The research utilizes fMRI scans from the ADHD 200 - Global Competitive dataset, comprising 776 subjects from three prominent data centers. The methodology entails data preparation, feature extraction via seed-based correlation, and classification with Convolutional Neural Networks (CNNs). Three classifiers were assessed: a Neural Network (Keras Sequential Model), a Support Vector Machine (SVM), and a Random Forest Classifier. The optimal outcome was achieved by the neural network, which harmonized precision, recall, and F1 scores, attaining an accuracy of 97 %. The SVM demonstrated considerable accuracy at 83 %, however the Random Forest Classifier exhibited a mere 50 % accuracy, underscoring the necessity for enhancement. These results underscore the merits and shortcomings of each classifier and offer suggestions for enhancement. The paper highlights the significance of Neural Networks for attaining precise and equitable forecasts, proposes enhancements for the Support Vector Machine, and stresses the imperative of optimizing the Random Forest Classifier. This study enhances ADHD diagnosis by methodically employing neuroimaging techniques and assessing several classifiers, leading to a reliable diagnostic system.