A Machine Learning Approach for Classification and Identification of Important ROIs for ADHD-I

Attention Deficit Hyperactivity Disorder (ADHD) is a condition due to deviated neurodevelopment. For analysis of ADHD, rs-fMRI is one of the forthcoming methods that have shown differences in brain activity for ADHD patients. The brain works in correlation with each of its regions through neural connections. Although previous research has estimated regions of the brain solely responsible for ADHD, work on correlation among brain regions that affect ADHD is still lacking. This research has explored the ADHD-200 dataset and sorted data for quality and comparison between ADHD-Inattentiveness (ADHD-I) and Typically Developing (TD) brains. Retaining literature in this sector have worked with multiclass data for ADHD, that is, classification between ADHD-Combined (ADHD-C) and TD reports. This research puts forth a consistent ensemble model for higher classification accuracy and for finding the most important correlations between brain regions that affect the subtype ADHD-I patients against TD subjects. The model proposed in this research comprises of six classifiers. The classifiers have been run through grid-search method to find the best hyperparameters to be used for the classification of this dataset. It has achieved 91.4% accuracy on 10-fold cross-validation with a Standard Deviation (SD) of 0.038 and 93% accuracy on a prediction set with a precision of 93% and recall of 91%. The model has also detected the top five most important brain region correlations as weighted data points in this classification task. This work puts the spotlight on a subtype of ADHD and achieves a consistently accurate model to classify the subtype and reveal important correlations among Region of Interest (ROI) correlations.