Classification of ADHD with fMRI data and multi-objective optimization.

Area of Science:

• Neuroimaging
• Machine Learning
• Medical Diagnostics

Background:

• Dataset imbalance is a significant challenge in neuroimaging classification.
• Imbalanced datasets can lead to classifier performance degradation, favoring majority classes.
• Accurate classification of neurodevelopmental disorders like ADHD is crucial.

Purpose of the Study:

• To develop an effective classification scheme for imbalanced neuroimaging datasets.
• To address the challenge of classifying attention deficit hyperactivity disorder (ADHD) using resting-state functional magnetic resonance imaging (fMRI).
• To improve classifier performance in the presence of significant class imbalance.

Main Methods:

• A multi-objective classification scheme utilizing Support Vector Machines (SVM) was proposed.
• The SVM model explicitly and separately handles positive and negative empirical errors.
• An interactive multi-objective method incorporating decision-maker preferences was employed to obtain Pareto optimal classifiers.

Main Results:

• The proposed multi-objective SVM scheme was evaluated on five ADHD-200 consortium datasets.
• Numerical results demonstrated superior performance compared to traditional classification methods.
• The approach effectively mitigates issues arising from imbalanced neuroimaging data.

Conclusions:

• The multi-objective classification scheme offers a robust solution for imbalanced datasets without requiring hyper-parameter tuning.
• This algorithmic-level approach effectively addresses data imbalance.
• The method has potential applications beyond ADHD diagnosis, including Alzheimer's and Autism spectrum disorder detection.