Diagnostic Classification of ADHD Versus Control: Support Vector Machine Classification Using Brief...

Area of Science:

• Neuroscience
• Developmental Psychology
• Computational Psychiatry

Background:

• Current ADHD diagnosis relies on subjective clinical interpretation, leading to variable reliability and accuracy.
• Brief neuropsychological measures are essential for assessing attention and concentration deficits in children.
• Machine learning offers potential for objective and precise diagnostic classification.

Purpose of the Study:

• To evaluate the efficacy of machine learning algorithms in classifying children with ADHD-Combined presentation (ADHD-C).
• To identify key neuropsychological features for accurate ADHD-C prediction.
• To assess the diagnostic utility of data-driven behavioral algorithms.

Main Methods:

• Utilized a forward feature selection method for Support Vector Machine (SVM) classification.
• Developed a decision tree model for rule-based classification.
• Included children with ADHD-C and typically developing controls, employing clinical interviews, neuropsychological tests (d2 Test of Attention), and parent questionnaires.

Main Results:

• The SVM model achieved 100% accuracy in classifying children with and without ADHD-C.
• Decision tree algorithms demonstrated 100% sensitivity and specificity in identifying ADHD-C.
• Brief neuropsychological data proved highly informative for diagnostic classification.

Conclusions:

• Machine learning models can achieve highly accurate, individual-level diagnostic classification for ADHD-C.
• Data-driven algorithms utilizing brief neuropsychological measures present a promising, efficient, and accurate diagnostic tool.
• This approach enhances diagnostic objectivity and reliability in pediatric ADHD assessment.