Towards interpretable machine learning models for diagnosis aid: A case study on attention deficit/hyperactivity...

Area of Science:

• Neuroscience
• Computational Psychiatry
• Developmental Psychology

Background:

• Attention Deficit/Hyperactivity Disorder (ADHD) is a neurodevelopmental disorder impacting academic, psychological, and relational well-being.
• Current ADHD diagnosis relies on clinical assessments and DSM-V criteria, with ongoing research seeking more objective methods.
• Machine Learning (ML) offers potential for predictive diagnosis using phenotypic and neuroimaging data.

Purpose of the Study:

• To develop an ML methodology for ADHD diagnosis that balances predictive performance with model interpretability.
• To identify key neurophysiological indicators relevant for ADHD diagnosis.
• To create a readable decision tree model for aiding medical diagnosis.

Main Methods:

• Applied an ML methodology focusing on explanatory power to a subset of the ADHD-200 dataset.
• Utilized decision trees, known for their interpretability, to analyze phenotypic and neuroimaging data.
• Evaluated model performance against existing literature and focused on the clarity of diagnostic explanations.

Main Results:

• The developed ML model identified the limbic system as relevant for ADHD diagnosis.
• The decision tree model provided meaningful explanations for its predictions.
• The model achieved favorable performance compared to recent studies in ADHD prediction.

Conclusions:

• An interpretable ML approach can effectively aid in ADHD diagnosis.
• The limbic system plays a significant role in the neurophysiology of ADHD.
• This methodology offers a promising balance between predictive accuracy and clinical interpretability for diagnostic support systems.