EEG/ERP-based biomarker/neuroalgorithms in adults with ADHD: Development, reliability, and application in clinical...

Area of Science:

• Neuroscience
• Computational Psychiatry
• Medical Informatics

Background:

• Attention-deficit/hyperactivity disorder (ADHD) diagnosis relies on subjective assessments.
• Electrophysiological measures like EEG and ERPs offer objective biomarkers.
• Machine learning (ML) can analyze complex neurophysiological data for diagnostic support.

Purpose of the Study:

• To evaluate the efficacy of neuroalgorithms, a machine learning approach, in diagnosing ADHD.
• To assess the performance of ML models using electrophysiological data from a large adult cohort.
• To determine if ML-based neuroalgorithms can improve diagnostic objectivity and reliability.

Main Methods:

• Collected spontaneous electroencephalographic (EEG) and event-related potential (ERP) data from 181 adults with ADHD and 147 healthy controls over two years.
• Utilized spectral power, ERP amplitude, and latency measures as input features for a semi-automatic machine learning framework.
• Employed logistic regression models to classify ADHD patients versus controls.

Main Results:

• The ML models achieved classification sensitivity ranging from 75% to 83% and specificity from 71% to 77%.
• Repeated measurements showed sustained high sensitivity (72%–76%) with a slight decrease in specificity over time (64%–67%).
• Demonstrated the potential of neuroalgorithms to distinguish between ADHD patients and healthy individuals.

Conclusions:

• Neuroalgorithms show significant potential to enhance ADHD diagnosis by reducing subjectivity and increasing reliability.
• Clinical implementation requires infrastructure for tracking neural networks and neuropathophysiological expertise.
• This approach links diagnostic criteria to underlying neuropathophysiology, improving diagnostic accuracy.