Explainable AI for Bipolar Disorder Diagnosis Using Hjorth Parameters.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Psychiatry

Background:

• Bipolar disorder (BD) diagnosis is subjective, lacking objective neurophysiological markers.
• Current diagnostic methods for BD are limited by their reliance on clinical observation.
• There is a need for objective biomarkers to improve BD detection and understanding.

Purpose of the Study:

• To develop an automated diagnostic framework for bipolar disorder (BD) using electroencephalography (EEG).
• To identify objective neurophysiological markers for BD detection through Hjorth parameters.
• To gain insights into the neural mechanisms underlying BD.

Main Methods:

• Utilized resting-state eyes-closed EEG data from 20 BD patients and 20 healthy controls.
• Extracted Hjorth parameters (activity, mobility, complexity) across multiple frequency bands.
• Employed leave-one-subject-out cross-validation and explainable artificial intelligence (XAI).

Main Results:

• Achieved a classification accuracy of 92.05% for BD detection.
• Identified Hjorth activity parameters in beta and gamma bands as key discriminative features.
• XAI highlighted anterior brain regions in higher frequency bands as significant for BD detection.

Conclusions:

• Hjorth parameters, especially in higher frequencies and anterior regions, show significant diagnostic utility for BD.
• The developed framework offers a promising tool for automated and objective BD diagnosis.
• Findings provide valuable insights into the neurophysiological basis of bipolar disorder.