Explainable AI for Bipolar Disorder Diagnosis Using Hjorth Parameters

Background: Despite the prevalence and severity of bipolar disorder (BD), current diagnostic approaches remain largely subjective. This study presents an automatic diagnostic framework using electroencephalography (EEG)-derived Hjorth parameters (activity, mobility, and complexity), aiming to establish objective neurophysiological markers for BD detection and provide insights into its underlying neural mechanisms. Methods: Using resting-state eyes-closed EEG data collected from 20 BD patients and 20 healthy controls (HCs), we developed a novel diagnostic approach based on Hjorth parameters extracted across multiple frequency bands. We employed a rigorous leave-one-subject-out cross-validation strategy to ensure robust, subject-independent assessment, combined with explainable artificial intelligence (XAI) to identify the most discriminative neural features. Results: Our approach achieved remarkable classification accuracy (92.05%), with the activity Hjorth parameters from beta and gamma frequency bands emerging as the most discriminative features. XAI analysis revealed that anterior brain regions in these higher frequency bands contributed most significantly to BD detection, providing new insights into the neurophysiological markers of BD. Conclusions: This study demonstrates the exceptional diagnostic utility of Hjorth parameters, particularly in higher frequency ranges and anterior brain regions, for BD detection. Our findings not only establish a promising framework for automated BD diagnosis but also offer valuable insights into the neurophysiological basis of bipolar and related disorders. The robust performance and interpretability of our approach suggest its potential as a clinical tool for objective BD diagnosis.