Voice-Activated Machine Learning for Screening Bipolar and Major Depressive Disorder in Youth

Overview

This study demonstrates that voice features can reliably differentiate bipolar disorder (BD), major depressive disorder (MDD), and healthy controls (HC) in children and adolescents using machine learning. The approach offers a non-invasive, efficient, and accurate diagnostic tool that may reduce misdiagnosis and diagnostic delays in youth psychiatric disorders.

Background

Bipolar disorder and major depressive disorder are prevalent psychiatric conditions affecting millions worldwide, including children and adolescents. Diagnosing BD is particularly challenging due to overlapping depressive symptoms and frequent misdiagnosis as MDD, often leading to inappropriate treatment and delayed care. Objective biomarkers for these disorders remain elusive, but voice analysis has emerged as a promising, non-intrusive method due to its sensitivity to physiological and neurobiological changes. Prior research has focused mainly on adults, highlighting the need for youth-specific diagnostic tools.

Data Highlights

Key Findings

• Voice features show stable, quantifiable differences among children and adolescents with BD, MDD, and HC.
• Machine learning classifiers trained on selected core voice features achieved high classification accuracy with low computational cost.
• The study included 150 subjects (50 per group) with rigorous DSM-5-based diagnoses confirmed by expert consensus.
• An external validation dataset (50 subjects) confirmed the robustness of the voice-based diagnostic model.
• Voice analysis offers a non-invasive, easily acquired biomarker alternative to neuroimaging or EEG for psychiatric disorder screening in youth.

Clinical Implications

Voice-based machine learning screening can facilitate earlier and more accurate differentiation between BD and MDD in children and adolescents, potentially reducing misdiagnosis and inappropriate treatment. This approach is practical for clinical settings due to its non-intrusive nature and low computational requirements, supporting broader implementation in psychiatric evaluation workflows.

Conclusion

Voice feature analysis combined with machine learning provides a reliable, simple, and efficient method for screening bipolar disorder and major depressive disorder in youth. This technique holds promise for improving early diagnosis and treatment outcomes in pediatric psychiatric care.

References

1. Global Burden of Disease Study 2019 -- Prevalence of BD and MDD
2. DSM-5 Diagnostic Criteria -- Bipolar and Major Depressive Disorders
3. Hamilton Depression Rating Scale (HAMD) and Young Mania Rating Scale (YMRS) -- Clinical Assessment Tools
4. Previous Voice-Based MDD Classification in Youth -- Prior Work by Authors