Machine learning analysis of posturography in panic disorder: a pilot study for objective physiological biomarker...

Area of Science:

• Neuroscience
• Biomedical Engineering
• Clinical Psychology

Background:

• Panic disorder (PD) is associated with subtle postural control deficits.
• Traditional statistical methods inadequately capture these abnormalities.
• Machine learning (ML) offers potential for enhanced detection of PD-related postural patterns.

Purpose of the Study:

• To evaluate static postural control in PD patients.
• To determine if ML analysis of stabilometric data can improve differentiation from healthy controls.

Main Methods:

• A cross-sectional case-control study involving 12 PD patients and 21 healthy volunteers.
• Stabilometry was performed under five sensory conditions.
• ML models (Logistic Regression, Linear Discriminant Analysis, etc.) were trained and validated using stratified, subject-grouped cross-validation.

Main Results:

• Logistic Regression achieved 93.8% accuracy and an AUC of 96% in differentiating groups.
• Linear Discriminant Analysis showed the highest specificity (91.7%).
• PD patients exhibited significantly reduced mediolateral sway compared to controls.

Conclusions:

• ML applied to posturography can identify physiological markers for panic disorder.
• ML analysis of stabilometric data improves classification accuracy over traditional methods.
• Static posturography with ML may be superior to current clinical screening tools for PD.