Machine learning algorithm-based estimation model for the severity of depression assessed using Montgomery-Asberg...

Area of Science:

• Psychiatry and Artificial Intelligence
• Computational Psychiatry
• Mental Health Technology

Background:

• Depressive disorder assessment traditionally relies on trained professionals using rating scales.
• Ensuring consistent data collection with these scales presents a significant challenge.
• AI-driven tools offer a potential solution for objective and reliable depression severity evaluation.

Purpose of the Study:

• To evaluate the rater and estimation-system reliability of the AI-MADRS (Artificial Intelligence-Montgomery-Asberg Depression Rating Scale) system.
• To compare the AI-MADRS estimation system's performance against consensus evaluations by trained psychiatrists.
• To assess the reliability of two AI-MADRS models: max estimation and average estimation.

Main Methods:

• Patients responded to AI-MADRS structured interview questions via oral prompts.
• Severity scores were generated by two AI-MADRS models (max and average estimation).
• AI-generated scores were compared with scores from consensus evaluations by trained psychiatrists.

Main Results:

• Analysis included 51 interviews from 30 patients.
• The average estimation model achieved a Pearson's correlation of 0.86 with psychiatrists' scores.
• The average estimation model demonstrated ANOVA ICC reliability of 0.75 with psychiatrists' evaluations.

Conclusions:

• The AI-MADRS average estimation model exhibits substantially acceptable reliability with trained psychiatrists.
• Further improvements are anticipated with larger datasets and refined AI-MADRS interviews.
• AI technologies show promise in modernizing and potentially revolutionizing depression assessment.