Facial Expression-Based Evaluation of the Emotion Estimation Software Kokoro Sensor in Healthy Individuals:...

Area of Science:

• Healthcare Technology
• Artificial Intelligence in Medicine
• Affective Computing

Background:

• Artificial intelligence (AI) systems are increasingly utilized in healthcare for objective emotional state assessment.
• AI offers a safe, quick, and user-friendly method for emotional evaluation.
• Evidence supporting AI implementation in healthcare for emotion recognition remains limited.

Purpose of the Study:

• To evaluate the concurrent validity of an emotion recognition AI system based on facial expressions.
• To determine the test-retest reliability of the AI system for emotion recognition.
• To explore the clinical utility of AI-driven facial expression analysis in healthcare.

Main Methods:

• Utilized the Kokoro Sensor, an automated facial expression recognition system.
• Assessed concurrent validity and test-retest reliability using participant and researcher subjective evaluations via a Likert scale.
• Employed the Japanese version of the Profile of Mood States-Short Form to control for mental state influence.

Main Results:

• Concurrent validity showed high agreement for positive expressions (CVI 95-100%) but lower for negative expressions (CVI 70-85%).
• Overall moderate agreement (κ=0.55) was found, with almost perfect agreement for positive vs. neutral expressions (κ=0.83).
• Test-retest reliability was substantial (κ=0.66), with high agreement for positive vs. neutral expressions (κ=0.85) but limited for negative vs. neutral (κ=0.36).

Conclusions:

• The Kokoro Sensor shows potential for identifying positive affect due to acceptable concurrent validity and high agreement for positive/neutral distinctions.
• Limited validity and reliability for negative expressions may constrain its clinical utility for detecting negative affect.
• Recommend using the Kokoro Sensor as an assistive tool, not a stand-alone method, in clinical settings.