A Clinical Decision Support System for Sleep Staging Tasks With Explanations From Artificial Intelligence:...

Area of Science:

• Artificial intelligence in medicine
• Clinical decision support systems
• Sleep science and polysomnography

Background:

• Deep learning in clinical settings requires human expert review.
• AI models function as clinical decision support systems (CDSSs).
• Lack of interpretability and user-centered design hinders AI adoption by clinicians.

Purpose of the Study:

• Develop an AI-based CDSS for sleep staging review.
• Provide clinically sound, user-centered explanations for AI predictions.
• Enhance polysomnographic technician's ability to review AI sleep staging results.

Main Methods:

• User-centered design framework for explanation generation.
• User interviews and iterative design to identify explanation needs.
• Evaluation of CDSS with polysomnographic technicians measuring accuracy and interrater reliability.

Main Results:

• Technicians require explanations linked to electroencephalogram (EEG) patterns.
• AI explanations were tailored to specific EEG patterns and technician workflows.
• Significant improvement in sleep staging performance (56.75 to 60.59, P=.05) for less experienced technicians.

Conclusions:

• User-centered design is effective for creating interpretable AI CDSS for sleep staging.
• Providing AI predictions with relevant clinical explanations aids adoption.
• The developed CDSS supports technicians in reviewing AI-generated sleep staging results.