Automatic characterization of user errors in spirometry.

Area of Science:

• Pulmonary Medicine
• Biomedical Engineering
• Artificial Intelligence

Background:

• Spirometry is crucial for diagnosing and managing chronic lung diseases.
• Patient errors during spirometry maneuvers can yield inaccurate results, necessitating specialist supervision.
• Reducing the need for direct specialist coaching can enhance spirometry accessibility.

Purpose of the Study:

• To develop an automated system for detecting common patient errors in spirometry.
• To reduce reliance on trained specialists for real-time error correction during spirometry tests.
• To improve the quality and availability of spirometry, particularly in underserved and telemedicine environments.

Main Methods:

• Machine learning classifiers were created for four distinct spirometry errors.
• Features were extracted from spirometry data to train the classifiers.
• The system was evaluated on its ability to label erroneous spirometry maneuvers.

Main Results:

• The developed machine learning models achieved high accuracy in detecting specific patient errors.
• F-scores for error detection ranged from 0.85 to 0.92, indicating robust performance.
• The study successfully demonstrated automated error detection in spirometry maneuvers.

Conclusions:

• Automated detection of spirometry errors is feasible using machine learning.
• This technology can reduce the need for specialist intervention, improving efficiency.
• The findings support increased availability of accurate spirometry, especially in low-resource and telemedicine settings.