Deep learning algorithm for visual quality assessment of the spirograms

Area of Science:

• Pulmonary Function Testing
• Medical Imaging Analysis
• Artificial Intelligence in Healthcare

Background:

• Spirometry quality is vital for accurate interpretation of pulmonary function parameters.
• Current American Thoracic Society and European Respiratory Society (ATS/ERS) standards require manual visual evaluation of spirometry curves.
• Automated assessment of quantitative criteria exists, but visual assessment remains a bottleneck.

Purpose of the Study:

• To develop and evaluate a convolutional neural network (CNN) algorithm for automatic quality assessment of spirometry curves.
• To provide an alternative to manual verification of spirogram acceptability and usability.
• To enhance the efficiency and consistency of spirometry quality control.

Main Methods:

• A dataset of 1998 spirograms meeting ATS/ERS quantitative criteria was used.
• Pulmonologists annotated each spirogram as 'confirm' or 'reject' for FEV1 and FVC.
• A CNN classification algorithm was developed and optimized using cross-validation on an 80% training and 20% test split.

Main Results:

• The CNN algorithm achieved high accuracy, sensitivity, and specificity for both FEV1 (92.6%, 93.1%, 90.0%) and FVC (94.1%, 95.6%, 88.3%).
• The algorithm demonstrated robust performance in classifying spirometry curve quality.
• Results indicate the algorithm's potential for reliable automated quality assessment.

Conclusions:

• The developed CNN algorithm offers a significant improvement in spirometry test quality assessment.
• It is particularly beneficial for unsupervised spirometry and can streamline quality control in clinical trials.
• This automated tool can serve as a valuable adjunct to specialist review for large-scale spirometry data analysis.