Deep-learning algorithm helps to standardise ATS/ERS spirometric acceptability and usability criteria.

Area of Science:

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

Background:

• Spirometry quality control relies on manual inspection, which is time-consuming and prone to inter-technician variability.
• Current American Thoracic Society (ATS)/European Respiratory Society (ERS) guidelines require subjective visual assessment alongside quantitative limits.
• Automating spirometry quality control is crucial for standardization and efficiency.

Purpose of the Study:

• To develop and validate a deep learning model, specifically a convolutional neural network (CNN), for standardizing spirometric maneuver acceptability and usability.
• To compare the performance of the CNN approach against traditional ATS/ERS criteria and rule-based models.
• To interpret the CNN model's decision-making process using Shapley values.

Main Methods:

• A dataset of 36,873 spirometry curves from the National Health and Nutritional Examination Survey (NHANES) was utilized.
• Raw spirometry data was converted into images of maximal expiratory flow-volume curves (MEFVC) for CNN processing.
• CNN models were trained on 90% of the data and tested on the remaining 10%, with Shapley values used for interpretation.

Main Results:

• The CNN model achieved 87% accuracy for maneuver acceptability and 92% for usability on the test set.
• For usability, the CNN demonstrated high sensitivity (92%) and specificity (96%).
• The CNN significantly outperformed ATS/ERS quantifiable rule-based models (p<0.0001), with MEFVC<1s and volume-time plateau being key factors for acceptability.

Conclusions:

• CNNs effectively standardize spirometric maneuver acceptability and usability, mimicking expert technician assessments.
• The developed algorithm automates a critical phase of spirometry quality control by integrating visual and quantitative data.
• This deep learning approach offers potential for individual maneuver feedback and improved diagnostic consistency.