An intelligent evaluation model for spirometry quality using wavelet and deep time-series methods.

Area of Science:

• Pulmonary Medicine
• Artificial Intelligence
• Biomedical Engineering

Background:

• Spirometry quality control (QC) is crucial for accurate pulmonary function tests (PFTs).
• Skilled personnel for PFTs are often scarce in primary care settings.
• Automated anomaly detection in spirometry reports is needed to ensure data reliability.

Purpose of the Study:

• To develop an intelligent evaluation model for automatic identification of spirometry report anomalies.
• To ensure the reliability and accuracy of diagnostic data from spirometry.

Main Methods:

• A hybrid model integrating artificial neural networks (ANN), wavelet analysis, and CNN-LSTM was proposed.
• The model was trained and validated on over 3,000 spirometry curves.
• The model was evaluated for its ability to identify various anomalies like cough, glottic closure, and leakage.

Main Results:

• The hybrid model achieved 96.73% average accuracy on the validation set and 95.84% agreement on the test set.
• The model outperformed existing benchmarks with high Precision (0.965), Recall (0.971), and F1 score (0.968).
• High consistency was observed in identifying diverse spirometry anomalies.

Conclusions:

• The developed model offers a high-precision, near real-time diagnostic tool for spirometry quality assessment.
• This approach can simplify data management and standardize evaluation processes.
• It supports primary healthcare institutions in improving testing quality and diagnostic reliability, especially in resource-limited environments.