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

Quality control (QC) of spirometry is a critical yet challenging prerequisite for accurate pulmonary function tests (PFTs), especially in primary care settings where skilled personnel are often unavailable. This study aims to develop an intelligent evaluation model to automatically identify anomaly types in spirometry reports, ensuring the reliability of diagnostic data.

Approach: We proposed a hybrid model, which integrates an artificial neural network (ANN) with wavelet analysis and integration of Convolutional Neural Network and Long Short-Term Memory models (CNN-LSTM). The model was trained and validated using over 3,000 curves collected from the First Affiliated Hospital of Guangzhou Medical University.

Main results: The proposed model achieved an average accuracy of 96.73\% on the validation set and 95.84\% agreement on the test set. Comparative analysis shows that our method outperforms existing benchmarks in Precision (0.965), Recall (0.971), and F1 score (0.968). The model successfully identifies a wide range of anomalies, including cough, glottic closure, and initial leakage, with high consistency.

Significance: This approach provides a high-precision, near real-time diagnostic tool for spirometry quality assessment. By simplifying data management and standardizing evaluation processes, the model supports primary healthcare institutions in enhancing testing quality and diagnostic reliability, particularly in resource-limited environments.