Time-Frequency Respiratory Impedance Maps Enable Within-Breath Deep Learning for Small Airway Dysfunction...

Area of Science:

• Respiratory Medicine
• Artificial Intelligence
• Biomedical Engineering

Background:

• Small airway dysfunction (SAD) is an early indicator of chronic airway diseases.
• Current spirometry methods for SAD detection are burdensome and effort-dependent.
• Impulse Oscillometry (IOS) offers a non-invasive, tidal-breathing alternative for respiratory assessment.

Purpose of the Study:

• To develop and validate a deep learning framework for SAD identification using IOS.
• To leverage within-breath impedance dynamics for enhanced SAD detection.
• To improve upon existing methods by reducing patient burden and increasing accuracy.

Main Methods:

• A dual-domain deep learning framework processing raw IOS time-series signals.
• Transformation of IOS data into time-frequency respiratory impedance maps (TFRIM).
• A two-stream architecture learning from TFRIM and time-series data, with adaptive feature modulation for calibration.

Main Results:

• The proposed framework achieved 81.39% accuracy in identifying SAD across 2510 subjects.
• The model outperformed representative baseline methods in experimental validation.
• Joint prediction of multiple small airway indices with decision-level fusion.

Conclusions:

• Combining within-breath IOS dynamics with subject-specific calibration shows potential for SAD identification.
• The deep learning framework offers a less burdensome alternative to spirometry.
• Further external validation is recommended before clinical screening deployment.